nim-chronos/chronos/transports/stream.nim

2677 lines
103 KiB
Nim
Raw Normal View History

2018-05-16 08:22:34 +00:00
#
# Chronos Stream Transport
# (c) Copyright 2018-Present
2018-05-16 08:22:34 +00:00
# Status Research & Development GmbH
#
# Licensed under either of
# Apache License, version 2.0, (LICENSE-APACHEv2)
# MIT license (LICENSE-MIT)
when (NimMajor, NimMinor) < (1, 4):
{.push raises: [Defect].}
else:
{.push raises: [].}
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
import std/deques
import ".."/[asyncloop, handles, osdefs, osutils]
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
import common
2018-05-16 08:22:34 +00:00
type
2018-05-21 21:52:57 +00:00
VectorKind = enum
2018-05-22 21:03:13 +00:00
DataBuffer, # Simple buffer pointer/length
DataFile # File handle for sendfile/TransmitFile
2018-05-21 21:52:57 +00:00
StreamVector = object
2018-05-22 21:03:13 +00:00
kind: VectorKind # Writer vector source kind
buf: pointer # Writer buffer pointer
buflen: int # Writer buffer size
offset: uint # Writer vector offset
2018-10-25 19:59:40 +00:00
size: int # Original size
writer: Future[int] # Writer vector completion Future
2018-05-21 21:52:57 +00:00
2018-05-16 08:22:34 +00:00
TransportKind* {.pure.} = enum
2018-05-22 21:03:13 +00:00
Socket, # Socket transport
Pipe, # Pipe transport
File # File transport
2018-05-16 08:22:34 +00:00
TransportFlags* = enum
None,
# Default value
WinServerPipe,
# This is internal flag which used to differentiate between server pipe
# handle and client pipe handle.
WinNoPipeFlash,
# By default `AddressFamily.Unix` transports in Windows are using
# `FlushFileBuffers()` when transport closing.
# This flag disables usage of `FlushFileBuffers()` on `AddressFamily.Unix`
# transport shutdown. If both server and client are running in the same
# thread, because of `FlushFileBuffers()` will ensure that all bytes
# or messages written to the pipe are read by the client, it is possible to
# get stuck on transport `close()`.
# Please use this flag only if you are making both client and server in
# the same thread.
TcpNoDelay # deprecated: Use SocketFlags.TcpNoDelay
SocketFlags* {.pure.} = enum
TcpNoDelay,
ReuseAddr,
ReusePort
StreamTransportTracker* = ref object of TrackerBase
opened*: int64
closed*: int64
StreamServerTracker* = ref object of TrackerBase
opened*: int64
closed*: int64
ReadMessagePredicate* = proc (data: openArray[byte]): tuple[consumed: int,
done: bool] {.
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
gcsafe, raises: [Defect].}
const
StreamTransportTrackerName = "stream.transport"
StreamServerTrackerName = "stream.server"
when defined(windows):
type
StreamTransport* = ref object of RootRef
fd*: AsyncFD # File descriptor
state: set[TransportState] # Current Transport state
reader: Future[void] # Current reader Future
buffer: seq[byte] # Reading buffer
offset: int # Reading buffer offset
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
error: ref CatchableError # Current error
queue: Deque[StreamVector] # Writer queue
future: Future[void] # Stream life future
# Windows specific part
rwsabuf: WSABUF # Reader WSABUF
wwsabuf: WSABUF # Writer WSABUF
rovl: CustomOverlapped # Reader OVERLAPPED structure
wovl: CustomOverlapped # Writer OVERLAPPED structure
roffset: int # Pending reading offset
flags: set[TransportFlags] # Internal flags
case kind*: TransportKind
of TransportKind.Socket:
domain: Domain # Socket transport domain (IPv4/IPv6)
local: TransportAddress # Local address
remote: TransportAddress # Remote address
of TransportKind.Pipe:
todo1: int
of TransportKind.File:
todo2: int
else:
type
StreamTransport* = ref object of RootRef
fd*: AsyncFD # File descriptor
state: set[TransportState] # Current Transport state
reader: Future[void] # Current reader Future
buffer: seq[byte] # Reading buffer
offset: int # Reading buffer offset
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
error: ref CatchableError # Current error
queue: Deque[StreamVector] # Writer queue
future: Future[void] # Stream life future
case kind*: TransportKind
of TransportKind.Socket:
domain: Domain # Socket transport domain (IPv4/IPv6)
local: TransportAddress # Local address
remote: TransportAddress # Remote address
of TransportKind.Pipe:
todo1: int
of TransportKind.File:
todo2: int
2018-05-16 08:22:34 +00:00
type
2018-05-22 22:29:07 +00:00
StreamCallback* = proc(server: StreamServer,
client: StreamTransport): Future[void] {.
gcsafe, raises: [Defect].}
2018-05-28 23:35:15 +00:00
## New remote client connection callback
## ``server`` - StreamServer object.
## ``client`` - accepted client transport.
2018-05-16 08:22:34 +00:00
TransportInitCallback* = proc(server: StreamServer,
fd: AsyncFD): StreamTransport {.
gcsafe, raises: [Defect].}
## Custom transport initialization procedure, which can allocate inherited
## StreamTransport object.
2018-05-16 08:22:34 +00:00
StreamServer* = ref object of SocketServer
## StreamServer object
function*: StreamCallback # callback which will be called after new
# client accepted
init*: TransportInitCallback # callback which will be called before
# transport for new client
2018-05-16 08:22:34 +00:00
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc remoteAddress*(transp: StreamTransport): TransportAddress {.
raises: [Defect, TransportError].} =
2018-05-16 08:22:34 +00:00
## Returns ``transp`` remote socket address.
if transp.kind != TransportKind.Socket:
raise newException(TransportError, "Socket required!")
if transp.remote.family == AddressFamily.None:
2018-05-16 08:22:34 +00:00
var saddr: Sockaddr_storage
var slen = SockLen(sizeof(saddr))
if getpeername(SocketHandle(transp.fd), cast[ptr SockAddr](addr saddr),
addr slen) != 0:
raiseTransportOsError(osLastError())
fromSAddr(addr saddr, slen, transp.remote)
transp.remote
2018-05-16 08:22:34 +00:00
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc localAddress*(transp: StreamTransport): TransportAddress {.
raises: [Defect, TransportError].} =
2018-05-16 08:22:34 +00:00
## Returns ``transp`` local socket address.
if transp.kind != TransportKind.Socket:
raise newException(TransportError, "Socket required!")
if transp.local.family == AddressFamily.None:
2018-05-16 08:22:34 +00:00
var saddr: Sockaddr_storage
var slen = SockLen(sizeof(saddr))
if getsockname(SocketHandle(transp.fd), cast[ptr SockAddr](addr saddr),
addr slen) != 0:
raiseTransportOsError(osLastError())
fromSAddr(addr saddr, slen, transp.local)
transp.local
2018-05-16 08:22:34 +00:00
proc localAddress*(server: StreamServer): TransportAddress =
## Returns ``server`` bound local socket address.
server.local
template completeReader(stream: StreamTransport) =
if not(isNil(transp.reader)) and not(transp.reader.finished()):
transp.reader.complete()
transp.reader = nil
2018-05-16 08:22:34 +00:00
template setReadError(t, e: untyped) =
(t).state.incl(ReadError)
(t).error = getTransportOsError(e)
2018-05-16 08:22:34 +00:00
template checkPending(t: untyped) =
if not(isNil((t).reader)):
2018-05-16 08:22:34 +00:00
raise newException(TransportError, "Read operation already pending!")
template shiftBuffer(t, c: untyped) =
if (t).offset > c:
if c > 0:
moveMem(addr((t).buffer[0]), addr((t).buffer[(c)]), (t).offset - (c))
(t).offset = (t).offset - (c)
2018-05-16 08:22:34 +00:00
else:
(t).offset = 0
template shiftVectorBuffer(v: var StreamVector, o: untyped) =
2018-05-21 21:52:57 +00:00
(v).buf = cast[pointer](cast[uint]((v).buf) + uint(o))
(v).buflen -= int(o)
template shiftVectorFile(v: var StreamVector, o: untyped) =
(v).buf = cast[pointer](cast[uint]((v).buf) - uint(o))
(v).offset += uint(o)
2018-05-21 21:52:57 +00:00
proc setupStreamTransportTracker(): StreamTransportTracker {.
gcsafe, raises: [Defect].}
proc setupStreamServerTracker(): StreamServerTracker {.
gcsafe, raises: [Defect].}
proc getStreamTransportTracker(): StreamTransportTracker {.inline.} =
var res = cast[StreamTransportTracker](getTracker(StreamTransportTrackerName))
if isNil(res):
res = setupStreamTransportTracker()
doAssert(not(isNil(res)))
res
proc getStreamServerTracker(): StreamServerTracker {.inline.} =
var res = cast[StreamServerTracker](getTracker(StreamServerTrackerName))
if isNil(res):
res = setupStreamServerTracker()
doAssert(not(isNil(res)))
res
proc dumpTransportTracking(): string {.gcsafe.} =
var tracker = getStreamTransportTracker()
"Opened transports: " & $tracker.opened & "\n" &
"Closed transports: " & $tracker.closed
proc dumpServerTracking(): string {.gcsafe.} =
var tracker = getStreamServerTracker()
"Opened servers: " & $tracker.opened & "\n" &
"Closed servers: " & $tracker.closed
proc leakTransport(): bool {.gcsafe.} =
var tracker = getStreamTransportTracker()
tracker.opened != tracker.closed
proc leakServer(): bool {.gcsafe.} =
var tracker = getStreamServerTracker()
tracker.opened != tracker.closed
proc trackStream(t: StreamTransport) {.inline.} =
var tracker = getStreamTransportTracker()
inc(tracker.opened)
proc untrackStream(t: StreamTransport) {.inline.} =
var tracker = getStreamTransportTracker()
inc(tracker.closed)
proc trackServer(s: StreamServer) {.inline.} =
var tracker = getStreamServerTracker()
inc(tracker.opened)
proc untrackServer(s: StreamServer) {.inline.} =
var tracker = getStreamServerTracker()
inc(tracker.closed)
proc setupStreamTransportTracker(): StreamTransportTracker {.gcsafe.} =
let res = StreamTransportTracker(
opened: 0, closed: 0, dump: dumpTransportTracking, isLeaked: leakTransport)
addTracker(StreamTransportTrackerName, res)
res
proc setupStreamServerTracker(): StreamServerTracker {.gcsafe.} =
let res = StreamServerTracker(
opened: 0, closed: 0, dump: dumpServerTracking, isLeaked: leakServer)
addTracker(StreamServerTrackerName, res)
res
proc completePendingWriteQueue(queue: var Deque[StreamVector],
v: int) {.inline.} =
while len(queue) > 0:
var vector = queue.popFirst()
2019-06-20 20:30:41 +00:00
if not(vector.writer.finished()):
vector.writer.complete(v)
2019-08-28 04:57:06 +00:00
proc failPendingWriteQueue(queue: var Deque[StreamVector],
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
error: ref CatchableError) {.inline.} =
2019-08-28 04:57:06 +00:00
while len(queue) > 0:
var vector = queue.popFirst()
if not(vector.writer.finished()):
vector.writer.fail(error)
proc clean(server: StreamServer) {.inline.} =
if not(server.loopFuture.finished()):
untrackServer(server)
server.loopFuture.complete()
if not(isNil(server.udata)) and (GCUserData in server.flags):
GC_unref(cast[ref int](server.udata))
GC_unref(server)
proc clean(transp: StreamTransport) {.inline.} =
if not(transp.future.finished()):
untrackStream(transp)
transp.future.complete()
GC_unref(transp)
when defined(windows):
2018-05-16 08:22:34 +00:00
template zeroOvelappedOffset(t: untyped) =
(t).offset = 0
(t).offsetHigh = 0
template setOverlappedOffset(t, o: untyped) =
(t).offset = cast[DWORD](cast[uint64](o) and 0xFFFFFFFF'u64)
(t).offsetHigh = cast[DWORD](cast[uint64](o) shr 32)
2018-05-16 08:22:34 +00:00
2018-05-21 21:52:57 +00:00
template getFileSize(v: untyped): uint =
cast[uint]((v).buf)
2018-05-16 08:22:34 +00:00
template getFileHandle(v: untyped): HANDLE =
cast[HANDLE]((v).buflen)
2018-05-16 08:22:34 +00:00
2018-05-21 21:52:57 +00:00
template setReaderWSABuffer(t: untyped) =
(t).rwsabuf.buf = cast[cstring](
2018-05-16 08:22:34 +00:00
cast[uint](addr t.buffer[0]) + uint((t).roffset))
(t).rwsabuf.len = ULONG(len((t).buffer) - (t).roffset)
2018-05-16 08:22:34 +00:00
2018-05-21 21:52:57 +00:00
template setWriterWSABuffer(t, v: untyped) =
(t).wwsabuf.buf = cast[cstring](v.buf)
(t).wwsabuf.len = cast[ULONG](v.buflen)
2018-05-16 08:22:34 +00:00
proc isConnResetError(err: OSErrorCode): bool {.inline.} =
result = (err == OSErrorCode(osdefs.WSAECONNRESET)) or
(err == OSErrorCode(osdefs.WSAECONNABORTED)) or
(err == OSErrorCode(osdefs.ERROR_PIPE_NOT_CONNECTED))
2018-05-22 16:27:20 +00:00
proc writeStreamLoop(udata: pointer) {.gcsafe, nimcall.} =
var bytesCount: uint32
2018-05-21 21:52:57 +00:00
var ovl = cast[PtrCustomOverlapped](udata)
var transp = cast[StreamTransport](ovl.data.udata)
2018-05-16 08:22:34 +00:00
if WriteClosed in transp.state:
transp.state.excl(WritePending)
transp.state.incl({WritePaused})
let error = getTransportUseClosedError()
failPendingWriteQueue(transp.queue, error)
else:
while len(transp.queue) > 0:
if WritePending in transp.state:
## Continuation
transp.state.excl(WritePending)
let err = transp.wovl.data.errCode
if err == OSErrorCode(-1):
bytesCount = transp.wovl.data.bytesCount
var vector = transp.queue.popFirst()
if bytesCount == 0:
if not(vector.writer.finished()):
vector.writer.complete(0)
else:
if transp.kind == TransportKind.Socket:
if vector.kind == VectorKind.DataBuffer:
if bytesCount < transp.wwsabuf.len:
vector.shiftVectorBuffer(bytesCount)
transp.queue.addFirst(vector)
else:
if not(vector.writer.finished()):
# This conversion to `int` safe, because its impossible
# to call write() with size bigger than `int`.
vector.writer.complete(int(transp.wwsabuf.len))
else:
if uint(bytesCount) < getFileSize(vector):
vector.shiftVectorFile(bytesCount)
transp.queue.addFirst(vector)
else:
if not(vector.writer.finished()):
vector.writer.complete(int(getFileSize(vector)))
elif transp.kind == TransportKind.Pipe:
if vector.kind == VectorKind.DataBuffer:
if bytesCount < transp.wwsabuf.len:
vector.shiftVectorBuffer(bytesCount)
transp.queue.addFirst(vector)
else:
if not(vector.writer.finished()):
# This conversion to `int` safe, because its impossible
# to call write() with size bigger than `int`.
vector.writer.complete(int(transp.wwsabuf.len))
elif int(err) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt
2019-08-28 04:57:06 +00:00
transp.state.incl({WritePaused, WriteEof})
let vector = transp.queue.popFirst()
2019-08-28 04:57:06 +00:00
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
else:
let vector = transp.queue.popFirst()
if isConnResetError(err):
# Soft error happens which indicates that remote peer got
# disconnected, complete all pending writes in queue with 0.
transp.state.incl({WritePaused, WriteEof})
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
else:
transp.state.incl({WritePaused, WriteError})
let error = getTransportOsError(err)
if not(vector.writer.finished()):
vector.writer.fail(error)
failPendingWriteQueue(transp.queue, error)
break
else:
## Initiation
transp.state.incl(WritePending)
if transp.kind == TransportKind.Socket:
let sock = SocketHandle(transp.fd)
var vector = transp.queue.popFirst()
if vector.kind == VectorKind.DataBuffer:
transp.wovl.zeroOvelappedOffset()
transp.setWriterWSABuffer(vector)
let ret = wsaSend(sock, addr transp.wwsabuf, 1,
addr bytesCount, DWORD(0),
cast[POVERLAPPED](addr transp.wovl), nil)
if ret != 0:
let err = osLastError()
if int(err) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt
transp.state.excl(WritePending)
transp.state.incl({WritePaused, WriteEof})
2019-06-20 20:30:41 +00:00
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
elif int(err) == osdefs.ERROR_IO_PENDING:
transp.queue.addFirst(vector)
else:
transp.state.excl(WritePending)
if isConnResetError(err):
# Soft error happens which indicates that remote peer got
# disconnected, complete all pending writes in queue with 0.
transp.state.incl({WritePaused, WriteEof})
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
else:
transp.state.incl({WritePaused, WriteError})
let error = getTransportOsError(err)
if not(vector.writer.finished()):
vector.writer.fail(error)
failPendingWriteQueue(transp.queue, error)
break
else:
transp.queue.addFirst(vector)
2018-05-16 08:22:34 +00:00
else:
let loop = getThreadDispatcher()
let size = min(uint32(getFileSize(vector)), 2_147_483_646'u32)
transp.wovl.setOverlappedOffset(vector.offset)
var ret = loop.transmitFile(sock, getFileHandle(vector), size,
0'u32,
cast[POVERLAPPED](addr transp.wovl),
nil, 0'u32)
if ret == 0:
let err = osLastError()
if int(err) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt
transp.state.excl(WritePending)
transp.state.incl({WritePaused, WriteEof})
2019-06-20 20:30:41 +00:00
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
elif int(err) == osdefs.ERROR_IO_PENDING:
transp.queue.addFirst(vector)
else:
transp.state.excl(WritePending)
if isConnResetError(err):
# Soft error happens which indicates that remote peer got
# disconnected, complete all pending writes in queue with 0.
transp.state.incl({WritePaused, WriteEof})
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
else:
transp.state.incl({WritePaused, WriteError})
let error = getTransportOsError(err)
if not(vector.writer.finished()):
vector.writer.fail(error)
failPendingWriteQueue(transp.queue, error)
break
else:
transp.queue.addFirst(vector)
elif transp.kind == TransportKind.Pipe:
let pipe = HANDLE(transp.fd)
var vector = transp.queue.popFirst()
if vector.kind == VectorKind.DataBuffer:
transp.wovl.zeroOvelappedOffset()
transp.setWriterWSABuffer(vector)
let ret = writeFile(pipe, cast[pointer](transp.wwsabuf.buf),
DWORD(transp.wwsabuf.len), addr bytesCount,
cast[POVERLAPPED](addr transp.wovl))
if ret == 0:
let err = osLastError()
if int(err) in {osdefs.ERROR_OPERATION_ABORTED,
osdefs.ERROR_NO_DATA}:
# CancelIO() interrupt
transp.state.excl(WritePending)
transp.state.incl({WritePaused, WriteEof})
2019-06-20 20:30:41 +00:00
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
elif int(err) == osdefs.ERROR_IO_PENDING:
transp.queue.addFirst(vector)
else:
transp.state.excl(WritePending)
if isConnResetError(err):
# Soft error happens which indicates that remote peer got
# disconnected, complete all pending writes in queue with 0.
transp.state.incl({WritePaused, WriteEof})
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
break
else:
transp.state.incl({WritePaused, WriteError})
let error = getTransportOsError(err)
if not(vector.writer.finished()):
vector.writer.fail(error)
failPendingWriteQueue(transp.queue, error)
break
else:
transp.queue.addFirst(vector)
break
2018-05-16 08:22:34 +00:00
if len(transp.queue) == 0:
transp.state.incl(WritePaused)
2018-05-16 08:22:34 +00:00
2018-05-22 16:27:20 +00:00
proc readStreamLoop(udata: pointer) {.gcsafe, nimcall.} =
2018-05-21 21:52:57 +00:00
var ovl = cast[PtrCustomOverlapped](udata)
var transp = cast[StreamTransport](ovl.data.udata)
2018-05-16 08:22:34 +00:00
while true:
if ReadPending in transp.state:
## Continuation
2018-05-21 21:52:57 +00:00
transp.state.excl(ReadPending)
2018-05-16 08:22:34 +00:00
let err = transp.rovl.data.errCode
if err == OSErrorCode(-1):
let bytesCount = transp.rovl.data.bytesCount
if bytesCount == 0:
transp.state.incl({ReadEof, ReadPaused})
2018-05-16 08:22:34 +00:00
else:
if transp.offset != transp.roffset:
moveMem(addr transp.buffer[transp.offset],
addr transp.buffer[transp.roffset],
bytesCount)
transp.offset += int(bytesCount)
2018-05-16 08:22:34 +00:00
transp.roffset = transp.offset
if transp.offset == len(transp.buffer):
transp.state.incl(ReadPaused)
elif int(err) in {osdefs.ERROR_OPERATION_ABORTED,
osdefs.ERROR_CONNECTION_ABORTED,
osdefs.ERROR_BROKEN_PIPE,
osdefs.ERROR_NETNAME_DELETED}:
# CancelIO() interrupt or closeSocket() call.
transp.state.incl(ReadPaused)
elif transp.kind == TransportKind.Socket and
(int(err) in {osdefs.ERROR_NETNAME_DELETED,
osdefs.WSAECONNABORTED}):
transp.state.incl({ReadEof, ReadPaused})
elif transp.kind == TransportKind.Pipe and
(int(err) in {osdefs.ERROR_PIPE_NOT_CONNECTED}):
transp.state.incl({ReadEof, ReadPaused})
2018-05-16 08:22:34 +00:00
else:
transp.setReadError(err)
transp.completeReader()
if ReadClosed in transp.state:
# Stop tracking transport
transp.clean()
2018-05-21 21:52:57 +00:00
if ReadPaused in transp.state:
# Transport buffer is full, so we will not continue on reading.
break
2018-05-16 08:22:34 +00:00
else:
## Initiation
2018-05-21 21:52:57 +00:00
if transp.state * {ReadEof, ReadClosed, ReadError} == {}:
2018-05-16 08:22:34 +00:00
var flags = DWORD(0)
var bytesCount = 0'u32
2018-05-16 08:22:34 +00:00
transp.state.excl(ReadPaused)
transp.state.incl(ReadPending)
if transp.kind == TransportKind.Socket:
let sock = SocketHandle(transp.fd)
2018-05-21 21:52:57 +00:00
transp.roffset = transp.offset
transp.setReaderWSABuffer()
let ret = wsaRecv(sock, addr transp.rwsabuf, 1,
2018-05-16 08:22:34 +00:00
addr bytesCount, addr flags,
cast[POVERLAPPED](addr transp.rovl), nil)
if ret != 0:
let err = osLastError()
if int32(err) == osdefs.ERROR_OPERATION_ABORTED:
2018-05-28 23:35:15 +00:00
# CancelIO() interrupt
2018-05-21 21:52:57 +00:00
transp.state.excl(ReadPending)
2018-05-16 08:22:34 +00:00
transp.state.incl(ReadPaused)
elif int32(err) in {osdefs.WSAECONNRESET, osdefs.WSAENETRESET,
osdefs.WSAECONNABORTED}:
2018-07-24 15:43:30 +00:00
transp.state.excl(ReadPending)
transp.state.incl({ReadEof, ReadPaused})
transp.completeReader()
elif int32(err) != osdefs.ERROR_IO_PENDING:
transp.state.excl(ReadPending)
transp.state.incl(ReadPaused)
2018-05-16 08:22:34 +00:00
transp.setReadError(err)
transp.completeReader()
elif transp.kind == TransportKind.Pipe:
let pipe = HANDLE(transp.fd)
transp.roffset = transp.offset
transp.setReaderWSABuffer()
let ret = readFile(pipe, cast[pointer](transp.rwsabuf.buf),
DWORD(transp.rwsabuf.len), addr bytesCount,
cast[POVERLAPPED](addr transp.rovl))
if ret == 0:
let err = osLastError()
if int32(err) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt
transp.state.excl(ReadPending)
transp.state.incl(ReadPaused)
elif int32(err) in {osdefs.ERROR_BROKEN_PIPE,
osdefs.ERROR_PIPE_NOT_CONNECTED}:
transp.state.excl(ReadPending)
transp.state.incl({ReadEof, ReadPaused})
transp.completeReader()
elif int32(err) != osdefs.ERROR_IO_PENDING:
transp.state.excl(ReadPending)
transp.state.incl(ReadPaused)
transp.setReadError(err)
transp.completeReader()
else:
transp.state.incl(ReadPaused)
transp.completeReader()
# Transport close happens in callback, and we not started new
# WSARecvFrom session.
if ReadClosed in transp.state:
2019-06-20 20:30:41 +00:00
if not(transp.future.finished()):
transp.future.complete()
2018-05-16 08:22:34 +00:00
## Finish Loop
break
proc newStreamSocketTransport(sock: AsyncFD, bufsize: int,
child: StreamTransport): StreamTransport =
var transp: StreamTransport
if not(isNil(child)):
transp = child
else:
transp = StreamTransport(kind: TransportKind.Socket)
2018-05-21 21:52:57 +00:00
transp.fd = sock
transp.rovl.data = CompletionData(cb: readStreamLoop,
2018-05-21 21:52:57 +00:00
udata: cast[pointer](transp))
transp.wovl.data = CompletionData(cb: writeStreamLoop,
2018-05-21 21:52:57 +00:00
udata: cast[pointer](transp))
transp.buffer = newSeq[byte](bufsize)
transp.state = {ReadPaused, WritePaused}
transp.queue = initDeque[StreamVector]()
transp.future = newFuture[void]("stream.socket.transport")
GC_ref(transp)
transp
2018-05-16 08:22:34 +00:00
proc newStreamPipeTransport(fd: AsyncFD, bufsize: int,
child: StreamTransport,
flags: set[TransportFlags] = {}): StreamTransport =
var transp: StreamTransport
if not(isNil(child)):
transp = child
else:
transp = StreamTransport(kind: TransportKind.Pipe)
transp.fd = fd
transp.rovl.data = CompletionData(cb: readStreamLoop,
udata: cast[pointer](transp))
transp.wovl.data = CompletionData(cb: writeStreamLoop,
udata: cast[pointer](transp))
transp.buffer = newSeq[byte](bufsize)
transp.flags = flags
transp.state = {ReadPaused, WritePaused}
transp.queue = initDeque[StreamVector]()
transp.future = newFuture[void]("stream.pipe.transport")
GC_ref(transp)
transp
2018-05-16 08:22:34 +00:00
proc bindToDomain(handle: AsyncFD, domain: Domain): bool =
if domain == Domain.AF_INET6:
var saddr: Sockaddr_in6
saddr.sin6_family = type(saddr.sin6_family)(osdefs.AF_INET6)
if osdefs.bindSocket(SocketHandle(handle),
cast[ptr SockAddr](addr(saddr)),
sizeof(saddr).SockLen) != 0'i32:
return false
true
elif domain == Domain.AF_INET:
2018-05-16 08:22:34 +00:00
var saddr: Sockaddr_in
saddr.sin_family = type(saddr.sin_family)(osdefs.AF_INET)
if osdefs.bindSocket(SocketHandle(handle),
cast[ptr SockAddr](addr(saddr)),
sizeof(saddr).SockLen) != 0'i32:
return false
true
else:
raiseAssert "Unsupported domain"
2018-05-16 08:22:34 +00:00
proc connect*(address: TransportAddress,
bufferSize = DefaultStreamBufferSize,
child: StreamTransport = nil,
localAddress = TransportAddress(),
flags: set[SocketFlags] = {},
): Future[StreamTransport] =
## Open new connection to remote peer with address ``address`` and create
## new transport object ``StreamTransport`` for established connection.
2018-05-28 23:35:15 +00:00
## ``bufferSize`` is size of internal buffer for transport.
let loop = getThreadDispatcher()
2018-05-16 08:22:34 +00:00
var retFuture = newFuture[StreamTransport]("stream.transport.connect")
if address.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
## Socket handling part
var
saddr: Sockaddr_storage
slen: SockLen
sock: AsyncFD
povl: RefCustomOverlapped
proto: Protocol
var raddress = windowsAnyAddressFix(address)
toSAddr(raddress, saddr, slen)
proto = Protocol.IPPROTO_TCP
sock = createAsyncSocket(raddress.getDomain(), SockType.SOCK_STREAM,
proto)
if sock == asyncInvalidSocket:
2019-06-20 20:30:41 +00:00
retFuture.fail(getTransportOsError(osLastError()))
return retFuture
2018-05-16 08:22:34 +00:00
if SocketFlags.ReuseAddr in flags:
if not(setSockOpt(sock, SOL_SOCKET, SO_REUSEADDR, 1)):
let err = osLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
if SocketFlags.ReusePort in flags:
if not(setSockOpt(sock, SOL_SOCKET, SO_REUSEPORT, 1)):
let err = osLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
if localAddress != TransportAddress():
if localAddress.family != address.family:
sock.closeSocket()
retFuture.fail(newException(TransportOsError,
"connect local address domain is not equal to target address domain"))
return retFuture
var
localAddr: Sockaddr_storage
localAddrLen: SockLen
localAddress.toSAddr(localAddr, localAddrLen)
if bindSocket(SocketHandle(sock),
cast[ptr SockAddr](addr localAddr), localAddrLen) != 0:
sock.closeSocket()
retFuture.fail(getTransportOsError(osLastError()))
return retFuture
elif not(bindToDomain(sock, raddress.getDomain())):
let err = wsaGetLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
2018-05-16 08:22:34 +00:00
2019-06-20 20:30:41 +00:00
proc socketContinuation(udata: pointer) {.gcsafe.} =
var ovl = cast[RefCustomOverlapped](udata)
2019-06-20 20:30:41 +00:00
if not(retFuture.finished()):
if ovl.data.errCode == OSErrorCode(-1):
if setsockopt(SocketHandle(sock), cint(osdefs.SOL_SOCKET),
cint(osdefs.SO_UPDATE_CONNECT_CONTEXT), nil,
SockLen(0)) != 0'i32:
let err = wsaGetLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
else:
let transp = newStreamSocketTransport(sock, bufferSize, child)
# Start tracking transport
trackStream(transp)
retFuture.complete(transp)
else:
sock.closeSocket()
retFuture.fail(getTransportOsError(ovl.data.errCode))
GC_unref(ovl)
2019-06-20 20:30:41 +00:00
proc cancel(udata: pointer) {.gcsafe.} =
sock.closeSocket()
2019-06-20 20:30:41 +00:00
povl = RefCustomOverlapped()
GC_ref(povl)
povl.data = CompletionData(cb: socketContinuation)
let res = loop.connectEx(SocketHandle(sock),
cast[ptr SockAddr](addr saddr),
cint(slen), nil, 0, nil,
cast[POVERLAPPED](povl))
# We will not process immediate completion, to avoid undefined behavior.
if res == osdefs.FALSE:
let err = osLastError()
if int32(err) != osdefs.ERROR_IO_PENDING:
GC_unref(povl)
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
retFuture.cancelCallback = cancel
elif address.family == AddressFamily.Unix:
## Unix domain socket emulation with Windows Named Pipes.
# For some reason Nim compiler does not detect `pipeHandle` usage in
# pipeContinuation() procedure, so we marking it as {.used.} here.
var pipeHandle {.used.} = osdefs.INVALID_HANDLE_VALUE
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
var pipeContinuation: proc (udata: pointer) {.gcsafe, raises: [Defect].}
pipeContinuation = proc (udata: pointer) {.gcsafe, raises: [Defect].} =
2019-06-20 20:30:41 +00:00
# Continue only if `retFuture` is not cancelled.
if not(retFuture.finished()):
let
pipeSuffix = $cast[cstring](unsafeAddr address.address_un[0])
pipeAsciiName = PipeHeaderName & pipeSuffix[1 .. ^1]
pipeName = toWideString(pipeAsciiName).valueOr:
retFuture.fail(getTransportOsError(error))
return
genericFlags = osdefs.GENERIC_READ or osdefs.GENERIC_WRITE
shareFlags = osdefs.FILE_SHARE_READ or osdefs.FILE_SHARE_WRITE
pipeHandle = createFile(pipeName, genericFlags, shareFlags,
nil, osdefs.OPEN_EXISTING,
osdefs.FILE_FLAG_OVERLAPPED, HANDLE(0))
free(pipeName)
if pipeHandle == osdefs.INVALID_HANDLE_VALUE:
2019-06-20 20:30:41 +00:00
let err = osLastError()
if int32(err) == osdefs.ERROR_PIPE_BUSY:
discard setTimer(Moment.fromNow(50.milliseconds),
pipeContinuation, nil)
2019-06-20 20:30:41 +00:00
else:
retFuture.fail(getTransportOsError(err))
2018-05-16 08:22:34 +00:00
else:
let res = register2(AsyncFD(pipeHandle))
if res.isErr():
retFuture.fail(getTransportOsError(res.error()))
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
return
let transp = newStreamPipeTransport(AsyncFD(pipeHandle),
2019-06-20 20:30:41 +00:00
bufferSize, child)
# Start tracking transport
trackStream(transp)
retFuture.complete(transp)
pipeContinuation(nil)
2018-05-16 08:22:34 +00:00
return retFuture
proc createAcceptPipe(server: StreamServer): Result[AsyncFD, OSErrorCode] =
let
pipeSuffix = $cast[cstring](addr server.local.address_un)
pipeName = ? toWideString(PipeHeaderName & pipeSuffix)
openMode =
if FirstPipe notin server.flags:
server.flags.incl(FirstPipe)
osdefs.PIPE_ACCESS_DUPLEX or osdefs.FILE_FLAG_OVERLAPPED or
osdefs.FILE_FLAG_FIRST_PIPE_INSTANCE
else:
osdefs.PIPE_ACCESS_DUPLEX or osdefs.FILE_FLAG_OVERLAPPED
pipeMode = osdefs.PIPE_TYPE_BYTE or osdefs.PIPE_READMODE_BYTE or
osdefs.PIPE_WAIT
pipeHandle = createNamedPipe(pipeName, openMode, pipeMode,
osdefs.PIPE_UNLIMITED_INSTANCES,
DWORD(server.bufferSize),
DWORD(server.bufferSize),
DWORD(0), nil)
free(pipeName)
if pipeHandle == osdefs.INVALID_HANDLE_VALUE:
return err(osLastError())
let res = register2(AsyncFD(pipeHandle))
if res.isErr():
discard osdefs.closeHandle(pipeHandle)
return err(res.error())
ok(AsyncFD(pipeHandle))
proc acceptPipeLoop(udata: pointer) {.gcsafe, nimcall.} =
var ovl = cast[PtrCustomOverlapped](udata)
var server = cast[StreamServer](ovl.data.udata)
while true:
if server.apending:
## Continuation
server.apending = false
if server.status notin {ServerStatus.Stopped, ServerStatus.Closed}:
if ovl.data.errCode == OSErrorCode(-1):
var ntransp: StreamTransport
var flags = {WinServerPipe}
if NoPipeFlash in server.flags:
flags.incl(WinNoPipeFlash)
if not(isNil(server.init)):
var transp = server.init(server, server.sock)
ntransp = newStreamPipeTransport(server.sock, server.bufferSize,
transp, flags)
else:
ntransp = newStreamPipeTransport(server.sock, server.bufferSize,
nil, flags)
# Start tracking transport
trackStream(ntransp)
asyncSpawn server.function(server, ntransp)
elif int32(ovl.data.errCode) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt or close call.
if server.status in {ServerStatus.Closed, ServerStatus.Stopped}:
server.clean()
break
else:
# We should not raise defects in this loop.
discard disconnectNamedPipe(HANDLE(server.sock))
discard closeHandle(HANDLE(server.sock))
raiseOsDefect(osLastError(), "acceptPipeLoop(): Unable to accept " &
"new pipe connection")
else:
# Server close happens in callback, and we are not started new
# connectNamedPipe session.
if not(server.loopFuture.finished()):
server.clean()
break
else:
## Initiation
if server.status notin {ServerStatus.Stopped, ServerStatus.Closed}:
server.apending = true
let
pipeSuffix = $cast[cstring](addr server.local.address_un)
pipeAsciiName = PipeHeaderName & pipeSuffix
pipeName = toWideString(pipeAsciiName).valueOr:
raiseOsDefect(error, "acceptPipeLoop(): Unable to create name " &
"for new pipe connection")
openMode =
if FirstPipe notin server.flags:
server.flags.incl(FirstPipe)
osdefs.PIPE_ACCESS_DUPLEX or osdefs.FILE_FLAG_OVERLAPPED or
osdefs.FILE_FLAG_FIRST_PIPE_INSTANCE
else:
osdefs.PIPE_ACCESS_DUPLEX or osdefs.FILE_FLAG_OVERLAPPED
pipeMode = osdefs.PIPE_TYPE_BYTE or osdefs.PIPE_READMODE_BYTE or
osdefs.PIPE_WAIT
pipeHandle = createNamedPipe(pipeName, openMode, pipeMode,
osdefs.PIPE_UNLIMITED_INSTANCES,
DWORD(server.bufferSize),
DWORD(server.bufferSize),
DWORD(0), nil)
free(pipeName)
if pipeHandle == osdefs.INVALID_HANDLE_VALUE:
raiseOsDefect(osLastError(), "acceptPipeLoop(): Unable to create " &
"new pipe")
server.sock = AsyncFD(pipeHandle)
let wres = register2(server.sock)
if wres.isErr():
raiseOsDefect(wres.error(), "acceptPipeLoop(): Unable to " &
"register new pipe in dispatcher")
let res = connectNamedPipe(pipeHandle,
cast[POVERLAPPED](addr server.aovl))
if res == 0:
let errCode = osLastError()
if errCode == osdefs.ERROR_OPERATION_ABORTED:
server.apending = false
break
elif errCode == osdefs.ERROR_IO_PENDING:
discard
elif errCode == osdefs.ERROR_PIPE_CONNECTED:
discard
else:
raiseOsDefect(errCode, "acceptPipeLoop(): Unable to establish " &
"pipe connection")
break
else:
# Server close happens in callback, and we are not started new
# connectNamedPipe session.
if not(server.loopFuture.finished()):
server.clean()
break
proc acceptLoop(udata: pointer) {.gcsafe, nimcall.} =
var ovl = cast[PtrCustomOverlapped](udata)
var server = cast[StreamServer](ovl.data.udata)
var loop = getThreadDispatcher()
2018-05-16 08:22:34 +00:00
while true:
if server.apending:
## Continuation
server.apending = false
if server.status notin {ServerStatus.Stopped, ServerStatus.Closed}:
if ovl.data.errCode == OSErrorCode(-1):
if setsockopt(SocketHandle(server.asock), cint(osdefs.SOL_SOCKET),
cint(osdefs.SO_UPDATE_ACCEPT_CONTEXT),
addr server.sock,
SockLen(sizeof(SocketHandle))) != 0'i32:
let errCode = OSErrorCode(wsaGetLastError())
server.asock.closeSocket()
raiseOsDefect(errCode, "acceptLoop(): Unable to set accept " &
"context for socket")
else:
var ntransp: StreamTransport
if not(isNil(server.init)):
let transp = server.init(server, server.asock)
ntransp = newStreamSocketTransport(server.asock,
server.bufferSize,
transp)
else:
ntransp = newStreamSocketTransport(server.asock,
server.bufferSize, nil)
# Start tracking transport
trackStream(ntransp)
asyncSpawn server.function(server, ntransp)
elif int32(ovl.data.errCode) == osdefs.ERROR_OPERATION_ABORTED:
# CancelIO() interrupt or close.
server.asock.closeSocket()
if server.status in {ServerStatus.Closed, ServerStatus.Stopped}:
# Stop tracking server
if not(server.loopFuture.finished()):
server.clean()
break
else:
server.asock.closeSocket()
raiseOsDefect(ovl.data.errCode, "acceptLoop(): Unable to accept " &
"new connection")
else:
# Server close happens in callback, and we are not started new
# AcceptEx session.
if not(server.loopFuture.finished()):
server.clean()
break
else:
## Initiation
if server.status notin {ServerStatus.Stopped, ServerStatus.Closed}:
server.apending = true
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
# TODO No way to report back errors!
server.asock =
block:
let sock = createAsyncSocket(server.domain, SockType.SOCK_STREAM,
Protocol.IPPROTO_TCP)
if sock == asyncInvalidSocket:
raiseOsDefect(osLastError(),
"acceptLoop(): Unablet to create new socket")
sock
var dwBytesReceived = DWORD(0)
let dwReceiveDataLength = DWORD(0)
let dwLocalAddressLength = DWORD(sizeof(Sockaddr_in6) + 16)
let dwRemoteAddressLength = DWORD(sizeof(Sockaddr_in6) + 16)
let res = loop.acceptEx(SocketHandle(server.sock),
SocketHandle(server.asock),
addr server.abuffer[0],
dwReceiveDataLength, dwLocalAddressLength,
dwRemoteAddressLength, addr dwBytesReceived,
cast[POVERLAPPED](addr server.aovl))
if res == osdefs.FALSE:
let errCode = osLastError()
if errCode == osdefs.ERROR_OPERATION_ABORTED:
server.apending = false
break
elif errCode == osdefs.ERROR_IO_PENDING:
discard
else:
raiseOsDefect(errCode, "acceptLoop(): Unable to accept " &
"connection")
break
else:
# Server close happens in callback, and we are not started new
# AcceptEx session.
if not(server.loopFuture.finished()):
server.clean()
break
2018-05-16 08:22:34 +00:00
proc resumeRead(transp: StreamTransport): Result[void, OSErrorCode] =
if ReadPaused in transp.state:
transp.state.excl(ReadPaused)
readStreamLoop(cast[pointer](addr transp.rovl))
ok()
2018-05-16 08:22:34 +00:00
proc resumeWrite(transp: StreamTransport): Result[void, OSErrorCode] =
if WritePaused in transp.state:
transp.state.excl(WritePaused)
writeStreamLoop(cast[pointer](addr transp.wovl))
ok()
2018-05-16 08:22:34 +00:00
proc pauseAccept(server: StreamServer): Result[void, OSErrorCode] =
if server.apending:
discard cancelIo(HANDLE(server.sock))
ok()
proc resumeAccept(server: StreamServer): Result[void, OSErrorCode] =
if not(server.apending):
server.aovl.data.cb(addr server.aovl)
ok()
proc accept*(server: StreamServer): Future[StreamTransport] =
var retFuture = newFuture[StreamTransport]("stream.server.accept")
doAssert(server.status != ServerStatus.Running,
"You could not use accept() if server was already started")
if server.status == ServerStatus.Closed:
retFuture.fail(getServerUseClosedError())
return retFuture
proc continuationSocket(udata: pointer) {.gcsafe.} =
2023-03-24 15:34:45 +00:00
if retFuture.finished():
# `retFuture` could become finished in 2 cases:
# 1. OS sends IOCP notification about failure, but we already failed
# `retFuture` with proper error.
# 2. `accept()` call has been cancelled. Cancellation callback closed
# accepting socket, so OS sends IOCP notification with an
# `ERROR_OPERATION_ABORTED` error.
return
2023-03-24 15:34:45 +00:00
var
ovl = cast[PtrCustomOverlapped](udata)
server = cast[StreamServer](ovl.data.udata)
server.apending = false
2023-03-24 15:34:45 +00:00
if server.status in {ServerStatus.Stopped, ServerStatus.Closed}:
retFuture.fail(getServerUseClosedError())
server.asock.closeSocket()
server.clean()
else:
case ovl.data.errCode
of OSErrorCode(-1):
if setsockopt(SocketHandle(server.asock), cint(osdefs.SOL_SOCKET),
cint(osdefs.SO_UPDATE_ACCEPT_CONTEXT),
addr server.sock,
SockLen(sizeof(SocketHandle))) != 0'i32:
let err = osLastError()
server.asock.closeSocket()
if err == osdefs.WSAENOTSOCK:
# This can be happened when server get closed, but continuation
# was already scheduled, so we failing it not with OS error.
retFuture.fail(getServerUseClosedError())
else:
let errorMsg = osErrorMsg(err)
retFuture.fail(getConnectionAbortedError(errorMsg))
else:
var ntransp: StreamTransport
if not(isNil(server.init)):
let transp = server.init(server, server.asock)
ntransp = newStreamSocketTransport(server.asock,
server.bufferSize,
transp)
else:
ntransp = newStreamSocketTransport(server.asock,
server.bufferSize, nil)
# Start tracking transport
trackStream(ntransp)
retFuture.complete(ntransp)
of OSErrorCode(osdefs.ERROR_OPERATION_ABORTED):
# CancelIO() interrupt or close.
server.asock.closeSocket()
retFuture.fail(getServerUseClosedError())
2023-03-24 15:34:45 +00:00
server.clean()
of OSErrorCode(osdefs.WSAENETDOWN),
OSErrorCode(osdefs.WSAENETRESET),
OSErrorCode(osdefs.WSAECONNABORTED),
OSErrorCode(osdefs.WSAECONNRESET),
OSErrorCode(osdefs.WSAETIMEDOUT):
server.asock.closeSocket()
2023-03-24 15:34:45 +00:00
retFuture.fail(getConnectionAbortedError(int(ovl.data.errCode)))
server.clean()
else:
2023-03-24 15:34:45 +00:00
server.asock.closeSocket()
retFuture.fail(getTransportOsError(ovl.data.errCode))
proc cancellationSocket(udata: pointer) {.gcsafe.} =
if server.apending:
server.apending = false
server.asock.closeSocket()
proc continuationPipe(udata: pointer) {.gcsafe.} =
2023-03-24 15:34:45 +00:00
if retFuture.finished():
# `retFuture` could become finished in 2 cases:
# 1. OS sends IOCP notification about failure, but we already failed
# `retFuture` with proper error.
# 2. `accept()` call has been cancelled. Cancellation callback closed
# accepting socket, so OS sends IOCP notification with an
# `ERROR_OPERATION_ABORTED` error.
return
2023-03-24 15:34:45 +00:00
var
ovl = cast[PtrCustomOverlapped](udata)
server = cast[StreamServer](ovl.data.udata)
server.apending = false
2023-03-24 15:34:45 +00:00
if server.status in {ServerStatus.Stopped, ServerStatus.Closed}:
retFuture.fail(getServerUseClosedError())
server.sock.closeHandle()
server.clean()
else:
if ovl.data.errCode == OSErrorCode(-1):
var ntransp: StreamTransport
var flags = {WinServerPipe}
if NoPipeFlash in server.flags:
flags.incl(WinNoPipeFlash)
if not(isNil(server.init)):
var transp = server.init(server, server.sock)
ntransp = newStreamPipeTransport(server.sock, server.bufferSize,
transp, flags)
else:
ntransp = newStreamPipeTransport(server.sock, server.bufferSize,
nil, flags)
server.sock = server.createAcceptPipe().valueOr:
server.sock = asyncInvalidSocket
server.errorCode = error
retFuture.fail(getTransportOsError(error))
return
trackStream(ntransp)
retFuture.complete(ntransp)
elif int32(ovl.data.errCode) in {osdefs.ERROR_OPERATION_ABORTED,
osdefs.ERROR_PIPE_NOT_CONNECTED}:
# CancelIO() interrupt or close call.
retFuture.fail(getServerUseClosedError())
server.clean()
else:
2023-03-24 15:34:45 +00:00
discard closeHandle(HANDLE(server.sock))
server.sock = server.createAcceptPipe().valueOr:
server.sock = asyncInvalidSocket
server.errorCode = error
retFuture.fail(getTransportOsError(error))
return
retFuture.fail(getTransportOsError(ovl.data.errCode))
proc cancellationPipe(udata: pointer) {.gcsafe.} =
if server.apending:
server.apending = false
server.sock.closeHandle()
if server.local.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
# TCP Sockets part
var loop = getThreadDispatcher()
server.asock = createAsyncSocket(server.domain, SockType.SOCK_STREAM,
Protocol.IPPROTO_TCP)
if server.asock == asyncInvalidSocket:
let err = osLastError()
case int(err)
of osdefs.ERROR_TOO_MANY_OPEN_FILES,
osdefs.WSAENOBUFS, osdefs.WSAEMFILE:
retFuture.fail(getTransportTooManyError(int(err)))
else:
retFuture.fail(getTransportOsError(err))
return retFuture
var dwBytesReceived = DWORD(0)
let dwReceiveDataLength = DWORD(0)
let dwLocalAddressLength = DWORD(sizeof(Sockaddr_in6) + 16)
let dwRemoteAddressLength = DWORD(sizeof(Sockaddr_in6) + 16)
server.aovl.data = CompletionData(cb: continuationSocket,
udata: cast[pointer](server))
server.apending = true
let res = loop.acceptEx(SocketHandle(server.sock),
SocketHandle(server.asock),
addr server.abuffer[0],
dwReceiveDataLength, dwLocalAddressLength,
dwRemoteAddressLength, addr dwBytesReceived,
cast[POVERLAPPED](addr server.aovl))
if res == osdefs.FALSE:
let err = osLastError()
case int(err)
of osdefs.ERROR_OPERATION_ABORTED:
server.apending = false
retFuture.fail(getServerUseClosedError())
return retFuture
of osdefs.ERROR_IO_PENDING:
discard
of osdefs.WSAECONNRESET, osdefs.WSAECONNABORTED, osdefs.WSAENETDOWN,
osdefs.WSAENETRESET, osdefs.WSAETIMEDOUT:
server.apending = false
retFuture.fail(getConnectionAbortedError(int(err)))
return retFuture
else:
server.apending = false
retFuture.fail(getTransportOsError(err))
return retFuture
retFuture.cancelCallback = cancellationSocket
elif server.local.family in {AddressFamily.Unix}:
# Unix domain sockets emulation via Windows Named pipes part.
server.apending = true
if server.sock == asyncInvalidPipe:
let err = server.errorCode
if int32(err) == osdefs.ERROR_TOO_MANY_OPEN_FILES:
retFuture.fail(getTransportTooManyError())
else:
retFuture.fail(getTransportOsError(err))
return retFuture
server.aovl.data = CompletionData(cb: continuationPipe,
udata: cast[pointer](server))
server.apending = true
let res = connectNamedPipe(HANDLE(server.sock),
cast[POVERLAPPED](addr server.aovl))
if res == 0:
let err = osLastError()
if int32(err) == osdefs.ERROR_OPERATION_ABORTED:
server.apending = false
retFuture.fail(getServerUseClosedError())
return retFuture
elif int32(err) in {osdefs.ERROR_IO_PENDING,
osdefs.ERROR_PIPE_CONNECTED}:
discard
else:
server.apending = false
retFuture.fail(getTransportOsError(err))
return retFuture
retFuture.cancelCallback = cancellationPipe
return retFuture
2018-05-16 08:22:34 +00:00
else:
import ../sendfile
2018-05-16 08:22:34 +00:00
proc isConnResetError(err: OSErrorCode): bool {.inline.} =
(err == OSErrorCode(ECONNRESET)) or (err == OSErrorCode(EPIPE))
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc writeStreamLoop(udata: pointer) =
if isNil(udata):
# TODO this is an if rather than an assert for historical reasons:
# it should not happen unless there are race conditions - but if there
# are race conditions, `transp` might be invalid even if it's not nil:
# it could have been released
return
let
transp = cast[StreamTransport](udata)
fd = SocketHandle(transp.fd)
if WriteClosed in transp.state:
if transp.queue.len > 0:
let error = getTransportUseClosedError()
discard removeWriter2(transp.fd)
failPendingWriteQueue(transp.queue, error)
return
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
# We exit this loop in two ways:
# * The queue is empty: we call removeWriter to disable further callbacks
# * EWOULDBLOCK is returned and we need to wait for a new notification
while len(transp.queue) > 0:
template handleError() =
let err = osLastError()
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
if cint(err) == EINTR:
# Signal happened while writing - try again with all data
transp.queue.addFirst(vector)
continue
if cint(err) in [EWOULDBLOCK, EAGAIN]:
# Socket buffer is full - wait until next write notification - in
# particular, ensure removeWriter is not called
transp.queue.addFirst(vector)
return
# The errors below will clear the write queue, meaning we'll exit the
# loop
if isConnResetError(err):
# Soft error happens which indicates that remote peer got
# disconnected, complete all pending writes in queue with 0.
transp.state.incl({WriteEof})
if not(vector.writer.finished()):
vector.writer.complete(0)
completePendingWriteQueue(transp.queue, 0)
else:
transp.state.incl({WriteError})
let error = getTransportOsError(err)
if not(vector.writer.finished()):
vector.writer.fail(error)
failPendingWriteQueue(transp.queue, error)
var vector = transp.queue.popFirst()
case vector.kind
of VectorKind.DataBuffer:
let res =
case transp.kind
of TransportKind.Socket:
osdefs.send(fd, vector.buf, vector.buflen, MSG_NOSIGNAL)
of TransportKind.Pipe:
osdefs.write(cint(fd), vector.buf, vector.buflen)
else: raiseAssert "Unsupported transport kind: " & $transp.kind
if res >= 0:
if vector.buflen == res:
if not(vector.writer.finished()):
vector.writer.complete(vector.size)
else:
vector.shiftVectorBuffer(res)
transp.queue.addFirst(vector) # Try again with rest of data
else:
handleError()
of VectorKind.DataFile:
var nbytes = cast[int](vector.buf)
let res = sendfile(int(fd), cast[int](vector.buflen),
int(vector.offset), nbytes)
# In case of some errors on some systems, some bytes may have been
# written (see sendfile.nim)
vector.size += nbytes
if res >= 0:
if cast[int](vector.buf) == nbytes:
if not(vector.writer.finished()):
vector.writer.complete(vector.size)
else:
vector.shiftVectorFile(nbytes)
transp.queue.addFirst(vector)
else:
vector.shiftVectorFile(nbytes)
handleError()
# Nothing left in the queue - no need for further write notifications
# All writers are already scheduled, so its impossible to notify about an
# error.
transp.state.incl(WritePaused)
discard removeWriter2(transp.fd)
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc readStreamLoop(udata: pointer) =
if isNil(udata):
# TODO this is an if rather than an assert for historical reasons:
# it should not happen unless there are race conditions - but if there
# are race conditions, `transp` might be invalid even if it's not nil:
# it could have been released
return
let
transp = cast[StreamTransport](udata)
fd = SocketHandle(transp.fd)
if ReadClosed in transp.state:
transp.state.incl({ReadPaused})
transp.completeReader()
else:
if transp.kind == TransportKind.Socket:
while true:
let res = handleEintr(
osdefs.recv(fd, addr transp.buffer[transp.offset],
len(transp.buffer) - transp.offset, cint(0)))
if res < 0:
let err = osLastError()
if int(err) == ECONNRESET:
transp.state.incl({ReadEof, ReadPaused})
let rres = removeReader2(transp.fd)
if rres.isErr():
transp.state.incl(ReadError)
transp.setReadError(rres.error())
else:
transp.state.incl({ReadPaused, ReadError})
transp.setReadError(err)
discard removeReader2(transp.fd)
elif res == 0:
transp.state.incl({ReadEof, ReadPaused})
let rres = removeReader2(transp.fd)
if rres.isErr():
transp.state.incl(ReadError)
transp.setReadError(rres.error())
else:
transp.offset += res
if transp.offset == len(transp.buffer):
transp.state.incl(ReadPaused)
let rres = removeReader2(transp.fd)
if rres.isErr():
transp.state.incl(ReadError)
transp.setReadError(rres.error())
transp.completeReader()
break
elif transp.kind == TransportKind.Pipe:
while true:
let res = handleEintr(
osdefs.read(cint(fd), addr transp.buffer[transp.offset],
len(transp.buffer) - transp.offset))
if res < 0:
let err = osLastError()
transp.state.incl(ReadPaused)
transp.setReadError(err)
discard removeReader2(transp.fd)
elif res == 0:
transp.state.incl({ReadEof, ReadPaused})
let rres = removeReader2(transp.fd)
if rres.isErr():
transp.state.incl(ReadError)
transp.setReadError(rres.error())
else:
transp.offset += res
if transp.offset == len(transp.buffer):
transp.state.incl(ReadPaused)
let rres = removeReader2(transp.fd)
if rres.isErr():
transp.state.incl(ReadError)
transp.setReadError(rres.error())
transp.completeReader()
break
2018-05-16 08:22:34 +00:00
proc newStreamSocketTransport(sock: AsyncFD, bufsize: int,
child: StreamTransport): StreamTransport =
var transp: StreamTransport
if not(isNil(child)):
transp = child
else:
transp = StreamTransport(kind: TransportKind.Socket)
2018-05-21 21:52:57 +00:00
transp.fd = sock
transp.buffer = newSeq[byte](bufsize)
transp.state = {ReadPaused, WritePaused}
transp.queue = initDeque[StreamVector]()
transp.future = newFuture[void]("socket.stream.transport")
GC_ref(transp)
transp
2018-05-16 08:22:34 +00:00
proc newStreamPipeTransport(fd: AsyncFD, bufsize: int,
child: StreamTransport): StreamTransport =
var transp: StreamTransport
if not(isNil(child)):
transp = child
else:
transp = StreamTransport(kind: TransportKind.Pipe)
transp.fd = fd
transp.buffer = newSeq[byte](bufsize)
transp.state = {ReadPaused, WritePaused}
transp.queue = initDeque[StreamVector]()
transp.future = newFuture[void]("pipe.stream.transport")
GC_ref(transp)
transp
2018-05-16 08:22:34 +00:00
proc connect*(address: TransportAddress,
bufferSize = DefaultStreamBufferSize,
child: StreamTransport = nil,
localAddress = TransportAddress(),
flags: set[SocketFlags] = {},
): Future[StreamTransport] =
## Open new connection to remote peer with address ``address`` and create
## new transport object ``StreamTransport`` for established connection.
## ``bufferSize`` - size of internal buffer for transport.
2018-05-16 08:22:34 +00:00
var
saddr: Sockaddr_storage
slen: SockLen
proto: Protocol
var retFuture = newFuture[StreamTransport]("stream.transport.connect")
address.toSAddr(saddr, slen)
proto = Protocol.IPPROTO_TCP
if address.family == AddressFamily.Unix:
# `Protocol` enum is missing `0` value, so we making here cast, until
# `Protocol` enum will not support IPPROTO_IP == 0.
proto = cast[Protocol](0)
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
let sock = createAsyncSocket(address.getDomain(), SockType.SOCK_STREAM,
proto)
2018-05-16 08:22:34 +00:00
if sock == asyncInvalidSocket:
let err = osLastError()
if int(err) == EMFILE:
retFuture.fail(getTransportTooManyError())
else:
retFuture.fail(getTransportOsError(err))
2018-06-07 19:07:17 +00:00
return retFuture
2018-05-16 08:22:34 +00:00
if address.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
if SocketFlags.TcpNoDelay in flags:
if not(setSockOpt(sock, osdefs.IPPROTO_TCP, osdefs.TCP_NODELAY, 1)):
let err = osLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
if SocketFlags.ReuseAddr in flags:
if not(setSockOpt(sock, SOL_SOCKET, SO_REUSEADDR, 1)):
let err = osLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
if SocketFlags.ReusePort in flags:
if not(setSockOpt(sock, SOL_SOCKET, SO_REUSEPORT, 1)):
let err = osLastError()
sock.closeSocket()
retFuture.fail(getTransportOsError(err))
return retFuture
if localAddress != TransportAddress():
if localAddress.family != address.family:
sock.closeSocket()
retFuture.fail(newException(TransportOsError,
"connect local address domain is not equal to target address domain"))
return retFuture
var
localAddr: Sockaddr_storage
localAddrLen: SockLen
localAddress.toSAddr(localAddr, localAddrLen)
if bindSocket(SocketHandle(sock),
cast[ptr SockAddr](addr localAddr), localAddrLen) != 0:
sock.closeSocket()
retFuture.fail(getTransportOsError(osLastError()))
return retFuture
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc continuation(udata: pointer) =
2019-06-20 20:30:41 +00:00
if not(retFuture.finished()):
var err = 0
let res = removeWriter2(sock)
if res.isErr():
discard unregisterAndCloseFd(sock)
retFuture.fail(getTransportOsError(res.error()))
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
return
if not(sock.getSocketError(err)):
discard unregisterAndCloseFd(sock)
retFuture.fail(getTransportOsError(res.error()))
return
if err != 0:
discard unregisterAndCloseFd(sock)
retFuture.fail(getTransportOsError(OSErrorCode(err)))
return
let transp = newStreamSocketTransport(sock, bufferSize, child)
# Start tracking transport
trackStream(transp)
retFuture.complete(transp)
2018-05-16 08:22:34 +00:00
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc cancel(udata: pointer) =
if not(retFuture.finished()):
closeSocket(sock)
2019-06-20 20:30:41 +00:00
2018-05-16 08:22:34 +00:00
while true:
let res = osdefs.connect(SocketHandle(sock),
cast[ptr SockAddr](addr saddr), slen)
2018-05-16 08:22:34 +00:00
if res == 0:
let transp = newStreamSocketTransport(sock, bufferSize, child)
# Start tracking transport
trackStream(transp)
retFuture.complete(transp)
2018-05-16 08:22:34 +00:00
break
else:
let errorCode = osLastError()
2019-06-20 20:30:41 +00:00
# If connect() is interrupted by a signal that is caught while blocked
# waiting to establish a connection, connect() shall fail and set
# connect() to [EINTR], but the connection request shall not be aborted,
# and the connection shall be established asynchronously.
#
# http://www.madore.org/~david/computers/connect-intr.html
if (errorCode == EINPROGRESS) or (errorCode == EINTR):
let res = addWriter2(sock, continuation)
if res.isErr():
discard unregisterAndCloseFd(sock)
retFuture.fail(getTransportOsError(res.error()))
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
return retFuture
2019-06-20 20:30:41 +00:00
retFuture.cancelCallback = cancel
2018-05-16 08:22:34 +00:00
break
else:
discard unregisterAndCloseFd(sock)
retFuture.fail(getTransportOsError(errorCode))
2018-05-16 08:22:34 +00:00
break
return retFuture
proc acceptLoop(udata: pointer) =
if isNil(udata):
# TODO this is an if rather than an assert for historical reasons:
# it should not happen unless there are race conditions - but if there
# are race conditions, `transp` might be invalid even if it's not nil:
# it could have been released
return
2018-05-16 08:22:34 +00:00
var
saddr: Sockaddr_storage
slen: SockLen
let server = cast[StreamServer](udata)
if server.status in {ServerStatus.Stopped, ServerStatus.Closed}:
return
let
flags = {DescriptorFlag.CloseOnExec, DescriptorFlag.NonBlock}
sres = acceptConn(cint(server.sock), cast[ptr SockAddr](addr saddr),
addr slen, flags)
if sres.isOk():
let sock = AsyncFD(sres.get())
let rres = register2(sock)
if rres.isOk():
let ntransp =
if not(isNil(server.init)):
let transp = server.init(server, sock)
newStreamSocketTransport(sock, server.bufferSize, transp)
else:
newStreamSocketTransport(sock, server.bufferSize, nil)
trackStream(ntransp)
asyncSpawn server.function(server, ntransp)
2018-05-16 08:22:34 +00:00
else:
# Client was accepted, so we not going to raise assertion, but
# we need to close the socket.
discard closeFd(cint(sock))
else:
let errorCode = sres.error()
if errorCode != EAGAIN:
# This EAGAIN error appears only when server get closed, while
# acceptLoop() reader callback is already scheduled.
raiseOsDefect(errorCode, "acceptLoop(): Unable to accept connection")
2018-05-16 08:22:34 +00:00
proc resumeAccept(server: StreamServer): Result[void, OSErrorCode] =
addReader2(server.sock, acceptLoop, cast[pointer](server))
proc pauseAccept(server: StreamServer): Result[void, OSErrorCode] =
removeReader2(server.sock)
2018-05-16 08:22:34 +00:00
proc resumeRead(transp: StreamTransport): Result[void, OSErrorCode] =
if ReadPaused in transp.state:
? addReader2(transp.fd, readStreamLoop, cast[pointer](transp))
transp.state.excl(ReadPaused)
ok()
proc resumeWrite(transp: StreamTransport): Result[void, OSErrorCode] =
if transp.queue.len() == 1:
# writeStreamLoop keeps writing until queue is empty - we should not call
# resumeWrite under any other condition than when the items are
# added to a queue - if the flag is not set here, it means that the socket
# was not removed from write notifications at the right time, and this
# would mean an imbalance in registration and deregistration
doAssert(WritePaused in transp.state)
? addWriter2(transp.fd, writeStreamLoop, cast[pointer](transp))
transp.state.excl(WritePaused)
ok()
2018-05-16 08:22:34 +00:00
proc accept*(server: StreamServer): Future[StreamTransport] =
var retFuture = newFuture[StreamTransport]("stream.server.accept")
doAssert(server.status != ServerStatus.Running,
"You could not use accept() if server was started with start()")
if server.status == ServerStatus.Closed:
retFuture.fail(getServerUseClosedError())
return retFuture
proc continuation(udata: pointer) {.gcsafe.} =
var
saddr: Sockaddr_storage
slen: SockLen
if not(retFuture.finished()):
if server.status in {ServerStatus.Stopped, ServerStatus.Closed}:
retFuture.fail(getServerUseClosedError())
else:
let
flags = {DescriptorFlag.CloseOnExec, DescriptorFlag.NonBlock}
sres = acceptConn(cint(server.sock), cast[ptr SockAddr](addr saddr),
addr slen, flags)
if sres.isErr():
let errorCode = sres.error()
if errorCode == EAGAIN:
# This error appears only when server get closed, while accept()
# continuation is already scheduled.
retFuture.fail(getServerUseClosedError())
elif cint(errorCode) in {EMFILE, ENFILE, ENOBUFS, ENOMEM}:
retFuture.fail(getTransportTooManyError(cint(errorCode)))
elif cint(errorCode) in {ECONNABORTED, EPERM, ETIMEDOUT}:
retFuture.fail(getConnectionAbortedError(cint(errorCode)))
else:
retFuture.fail(getTransportOsError(errorCode))
# Error is already happened so we ignore removeReader2() errors.
discard removeReader2(server.sock)
else:
let
sock = AsyncFD(sres.get())
rres = register2(sock)
if rres.isOk():
let res = removeReader2(server.sock)
if res.isOk():
let ntransp =
if not(isNil(server.init)):
let transp = server.init(server, sock)
newStreamSocketTransport(sock, server.bufferSize, transp)
else:
newStreamSocketTransport(sock, server.bufferSize, nil)
# Start tracking transport
trackStream(ntransp)
retFuture.complete(ntransp)
else:
discard closeFd(cint(sock))
let errorMsg = osErrorMsg(res.error())
retFuture.fail(getConnectionAbortedError(errorMsg))
else:
# Error is already happened so we ignore errors.
discard removeReader2(server.sock)
discard closeFd(cint(sock))
let errorMsg = osErrorMsg(rres.error())
retFuture.fail(getConnectionAbortedError(errorMsg))
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc cancellation(udata: pointer) =
if not(retFuture.finished()):
discard removeReader2(server.sock)
let res = addReader2(server.sock, continuation, nil)
if res.isErr():
retFuture.fail(getTransportOsError(res.error()))
else:
retFuture.cancelCallback = cancellation
return retFuture
proc start2*(server: StreamServer): Result[void, OSErrorCode] =
2018-05-21 21:52:57 +00:00
## Starts ``server``.
doAssert(not(isNil(server.function)), "You should not start the server " &
"unless you have processing callback configured!")
if server.status == ServerStatus.Starting:
? server.resumeAccept()
server.status = ServerStatus.Running
ok()
2018-05-16 08:22:34 +00:00
proc stop2*(server: StreamServer): Result[void, OSErrorCode] =
2018-05-22 21:03:13 +00:00
## Stops ``server``.
if server.status == ServerStatus.Running:
if not(isNil(server.function)):
? server.pauseAccept()
server.status = ServerStatus.Stopped
elif server.status == ServerStatus.Starting:
server.status = ServerStatus.Stopped
ok()
proc start*(server: StreamServer) {.raises: [Defect, TransportOsError].} =
## Starts ``server``.
let res = start2(server)
if res.isErr(): raiseTransportOsError(res.error())
proc stop*(server: StreamServer) {.raises: [Defect, TransportOsError].} =
## Stops ``server``.
let res = stop2(server)
if res.isErr(): raiseTransportOsError(res.error())
2018-05-16 08:22:34 +00:00
proc join*(server: StreamServer): Future[void] =
## Waits until ``server`` is not closed.
var retFuture = newFuture[void]("stream.transport.server.join")
2019-06-20 20:30:41 +00:00
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc continuation(udata: pointer) =
2019-06-20 20:30:41 +00:00
retFuture.complete()
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc cancel(udata: pointer) =
2019-06-20 20:30:41 +00:00
server.loopFuture.removeCallback(continuation, cast[pointer](retFuture))
if not(server.loopFuture.finished()):
server.loopFuture.addCallback(continuation, cast[pointer](retFuture))
retFuture.cancelCallback = cancel
else:
retFuture.complete()
return retFuture
2018-05-21 21:52:57 +00:00
proc connect*(address: TransportAddress,
bufferSize = DefaultStreamBufferSize,
child: StreamTransport = nil,
flags: set[TransportFlags],
localAddress = TransportAddress()): Future[StreamTransport] =
# Retro compatibility with TransportFlags
var mappedFlags: set[SocketFlags]
if TcpNoDelay in flags: mappedFlags.incl(SocketFlags.TcpNoDelay)
address.connect(bufferSize, child, localAddress, mappedFlags)
proc close*(server: StreamServer) =
2018-05-21 21:52:57 +00:00
## Release ``server`` resources.
##
## Please note that release of resources is not completed immediately, to be
## sure all resources got released please use ``await server.join()``.
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
proc continuation(udata: pointer) =
2021-01-22 13:02:13 +00:00
# Stop tracking server
if not(server.loopFuture.finished()):
server.clean()
if server.status in {ServerStatus.Starting, ServerStatus.Stopped}:
server.status = ServerStatus.Closed
when defined(windows):
if server.local.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
if server.apending:
server.asock.closeSocket()
server.apending = false
server.sock.closeSocket(continuation)
elif server.local.family in {AddressFamily.Unix}:
if NoPipeFlash notin server.flags:
discard flushFileBuffers(HANDLE(server.sock))
discard disconnectNamedPipe(HANDLE(server.sock))
server.sock.closeHandle(continuation)
else:
server.sock.closeSocket(continuation)
proc closeWait*(server: StreamServer): Future[void] =
## Close server ``server`` and release all resources.
2018-09-13 18:15:57 +00:00
server.close()
result = server.join()
2018-05-16 08:22:34 +00:00
proc createStreamServer*(host: TransportAddress,
cbproc: StreamCallback,
flags: set[ServerFlags] = {},
2018-05-16 08:22:34 +00:00
sock: AsyncFD = asyncInvalidSocket,
backlog: int = 100,
bufferSize: int = DefaultStreamBufferSize,
child: StreamServer = nil,
init: TransportInitCallback = nil,
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
udata: pointer = nil): StreamServer {.
raises: [Defect, TransportOsError].} =
2018-05-28 23:35:15 +00:00
## Create new TCP stream server.
##
## ``host`` - address to which server will be bound.
## ``flags`` - flags to apply to server socket.
## ``cbproc`` - callback function which will be called, when new client
## connection will be established.
2018-05-28 23:35:15 +00:00
## ``sock`` - user-driven socket to use.
## ``backlog`` - number of outstanding connections in the socket's listen
## queue.
## ``bufferSize`` - size of internal buffer for transport.
## ``child`` - existing object ``StreamServer``object to initialize, can be
## used to initalize ``StreamServer`` inherited objects.
2018-05-28 23:35:15 +00:00
## ``udata`` - user-defined pointer.
2018-05-16 08:22:34 +00:00
var
saddr: Sockaddr_storage
slen: SockLen
serverSocket: AsyncFD
localAddress: TransportAddress
2018-05-16 08:22:34 +00:00
2022-01-04 17:00:17 +00:00
when defined(nimdoc):
discard
elif defined(windows):
# Windows
if host.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
if sock == asyncInvalidSocket:
serverSocket = createAsyncSocket(host.getDomain(),
SockType.SOCK_STREAM,
Protocol.IPPROTO_TCP)
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
if serverSocket == asyncInvalidSocket:
raiseTransportOsError(osLastError())
else:
let bres = setDescriptorBlocking(SocketHandle(sock), false)
if bres.isErr():
raiseTransportOsError(bres.error())
let wres = register2(sock)
if wres.isErr():
raiseTransportOsError(wres.error())
serverSocket = sock
# SO_REUSEADDR is not useful for Unix domain sockets.
if ServerFlags.ReuseAddr in flags:
if not(setSockOpt(serverSocket, osdefs.SOL_SOCKET,
osdefs.SO_REUSEADDR, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
if ServerFlags.ReusePort in flags:
if not(setSockOpt(serverSocket, osdefs.SOL_SOCKET,
osdefs.SO_REUSEPORT, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
# TCP flags are not useful for Unix domain sockets.
if ServerFlags.TcpNoDelay in flags:
if not(setSockOpt(serverSocket, osdefs.IPPROTO_TCP,
osdefs.TCP_NODELAY, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
host.toSAddr(saddr, slen)
if osdefs.bindSocket(SocketHandle(serverSocket),
cast[ptr SockAddr](addr saddr), slen) != 0:
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
slen = SockLen(sizeof(saddr))
if getsockname(SocketHandle(serverSocket), cast[ptr SockAddr](addr saddr),
addr slen) != 0:
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
fromSAddr(addr saddr, slen, localAddress)
if listen(SocketHandle(serverSocket), cint(backlog)) != 0:
let err = osLastError()
if sock == asyncInvalidSocket:
discard closeFd(SocketHandle(serverSocket))
raiseTransportOsError(err)
elif host.family == AddressFamily.Unix:
serverSocket = AsyncFD(0)
else:
# Posix
if sock == asyncInvalidSocket:
var proto = Protocol.IPPROTO_TCP
if host.family == AddressFamily.Unix:
# `Protocol` enum is missing `0` value, so we making here cast, until
# `Protocol` enum will not support IPPROTO_IP == 0.
proto = cast[Protocol](0)
serverSocket = createAsyncSocket(host.getDomain(),
SockType.SOCK_STREAM,
proto)
if serverSocket == asyncInvalidSocket:
raiseTransportOsError(osLastError())
else:
let bres = setDescriptorFlags(cint(sock), true, true)
if bres.isErr():
raiseTransportOsError(osLastError())
let rres = register2(sock)
if rres.isErr():
raiseTransportOsError(osLastError())
serverSocket = sock
if host.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
# SO_REUSEADDR and SO_REUSEPORT are not useful for Unix domain sockets.
if ServerFlags.ReuseAddr in flags:
if not(setSockOpt(serverSocket, SOL_SOCKET, SO_REUSEADDR, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard unregisterAndCloseFd(serverSocket)
raiseTransportOsError(err)
if ServerFlags.ReusePort in flags:
if not(setSockOpt(serverSocket, SOL_SOCKET, SO_REUSEPORT, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard unregisterAndCloseFd(serverSocket)
raiseTransportOsError(err)
# TCP flags are not useful for Unix domain sockets.
if ServerFlags.TcpNoDelay in flags:
if not(setSockOpt(serverSocket, osdefs.IPPROTO_TCP,
osdefs.TCP_NODELAY, 1)):
let err = osLastError()
if sock == asyncInvalidSocket:
discard unregisterAndCloseFd(serverSocket)
raiseTransportOsError(err)
elif host.family in {AddressFamily.Unix}:
# We do not care about result here, because if file cannot be removed,
# `bindSocket` will return EADDRINUSE.
discard osdefs.unlink(cast[cstring](unsafeAddr host.address_un[0]))
host.toSAddr(saddr, slen)
if osdefs.bindSocket(SocketHandle(serverSocket),
cast[ptr SockAddr](addr saddr), slen) != 0:
2018-05-16 08:22:34 +00:00
let err = osLastError()
if sock == asyncInvalidSocket:
discard unregisterAndCloseFd(serverSocket)
raiseTransportOsError(err)
2018-05-16 08:22:34 +00:00
# Obtain real address
slen = SockLen(sizeof(saddr))
if getsockname(SocketHandle(serverSocket), cast[ptr SockAddr](addr saddr),
addr slen) != 0:
let err = osLastError()
if sock == asyncInvalidSocket:
serverSocket.closeSocket()
raiseTransportOsError(err)
fromSAddr(addr saddr, slen, localAddress)
if listen(SocketHandle(serverSocket), cint(backlog)) != 0:
let err = osLastError()
if sock == asyncInvalidSocket:
discard unregisterAndCloseFd(serverSocket)
raiseTransportOsError(err)
var sres = if not(isNil(child)): child else: StreamServer()
sres.sock = serverSocket
sres.flags = flags
sres.function = cbproc
sres.init = init
sres.bufferSize = bufferSize
sres.status = Starting
sres.loopFuture = newFuture[void]("stream.transport.server")
sres.udata = udata
if localAddress.family == AddressFamily.None:
sres.local = host
else:
sres.local = localAddress
when defined(windows):
var cb: CallbackFunc
if host.family in {AddressFamily.IPv4, AddressFamily.IPv6}:
cb = acceptLoop
elif host.family == AddressFamily.Unix:
cb = acceptPipeLoop
if not(isNil(cbproc)):
sres.aovl.data = CompletionData(cb: cb,
udata: cast[pointer](sres))
else:
if host.family == AddressFamily.Unix:
sres.sock =
block:
let res = sres.createAcceptPipe()
if res.isErr():
raiseTransportOsError(res.error())
res.get()
sres.domain = host.getDomain()
sres.apending = false
# Start tracking server
trackServer(sres)
GC_ref(sres)
sres
2018-05-16 08:22:34 +00:00
proc createStreamServer*(host: TransportAddress,
flags: set[ServerFlags] = {},
sock: AsyncFD = asyncInvalidSocket,
backlog: int = 100,
bufferSize: int = DefaultStreamBufferSize,
child: StreamServer = nil,
init: TransportInitCallback = nil,
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
udata: pointer = nil): StreamServer {.
raises: [Defect, CatchableError].} =
createStreamServer(host, nil, flags, sock, backlog, bufferSize,
child, init, cast[pointer](udata))
proc createStreamServer*[T](host: TransportAddress,
cbproc: StreamCallback,
flags: set[ServerFlags] = {},
udata: ref T,
sock: AsyncFD = asyncInvalidSocket,
backlog: int = 100,
bufferSize: int = DefaultStreamBufferSize,
child: StreamServer = nil,
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
init: TransportInitCallback = nil): StreamServer {.
raises: [Defect, CatchableError].} =
var fflags = flags + {GCUserData}
GC_ref(udata)
createStreamServer(host, cbproc, fflags, sock, backlog, bufferSize,
child, init, cast[pointer](udata))
proc createStreamServer*[T](host: TransportAddress,
flags: set[ServerFlags] = {},
udata: ref T,
sock: AsyncFD = asyncInvalidSocket,
backlog: int = 100,
bufferSize: int = DefaultStreamBufferSize,
child: StreamServer = nil,
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
init: TransportInitCallback = nil): StreamServer {.
raises: [Defect, CatchableError].} =
var fflags = flags + {GCUserData}
GC_ref(udata)
createStreamServer(host, nil, fflags, sock, backlog, bufferSize,
child, init, cast[pointer](udata))
proc getUserData*[T](server: StreamServer): T {.inline.} =
## Obtain user data stored in ``server`` object.
result = cast[T](server.udata)
template fastWrite(transp: auto, pbytes: var ptr byte, rbytes: var int,
nbytes: int) =
# On windows, the write could be initiated here if there is no other write
# ongoing, but the queue is still needed due to the mechanics of iocp
2022-01-04 17:00:17 +00:00
when not defined(windows) and not defined(nimdoc):
if transp.queue.len == 0:
while rbytes > 0:
let res =
case transp.kind
of TransportKind.Socket:
osdefs.send(SocketHandle(transp.fd), pbytes, rbytes,
MSG_NOSIGNAL)
of TransportKind.Pipe:
osdefs.write(cint(transp.fd), pbytes, rbytes)
else:
raiseAssert "Unsupported transport kind: " & $transp.kind
if res > 0:
pbytes = cast[ptr byte](cast[uint](pbytes) + cast[uint](res))
rbytes -= res
if rbytes == 0:
retFuture.complete(nbytes)
return retFuture
# Not all bytes written - keep going
else:
let err = osLastError()
if cint(err) in [EAGAIN, EWOULDBLOCK]:
break # No bytes written, add to queue
if cint(err) == EINTR:
continue
if isConnResetError(err):
transp.state.incl({WriteEof})
retFuture.complete(0)
return retFuture
else:
transp.state.incl({WriteError})
let error = getTransportOsError(err)
retFuture.fail(error)
return retFuture
2018-05-16 08:22:34 +00:00
proc write*(transp: StreamTransport, pbytes: pointer,
nbytes: int): Future[int] =
## Write data from buffer ``pbytes`` with size ``nbytes`` using transport
2018-05-22 21:03:13 +00:00
## ``transp``.
var retFuture = newFuture[int]("stream.transport.write(pointer)")
transp.checkClosed(retFuture)
2019-10-23 11:13:23 +00:00
transp.checkWriteEof(retFuture)
var
pbytes = cast[ptr byte](pbytes)
rbytes = nbytes # Remaining bytes
fastWrite(transp, pbytes, rbytes, nbytes)
var vector = StreamVector(kind: DataBuffer, writer: retFuture,
buf: pbytes, buflen: rbytes, size: nbytes)
2018-05-16 08:22:34 +00:00
transp.queue.addLast(vector)
let wres = transp.resumeWrite()
if wres.isErr():
retFuture.fail(getTransportOsError(wres.error()))
return retFuture
proc write*(transp: StreamTransport, msg: sink string,
msglen = -1): Future[int] =
## Write data from string ``msg`` using transport ``transp``.
var retFuture = newFutureStr[int]("stream.transport.write(string)")
transp.checkClosed(retFuture)
2019-10-23 11:13:23 +00:00
transp.checkWriteEof(retFuture)
let
nbytes = if msglen <= 0: len(msg) else: msglen
var
pbytes = cast[ptr byte](unsafeAddr msg[0])
rbytes = nbytes
fastWrite(transp, pbytes, rbytes, nbytes)
let
written = nbytes - rbytes # In case fastWrite wrote some
pbytes =
when declared(shallowCopy):
if not(isLiteral(msg)):
shallowCopy(retFuture.gcholder, msg)
cast[ptr byte](addr retFuture.gcholder[written])
else:
retFuture.gcholder = msg[written ..< nbytes]
cast[ptr byte](addr retFuture.gcholder[0])
else:
retFuture.gcholder = msg[written ..< nbytes]
cast[ptr byte](addr retFuture.gcholder[0])
var vector = StreamVector(kind: DataBuffer, writer: retFuture,
buf: pbytes, buflen: rbytes, size: nbytes)
transp.queue.addLast(vector)
let wres = transp.resumeWrite()
if wres.isErr():
retFuture.fail(getTransportOsError(wres.error()))
return retFuture
proc write*[T](transp: StreamTransport, msg: sink seq[T],
msglen = -1): Future[int] =
## Write sequence ``msg`` using transport ``transp``.
var retFuture = newFutureSeq[int, T]("stream.transport.write(seq)")
transp.checkClosed(retFuture)
2019-10-23 11:13:23 +00:00
transp.checkWriteEof(retFuture)
let
nbytes = if msglen <= 0: (len(msg) * sizeof(T)) else: (msglen * sizeof(T))
var
pbytes = cast[ptr byte](unsafeAddr msg[0])
rbytes = nbytes
fastWrite(transp, pbytes, rbytes, nbytes)
let
written = nbytes - rbytes # In case fastWrite wrote some
pbytes =
when declared(shallowCopy):
if not(isLiteral(msg)):
shallowCopy(retFuture.gcholder, msg)
cast[ptr byte](addr retFuture.gcholder[written])
else:
retFuture.gcholder = msg[written ..< nbytes]
cast[ptr byte](addr retFuture.gcholder[0])
else:
retFuture.gcholder = msg[written ..< nbytes]
cast[ptr byte](addr retFuture.gcholder[0])
var vector = StreamVector(kind: DataBuffer, writer: retFuture,
buf: pbytes, buflen: rbytes, size: nbytes)
transp.queue.addLast(vector)
let wres = transp.resumeWrite()
if wres.isErr():
retFuture.fail(getTransportOsError(wres.error()))
return retFuture
2018-05-21 21:52:57 +00:00
proc writeFile*(transp: StreamTransport, handle: int,
offset: uint = 0, size: int = 0): Future[int] =
2018-05-22 21:03:13 +00:00
## Write data from file descriptor ``handle`` to transport ``transp``.
##
2018-05-22 21:03:13 +00:00
## You can specify starting ``offset`` in opened file and number of bytes
## to transfer from file to transport via ``size``.
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
var retFuture = newFuture[int]("stream.transport.writeFile")
when defined(windows):
if transp.kind != TransportKind.Socket:
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
retFuture.fail(newException(
TransportNoSupport, "writeFile() is not supported!"))
return retFuture
transp.checkClosed(retFuture)
2019-10-23 11:13:23 +00:00
transp.checkWriteEof(retFuture)
var vector = StreamVector(kind: DataFile, writer: retFuture,
2018-05-21 21:52:57 +00:00
buf: cast[pointer](size), offset: offset,
buflen: handle)
transp.queue.addLast(vector)
let wres = transp.resumeWrite()
if wres.isErr():
retFuture.fail(getTransportOsError(wres.error()))
return retFuture
2018-05-16 08:22:34 +00:00
proc atEof*(transp: StreamTransport): bool {.inline.} =
## Returns ``true`` if ``transp`` is at EOF.
(transp.offset == 0) and (ReadEof in transp.state) and
(ReadPaused in transp.state)
template readLoop(name, body: untyped): untyped =
# Read data until a predicate is satisfied - the body should return a tuple
# signalling how many bytes have been processed and whether we're done reading
checkClosed(transp)
checkPending(transp)
while true:
if ReadClosed in transp.state:
raise newException(TransportUseClosedError,
"Attempt to read data from closed stream")
if transp.offset == 0:
# We going to raise an error, only if transport buffer is empty.
if ReadError in transp.state:
raise transp.getError()
let (consumed, done) = body
transp.shiftBuffer(consumed)
if done:
break
else:
checkPending(transp)
var fut = newFuture[void](name)
transp.reader = fut
let res = resumeRead(transp)
if res.isErr():
let errorCode = res.error()
when defined(windows):
# This assertion could be changed, because at this moment
# resumeRead() could not return any error.
raiseOsDefect(errorCode, "readLoop(): Unable to resume reading")
else:
transp.reader.complete()
if errorCode == ESRCH:
# ESRCH 3 "No such process"
# This error could be happened on pipes only, when process which
# owns and communicates through this pipe (stdin, stdout, stderr) is
# already dead. In such case we need to send notification that this
# pipe is at EOF.
transp.state.incl({ReadEof, ReadPaused})
else:
raiseTransportOsError(errorCode)
else:
await fut
2018-05-16 08:22:34 +00:00
proc readExactly*(transp: StreamTransport, pbytes: pointer,
2018-05-21 21:52:57 +00:00
nbytes: int) {.async.} =
2018-05-22 21:03:13 +00:00
## Read exactly ``nbytes`` bytes from transport ``transp`` and store it to
## ``pbytes``. ``pbytes`` must not be ``nil`` pointer and ``nbytes`` should
## be Natural.
##
## If ``nbytes == 0`` this operation will return immediately.
##
2018-05-22 21:03:13 +00:00
## If EOF is received and ``nbytes`` is not yet readed, the procedure
## will raise ``TransportIncompleteError``, potentially with some bytes
## already written.
doAssert(not(isNil(pbytes)), "pbytes must not be nil")
doAssert(nbytes >= 0, "nbytes must be non-negative integer")
if nbytes == 0:
return
2018-05-16 08:22:34 +00:00
var index = 0
var pbuffer = cast[ptr UncheckedArray[byte]](pbytes)
readLoop("stream.transport.readExactly"):
2018-05-16 08:22:34 +00:00
if transp.offset == 0:
if transp.atEof():
2018-05-16 08:22:34 +00:00
raise newException(TransportIncompleteError, "Data incomplete!")
let count = min(nbytes - index, transp.offset)
if count > 0:
copyMem(addr pbuffer[index], addr(transp.buffer[0]), count)
index += count
(consumed: count, done: index == nbytes)
2018-05-16 08:22:34 +00:00
proc readOnce*(transp: StreamTransport, pbytes: pointer,
nbytes: int): Future[int] {.async.} =
## Perform one read operation on transport ``transp``.
##
## If internal buffer is not empty, ``nbytes`` bytes will be transferred from
## internal buffer, otherwise it will wait until some bytes will be received.
doAssert(not(isNil(pbytes)), "pbytes must not be nil")
doAssert(nbytes > 0, "nbytes must be positive integer")
var count = 0
readLoop("stream.transport.readOnce"):
2018-05-16 08:22:34 +00:00
if transp.offset == 0:
(0, transp.atEof())
2018-05-16 08:22:34 +00:00
else:
count = min(transp.offset, nbytes)
copyMem(pbytes, addr(transp.buffer[0]), count)
(count, true)
return count
2018-05-16 08:22:34 +00:00
proc readUntil*(transp: StreamTransport, pbytes: pointer, nbytes: int,
sep: seq[byte]): Future[int] {.async.} =
2018-05-22 21:03:13 +00:00
## Read data from the transport ``transp`` until separator ``sep`` is found.
##
2018-05-22 21:03:13 +00:00
## On success, the data and separator will be removed from the internal
## buffer (consumed). Returned data will include the separator at the end.
##
2018-05-28 23:35:15 +00:00
## If EOF is received, and `sep` was not found, procedure will raise
## ``TransportIncompleteError``.
##
2018-05-28 23:35:15 +00:00
## If ``nbytes`` bytes has been received and `sep` was not found, procedure
## will raise ``TransportLimitError``.
##
2018-05-28 23:35:15 +00:00
## Procedure returns actual number of bytes read.
doAssert(not(isNil(pbytes)), "pbytes must not be nil")
doAssert(len(sep) > 0, "separator must not be empty")
doAssert(nbytes >= 0, "nbytes must be non-negative integer")
if nbytes == 0:
raise newException(TransportLimitError, "Limit reached!")
var pbuffer = cast[ptr UncheckedArray[byte]](pbytes)
2018-05-16 08:22:34 +00:00
var state = 0
var k = 0
readLoop("stream.transport.readUntil"):
if transp.atEof():
raise newException(TransportIncompleteError, "Data incomplete!")
2018-05-16 08:22:34 +00:00
var index = 0
2018-05-16 08:22:34 +00:00
while index < transp.offset:
if k >= nbytes:
raise newException(TransportLimitError, "Limit reached!")
2018-05-16 08:22:34 +00:00
let ch = transp.buffer[index]
inc(index)
pbuffer[k] = ch
inc(k)
2018-05-16 08:22:34 +00:00
if sep[state] == ch:
inc(state)
if state == len(sep):
break
2018-05-16 08:22:34 +00:00
else:
state = 0
(index, state == len(sep))
return k
2018-05-16 08:22:34 +00:00
proc readLine*(transp: StreamTransport, limit = 0,
sep = "\r\n"): Future[string] {.async.} =
2018-05-22 21:03:13 +00:00
## Read one line from transport ``transp``, where "line" is a sequence of
## bytes ending with ``sep`` (default is "\r\n").
##
2018-05-22 21:03:13 +00:00
## If EOF is received, and ``sep`` was not found, the method will return the
## partial read bytes.
##
2018-05-22 21:03:13 +00:00
## If the EOF was received and the internal buffer is empty, return an
## empty string.
##
2018-05-22 21:03:13 +00:00
## If ``limit`` more then 0, then read is limited to ``limit`` bytes.
let lim = if limit <= 0: -1 else: limit
2018-05-16 08:22:34 +00:00
var state = 0
readLoop("stream.transport.readLine"):
if transp.atEof():
(0, true)
else:
var index = 0
while index < transp.offset:
let ch = char(transp.buffer[index])
index += 1
if sep[state] == ch:
inc(state)
if state == len(sep):
break
else:
if state != 0:
if limit > 0:
let missing = min(state, lim - len(result) - 1)
result.add(sep[0 ..< missing])
else:
result.add(sep[0 ..< state])
state = 0
2018-05-16 08:22:34 +00:00
result.add(ch)
if len(result) == lim:
break
2018-05-16 08:22:34 +00:00
(index, (state == len(sep)) or (lim == len(result)))
proc read*(transp: StreamTransport): Future[seq[byte]] {.async.} =
## Read all bytes from transport ``transp``.
##
## This procedure allocates buffer seq[byte] and return it as result.
readLoop("stream.transport.read"):
if transp.atEof():
(0, true)
2018-05-16 08:22:34 +00:00
else:
result.add(transp.buffer.toOpenArray(0, transp.offset - 1))
(transp.offset, false)
proc read*(transp: StreamTransport, n: int): Future[seq[byte]] {.async.} =
## Read all bytes (n <= 0) or exactly `n` bytes from transport ``transp``.
##
2018-05-28 23:35:15 +00:00
## This procedure allocates buffer seq[byte] and return it as result.
if n <= 0:
return await transp.read()
else:
readLoop("stream.transport.read"):
if transp.atEof():
(0, true)
2018-05-16 08:22:34 +00:00
else:
let count = min(transp.offset, n - len(result))
result.add(transp.buffer.toOpenArray(0, count - 1))
(count, len(result) == n)
2018-05-16 08:22:34 +00:00
proc consume*(transp: StreamTransport): Future[int] {.async.} =
## Consume all bytes from transport ``transp`` and discard it.
##
## Return number of bytes actually consumed and discarded.
readLoop("stream.transport.consume"):
if transp.atEof():
(0, true)
else:
result += transp.offset
(transp.offset, false)
proc consume*(transp: StreamTransport, n: int): Future[int] {.async.} =
## Consume all bytes (n <= 0) or ``n`` bytes from transport ``transp`` and
## discard it.
##
## Return number of bytes actually consumed and discarded.
if n <= 0:
return await transp.consume()
else:
readLoop("stream.transport.consume"):
if transp.atEof():
(0, true)
else:
let count = min(transp.offset, n - result)
result += count
(count, result == n)
proc readMessage*(transp: StreamTransport,
predicate: ReadMessagePredicate) {.async.} =
## Read all bytes from transport ``transp`` until ``predicate`` callback
## will not be satisfied.
##
## ``predicate`` callback should return tuple ``(consumed, result)``, where
## ``consumed`` is the number of bytes processed and ``result`` is a
## completion flag (``true`` if readMessage() should stop reading data,
## or ``false`` if readMessage() should continue to read data from transport).
##
## ``predicate`` callback must copy all the data from ``data`` array and
## return number of bytes it is going to consume.
## ``predicate`` callback will receive (zero-length) openArray, if transport
## is at EOF.
readLoop("stream.transport.readMessage"):
if transp.offset == 0:
if transp.atEof():
predicate([])
else:
# Case, when transport's buffer is not yet filled with data.
(0, false)
else:
predicate(transp.buffer.toOpenArray(0, transp.offset - 1))
2018-05-16 08:22:34 +00:00
proc join*(transp: StreamTransport): Future[void] =
2018-05-16 08:22:34 +00:00
## Wait until ``transp`` will not be closed.
var retFuture = newFuture[void]("stream.transport.join")
2019-06-20 20:30:41 +00:00
proc continuation(udata: pointer) {.gcsafe.} =
retFuture.complete()
proc cancel(udata: pointer) {.gcsafe.} =
transp.future.removeCallback(continuation, cast[pointer](retFuture))
if not(transp.future.finished()):
transp.future.addCallback(continuation, cast[pointer](retFuture))
retFuture.cancelCallback = cancel
else:
retFuture.complete()
return retFuture
2018-05-16 08:22:34 +00:00
proc close*(transp: StreamTransport) =
## Closes and frees resources of transport ``transp``.
##
## Please note that release of resources is not completed immediately, to be
## sure all resources got released please use ``await transp.join()``.
2019-06-20 20:30:41 +00:00
proc continuation(udata: pointer) {.gcsafe.} =
transp.clean()
if {ReadClosed, WriteClosed} * transp.state == {}:
transp.state.incl({WriteClosed, ReadClosed})
2018-05-16 08:22:34 +00:00
when defined(windows):
if transp.kind == TransportKind.Pipe:
if WinServerPipe in transp.flags:
if WinNoPipeFlash notin transp.flags:
discard flushFileBuffers(HANDLE(transp.fd))
discard disconnectNamedPipe(HANDLE(transp.fd))
else:
if WinNoPipeFlash notin transp.flags:
discard flushFileBuffers(HANDLE(transp.fd))
if ReadPaused in transp.state:
# If readStreamLoop() is not running we need to finish in
# continuation step.
closeHandle(transp.fd, continuation)
else:
# If readStreamLoop() is running, it will be properly finished inside
# of readStreamLoop().
closeHandle(transp.fd)
elif transp.kind == TransportKind.Socket:
if ReadPaused in transp.state:
# If readStreamLoop() is not running we need to finish in
# continuation step.
closeSocket(transp.fd, continuation)
else:
# If readStreamLoop() is running, it will be properly finished inside
# of readStreamLoop().
closeSocket(transp.fd)
else:
if transp.kind == TransportKind.Pipe:
closeHandle(transp.fd, continuation)
elif transp.kind == TransportKind.Socket:
closeSocket(transp.fd, continuation)
proc closeWait*(transp: StreamTransport): Future[void] =
## Close and frees resources of transport ``transp``.
2018-09-13 18:15:57 +00:00
transp.close()
transp.join()
proc closed*(transp: StreamTransport): bool {.inline.} =
## Returns ``true`` if transport in closed state.
({ReadClosed, WriteClosed} * transp.state != {})
proc finished*(transp: StreamTransport): bool {.inline.} =
## Returns ``true`` if transport in finished (EOF) state.
({ReadEof, WriteEof} * transp.state != {})
proc failed*(transp: StreamTransport): bool {.inline.} =
## Returns ``true`` if transport in error state.
({ReadError, WriteError} * transp.state != {})
proc running*(transp: StreamTransport): bool {.inline.} =
## Returns ``true`` if transport is still pending.
({ReadClosed, ReadEof, ReadError,
WriteClosed, WriteEof, WriteError} * transp.state == {})
proc fromPipe2*(fd: AsyncFD, child: StreamTransport = nil,
bufferSize = DefaultStreamBufferSize
): Result[StreamTransport, OSErrorCode] =
## Create new transport object using pipe's file descriptor.
##
## ``bufferSize`` is size of internal buffer for transport.
? register2(fd)
var res = newStreamPipeTransport(fd, bufferSize, child)
# Start tracking transport
trackStream(res)
ok(res)
proc fromPipe*(fd: AsyncFD, child: StreamTransport = nil,
exception tracking (#166) * exception tracking This PR adds minimal exception tracking to chronos, moving the goalpost one step further. In particular, it becomes invalid to raise exceptions from `callSoon` callbacks: this is critical for writing correct error handling because there's no reasonable way that a user of chronos can possibly _reason_ about exceptions coming out of there: the event loop will be in an indeterminite state when the loop is executing an _random_ callback. As expected, there are several issues in the error handling of chronos: in particular, it will end up in an inconsistent internal state whenever the selector loop operations fail, because the internal state update functions are not written in an exception-safe way. This PR turns this into a Defect, which probably is not the optimal way of handling things - expect more work to be done here. Some API have no way of reporting back errors to callers - for example, when something fails in the accept loop, there's not much it can do, and no way to report it back to the user of the API - this has been fixed with the new accept flow - the old one should be deprecated. Finally, there is information loss in the API: in composite operations like `poll` and `waitFor` there's no way to differentiate internal errors from user-level errors originating from callbacks. * store `CatchableError` in future * annotate proc's with correct raises information * `selectors2` to avoid non-CatchableError IOSelectorsException * `$` should never raise * remove unnecessary gcsafe annotations * fix exceptions leaking out of timer waits * fix some imports * functions must signal raising the union of all exceptions across all platforms to enable cross-platform code * switch to unittest2 * add `selectors2` which supercedes the std library version and fixes several exception handling issues in there * fixes * docs, platform-independent eh specifiers for some functions * add feature flag for strict exception mode also bump version to 3.0.0 - _most_ existing code should be compatible with this version of exception handling but some things might need fixing - callbacks, existing raises specifications etc. * fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
bufferSize = DefaultStreamBufferSize): StreamTransport {.
raises: [Defect, TransportOsError].} =
## Create new transport object using pipe's file descriptor.
##
## ``bufferSize`` is size of internal buffer for transport.
let res = fromPipe2(fd, child, bufferSize)
if res.isErr(): raiseTransportOsError(res.error())
res.get()