2018-05-16 08:22:34 +00:00
|
|
|
#
|
2019-02-06 14:49:11 +00:00
|
|
|
# Chronos
|
2018-05-16 08:22:34 +00:00
|
|
|
#
|
2021-05-06 07:49:55 +00:00
|
|
|
# (c) Copyright 2015 Dominik Picheta
|
|
|
|
# (c) Copyright 2018-2021 Status Research & Development GmbH
|
2018-05-16 08:22:34 +00:00
|
|
|
#
|
|
|
|
# Licensed under either of
|
|
|
|
# Apache License, version 2.0, (LICENSE-APACHEv2)
|
|
|
|
# MIT license (LICENSE-MIT)
|
|
|
|
|
2021-05-07 20:52:24 +00:00
|
|
|
import std/[os, tables, strutils, heapqueue, options, deques, sequtils]
|
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 ./srcloc
|
2019-04-08 00:59:49 +00:00
|
|
|
export srcloc
|
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
when defined(nimHasStacktracesModule):
|
|
|
|
import system/stacktraces
|
|
|
|
else:
|
|
|
|
const
|
|
|
|
reraisedFromBegin = -10
|
|
|
|
reraisedFromEnd = -100
|
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
const
|
2020-07-06 06:33:13 +00:00
|
|
|
LocCreateIndex* = 0
|
|
|
|
LocCompleteIndex* = 1
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
type StackTrace = string
|
2018-05-29 18:04:11 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
type
|
2019-06-20 20:30:41 +00:00
|
|
|
FutureState* {.pure.} = enum
|
|
|
|
Pending, Finished, Cancelled, Failed
|
|
|
|
|
2018-05-16 08:22:34 +00:00
|
|
|
FutureBase* = ref object of RootObj ## Untyped future.
|
2020-07-03 12:03:59 +00:00
|
|
|
location*: array[2, ptr SrcLoc]
|
2020-09-15 07:55:43 +00:00
|
|
|
callbacks: seq[AsyncCallback]
|
2019-06-20 20:30:41 +00:00
|
|
|
cancelcb*: CallbackFunc
|
|
|
|
child*: FutureBase
|
|
|
|
state*: FutureState
|
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 ## Stored exception
|
2020-07-03 12:03:59 +00:00
|
|
|
mustCancel*: bool
|
|
|
|
id*: int
|
2020-07-08 16:48:01 +00:00
|
|
|
|
|
|
|
when defined(chronosStackTrace):
|
|
|
|
errorStackTrace*: StackTrace
|
|
|
|
stackTrace: StackTrace ## For debugging purposes only.
|
|
|
|
|
|
|
|
when defined(chronosFutureTracking):
|
|
|
|
next*: FutureBase
|
|
|
|
prev*: FutureBase
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2018-05-29 18:04:11 +00:00
|
|
|
# ZAH: we have discussed some possible optimizations where
|
|
|
|
# the future can be stored within the caller's stack frame.
|
|
|
|
# How much refactoring is needed to make this a regular non-ref type?
|
|
|
|
# Obviously, it will still be allocated on the heap when necessary.
|
2018-05-16 08:22:34 +00:00
|
|
|
Future*[T] = ref object of FutureBase ## Typed future.
|
|
|
|
value: T ## Stored value
|
|
|
|
|
2019-03-30 22:31:10 +00:00
|
|
|
FutureStr*[T] = ref object of Future[T]
|
|
|
|
## Future to hold GC strings
|
|
|
|
gcholder*: string
|
|
|
|
|
|
|
|
FutureSeq*[A, B] = ref object of Future[A]
|
|
|
|
## Future to hold GC seqs
|
|
|
|
gcholder*: seq[B]
|
|
|
|
|
2018-05-16 08:22:34 +00:00
|
|
|
FutureVar*[T] = distinct Future[T]
|
|
|
|
|
2020-02-17 22:34:57 +00:00
|
|
|
FutureDefect* = object of Defect
|
2018-05-16 08:22:34 +00:00
|
|
|
cause*: FutureBase
|
|
|
|
|
2019-12-02 10:40:41 +00:00
|
|
|
FutureError* = object of CatchableError
|
|
|
|
|
2020-08-15 22:43:44 +00:00
|
|
|
CancelledError* = object of FutureError
|
|
|
|
|
2020-07-06 06:33:13 +00:00
|
|
|
FutureList* = object
|
|
|
|
head*: FutureBase
|
|
|
|
tail*: FutureBase
|
|
|
|
count*: int
|
|
|
|
|
2019-03-29 09:53:24 +00:00
|
|
|
var currentID* {.threadvar.}: int
|
|
|
|
currentID = 0
|
2020-07-08 16:48:01 +00:00
|
|
|
|
|
|
|
when defined(chronosFutureTracking):
|
|
|
|
var futureList* {.threadvar.}: FutureList
|
|
|
|
futureList = FutureList()
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 13:46:22 +00:00
|
|
|
template setupFutureBase(loc: ptr SrcLoc) =
|
2018-05-16 08:22:34 +00:00
|
|
|
new(result)
|
2019-06-20 20:30:41 +00:00
|
|
|
result.state = FutureState.Pending
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
result.stackTrace = getStackTrace()
|
2019-03-29 09:53:24 +00:00
|
|
|
result.id = currentID
|
2019-04-08 00:59:49 +00:00
|
|
|
result.location[LocCreateIndex] = loc
|
2019-03-29 09:53:24 +00:00
|
|
|
currentID.inc()
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosFutureTracking):
|
2020-08-15 22:45:41 +00:00
|
|
|
result.next = nil
|
|
|
|
result.prev = futureList.tail
|
|
|
|
if not(isNil(futureList.tail)):
|
|
|
|
futureList.tail.next = result
|
|
|
|
futureList.tail = result
|
|
|
|
if isNil(futureList.head):
|
|
|
|
futureList.head = result
|
|
|
|
futureList.count.inc()
|
2020-07-06 06:33:13 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
proc newFutureImpl[T](loc: ptr SrcLoc): Future[T] =
|
2019-04-08 13:46:22 +00:00
|
|
|
setupFutureBase(loc)
|
2019-04-08 00:59:49 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
proc newFutureSeqImpl[A, B](loc: ptr SrcLoc): FutureSeq[A, B] =
|
2019-04-08 13:46:22 +00:00
|
|
|
setupFutureBase(loc)
|
2019-04-08 00:59:49 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
proc newFutureStrImpl[T](loc: ptr SrcLoc): FutureStr[T] =
|
2019-04-08 13:46:22 +00:00
|
|
|
setupFutureBase(loc)
|
2019-04-08 00:59:49 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
proc newFutureVarImpl[T](loc: ptr SrcLoc): FutureVar[T] =
|
|
|
|
FutureVar[T](newFutureImpl[T](loc))
|
2019-04-08 13:46:22 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
template newFuture*[T](fromProc: static[string] = ""): Future[T] =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Creates a new future.
|
|
|
|
##
|
|
|
|
## Specifying ``fromProc``, which is a string specifying the name of the proc
|
|
|
|
## that this future belongs to, is a good habit as it helps with debugging.
|
2020-11-15 18:46:42 +00:00
|
|
|
newFutureImpl[T](getSrcLocation(fromProc))
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
template newFutureSeq*[A, B](fromProc: static[string] = ""): FutureSeq[A, B] =
|
2019-04-08 00:59:49 +00:00
|
|
|
## Create a new future which can hold/preserve GC sequence until future will
|
|
|
|
## not be completed.
|
2018-05-16 08:22:34 +00:00
|
|
|
##
|
|
|
|
## Specifying ``fromProc``, which is a string specifying the name of the proc
|
|
|
|
## that this future belongs to, is a good habit as it helps with debugging.
|
2020-11-15 18:46:42 +00:00
|
|
|
newFutureSeqImpl[A, B](getSrcLocation(fromProc))
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
template newFutureStr*[T](fromProc: static[string] = ""): FutureStr[T] =
|
2019-03-30 22:31:10 +00:00
|
|
|
## Create a new future which can hold/preserve GC string until future will
|
|
|
|
## not be completed.
|
|
|
|
##
|
|
|
|
## Specifying ``fromProc``, which is a string specifying the name of the proc
|
|
|
|
## that this future belongs to, is a good habit as it helps with debugging.
|
2020-11-15 18:46:42 +00:00
|
|
|
newFutureStrImpl[T](getSrcLocation(fromProc))
|
2019-03-30 22:31:10 +00:00
|
|
|
|
2020-11-15 18:46:42 +00:00
|
|
|
template newFutureVar*[T](fromProc: static[string] = ""): FutureVar[T] =
|
2019-04-08 00:59:49 +00:00
|
|
|
## Create a new ``FutureVar``. This Future type is ideally suited for
|
|
|
|
## situations where you want to avoid unnecessary allocations of Futures.
|
2019-03-30 22:31:10 +00:00
|
|
|
##
|
|
|
|
## Specifying ``fromProc``, which is a string specifying the name of the proc
|
|
|
|
## that this future belongs to, is a good habit as it helps with debugging.
|
2020-11-15 18:46:42 +00:00
|
|
|
newFutureVarImpl[T](getSrcLocation(fromProc))
|
2019-03-30 22:31:10 +00:00
|
|
|
|
2018-05-16 08:22:34 +00:00
|
|
|
proc clean*[T](future: FutureVar[T]) =
|
|
|
|
## Resets the ``finished`` status of ``future``.
|
2019-06-20 20:30:41 +00:00
|
|
|
Future[T](future).state = FutureState.Pending
|
2020-09-15 07:55:43 +00:00
|
|
|
Future[T](future).value = default(T)
|
2018-05-16 08:22:34 +00:00
|
|
|
Future[T](future).error = nil
|
|
|
|
|
2019-06-20 20:30:41 +00:00
|
|
|
proc finished*(future: FutureBase | FutureVar): bool {.inline.} =
|
2020-11-13 12:22:58 +00:00
|
|
|
## Determines whether ``future`` has completed, i.e. ``future`` state changed
|
|
|
|
## from state ``Pending`` to one of the states (``Finished``, ``Cancelled``,
|
|
|
|
## ``Failed``).
|
2019-06-20 20:30:41 +00:00
|
|
|
when future is FutureVar:
|
|
|
|
result = (FutureBase(future).state != FutureState.Pending)
|
|
|
|
else:
|
|
|
|
result = (future.state != FutureState.Pending)
|
|
|
|
|
|
|
|
proc cancelled*(future: FutureBase): bool {.inline.} =
|
|
|
|
## Determines whether ``future`` has cancelled.
|
2020-11-13 12:22:58 +00:00
|
|
|
(future.state == FutureState.Cancelled)
|
2019-06-20 20:30:41 +00:00
|
|
|
|
|
|
|
proc failed*(future: FutureBase): bool {.inline.} =
|
|
|
|
## Determines whether ``future`` completed with an error.
|
2020-11-13 12:22:58 +00:00
|
|
|
(future.state == FutureState.Failed)
|
|
|
|
|
|
|
|
proc completed*(future: FutureBase): bool {.inline.} =
|
|
|
|
## Determines whether ``future`` completed without an error.
|
|
|
|
(future.state == FutureState.Finished)
|
|
|
|
|
|
|
|
proc done*(future: FutureBase): bool {.inline.} =
|
|
|
|
## This is an alias for ``completed(future)`` procedure.
|
|
|
|
completed(future)
|
2019-06-20 20:30:41 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosFutureTracking):
|
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 futureDestructor(udata: pointer) =
|
2020-07-08 16:48:01 +00:00
|
|
|
## This procedure will be called when Future[T] got finished, cancelled or
|
|
|
|
## failed and all Future[T].callbacks are already scheduled and processed.
|
2020-08-15 22:45:41 +00:00
|
|
|
let future = cast[FutureBase](udata)
|
|
|
|
if future == futureList.tail: futureList.tail = future.prev
|
|
|
|
if future == futureList.head: futureList.head = future.next
|
|
|
|
if not(isNil(future.next)): future.next.prev = future.prev
|
|
|
|
if not(isNil(future.prev)): future.prev.next = future.next
|
|
|
|
futureList.count.dec()
|
2020-07-08 16:48:01 +00:00
|
|
|
|
|
|
|
proc scheduleDestructor(future: FutureBase) {.inline.} =
|
|
|
|
callSoon(futureDestructor, cast[pointer](future))
|
2020-07-06 06:33:13 +00:00
|
|
|
|
2020-07-03 12:03:59 +00:00
|
|
|
proc checkFinished(future: FutureBase, loc: ptr SrcLoc) =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Checks whether `future` is finished. If it is then raises a
|
2019-12-02 10:40:41 +00:00
|
|
|
## ``FutureDefect``.
|
2019-06-20 20:30:41 +00:00
|
|
|
if future.finished():
|
2019-03-29 09:53:24 +00:00
|
|
|
var msg = ""
|
|
|
|
msg.add("An attempt was made to complete a Future more than once. ")
|
|
|
|
msg.add("Details:")
|
|
|
|
msg.add("\n Future ID: " & $future.id)
|
2019-04-08 00:59:49 +00:00
|
|
|
msg.add("\n Creation location:")
|
|
|
|
msg.add("\n " & $future.location[LocCreateIndex])
|
|
|
|
msg.add("\n First completion location:")
|
|
|
|
msg.add("\n " & $future.location[LocCompleteIndex])
|
|
|
|
msg.add("\n Second completion location:")
|
|
|
|
msg.add("\n " & $loc)
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
msg.add("\n Stack trace to moment of creation:")
|
|
|
|
msg.add("\n" & indent(future.stackTrace.strip(), 4))
|
|
|
|
msg.add("\n Stack trace to moment of secondary completion:")
|
|
|
|
msg.add("\n" & indent(getStackTrace().strip(), 4))
|
2019-04-08 00:59:49 +00:00
|
|
|
msg.add("\n\n")
|
2019-12-02 10:40:41 +00:00
|
|
|
var err = newException(FutureDefect, msg)
|
2019-03-29 09:53:24 +00:00
|
|
|
err.cause = future
|
|
|
|
raise err
|
2019-04-08 00:59:49 +00:00
|
|
|
else:
|
|
|
|
future.location[LocCompleteIndex] = loc
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-09-15 07:55:43 +00:00
|
|
|
proc finish(fut: FutureBase, state: FutureState) =
|
|
|
|
# We do not perform any checks here, because:
|
|
|
|
# 1. `finish()` is a private procedure and `state` is under our control.
|
|
|
|
# 2. `fut.state` is checked by `checkFinished()`.
|
|
|
|
fut.state = state
|
|
|
|
fut.cancelcb = nil # release cancellation callback memory
|
|
|
|
for item in fut.callbacks.mitems():
|
|
|
|
if not(isNil(item.function)):
|
2021-01-11 21:15:21 +00:00
|
|
|
callSoon(item)
|
2020-09-15 07:55:43 +00:00
|
|
|
item = default(AsyncCallback) # release memory as early as possible
|
|
|
|
fut.callbacks = default(seq[AsyncCallback]) # release seq as well
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-09-15 07:55:43 +00:00
|
|
|
when defined(chronosFutureTracking):
|
|
|
|
scheduleDestructor(fut)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
proc complete[T](future: Future[T], val: T, loc: ptr SrcLoc) =
|
2019-06-25 07:18:47 +00:00
|
|
|
if not(future.cancelled()):
|
2020-07-03 12:03:59 +00:00
|
|
|
checkFinished(FutureBase(future), loc)
|
2019-06-25 07:18:47 +00:00
|
|
|
doAssert(isNil(future.error))
|
|
|
|
future.value = val
|
2020-09-15 07:55:43 +00:00
|
|
|
future.finish(FutureState.Finished)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
template complete*[T](future: Future[T], val: T) =
|
|
|
|
## Completes ``future`` with value ``val``.
|
|
|
|
complete(future, val, getSrcLocation())
|
|
|
|
|
|
|
|
proc complete(future: Future[void], loc: ptr SrcLoc) =
|
2019-06-25 07:18:47 +00:00
|
|
|
if not(future.cancelled()):
|
2020-07-03 12:03:59 +00:00
|
|
|
checkFinished(FutureBase(future), loc)
|
2019-06-25 07:18:47 +00:00
|
|
|
doAssert(isNil(future.error))
|
2020-09-15 07:55:43 +00:00
|
|
|
future.finish(FutureState.Finished)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
template complete*(future: Future[void]) =
|
2019-06-25 07:18:47 +00:00
|
|
|
## Completes a void ``future``.
|
2019-04-08 00:59:49 +00:00
|
|
|
complete(future, getSrcLocation())
|
|
|
|
|
|
|
|
proc complete[T](future: FutureVar[T], loc: ptr SrcLoc) =
|
2019-06-25 07:18:47 +00:00
|
|
|
if not(future.cancelled()):
|
|
|
|
template fut: untyped = Future[T](future)
|
2020-07-03 12:03:59 +00:00
|
|
|
checkFinished(FutureBase(fut), loc)
|
2019-06-25 07:18:47 +00:00
|
|
|
doAssert(isNil(fut.error))
|
2020-09-15 07:55:43 +00:00
|
|
|
fut.finish(FutureState.Finished)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
template complete*[T](futvar: FutureVar[T]) =
|
|
|
|
## Completes a ``FutureVar``.
|
|
|
|
complete(futvar, getSrcLocation())
|
|
|
|
|
|
|
|
proc complete[T](futvar: FutureVar[T], val: T, loc: ptr SrcLoc) =
|
2019-06-25 07:18:47 +00:00
|
|
|
if not(futvar.cancelled()):
|
|
|
|
template fut: untyped = Future[T](futvar)
|
2020-07-03 12:03:59 +00:00
|
|
|
checkFinished(FutureBase(fut), loc)
|
2019-06-25 07:18:47 +00:00
|
|
|
doAssert(isNil(fut.error))
|
|
|
|
fut.value = val
|
2020-09-15 07:55:43 +00:00
|
|
|
fut.finish(FutureState.Finished)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-04-08 00:59:49 +00:00
|
|
|
template complete*[T](futvar: FutureVar[T], val: T) =
|
|
|
|
## Completes a ``FutureVar`` with value ``val``.
|
|
|
|
##
|
|
|
|
## Any previously stored value will be overwritten.
|
|
|
|
complete(futvar, val, getSrcLocation())
|
|
|
|
|
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 fail[T](future: Future[T], error: ref CatchableError, loc: ptr SrcLoc) =
|
2019-06-25 07:18:47 +00:00
|
|
|
if not(future.cancelled()):
|
2020-07-03 12:03:59 +00:00
|
|
|
checkFinished(FutureBase(future), loc)
|
2019-06-25 07:18:47 +00:00
|
|
|
future.error = error
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
future.errorStackTrace = if getStackTrace(error) == "":
|
|
|
|
getStackTrace()
|
|
|
|
else:
|
|
|
|
getStackTrace(error)
|
2020-09-15 07:55:43 +00:00
|
|
|
future.finish(FutureState.Failed)
|
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
|
|
|
template fail*[T](future: Future[T], error: ref CatchableError) =
|
2019-04-08 00:59:49 +00:00
|
|
|
## Completes ``future`` with ``error``.
|
|
|
|
fail(future, error, getSrcLocation())
|
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
template newCancelledError(): ref CancelledError =
|
|
|
|
(ref CancelledError)(msg: "Future operation cancelled!")
|
|
|
|
|
2020-07-03 12:03:59 +00:00
|
|
|
proc cancelAndSchedule(future: FutureBase, loc: ptr SrcLoc) =
|
|
|
|
if not(future.finished()):
|
2019-06-20 20:30:41 +00:00
|
|
|
checkFinished(future, loc)
|
2020-07-08 16:48:01 +00:00
|
|
|
future.error = newCancelledError()
|
|
|
|
when defined(chronosStackTrace):
|
|
|
|
future.errorStackTrace = getStackTrace()
|
2020-09-15 07:55:43 +00:00
|
|
|
future.finish(FutureState.Cancelled)
|
2020-07-03 12:03:59 +00:00
|
|
|
|
|
|
|
template cancelAndSchedule*[T](future: Future[T]) =
|
|
|
|
cancelAndSchedule(FutureBase(future), getSrcLocation())
|
|
|
|
|
2020-11-13 12:22:58 +00:00
|
|
|
proc cancel(future: FutureBase, loc: ptr SrcLoc): bool =
|
|
|
|
## Request that Future ``future`` cancel itself.
|
|
|
|
##
|
|
|
|
## This arranges for a `CancelledError` to be thrown into procedure which
|
|
|
|
## waits for ``future`` on the next cycle through the event loop.
|
|
|
|
## The procedure then has a chance to clean up or even deny the request
|
|
|
|
## using `try/except/finally`.
|
|
|
|
##
|
|
|
|
## This call do not guarantee that the ``future`` will be cancelled: the
|
|
|
|
## exception might be caught and acted upon, delaying cancellation of the
|
|
|
|
## ``future`` or preventing cancellation completely. The ``future`` may also
|
|
|
|
## return value or raise different exception.
|
|
|
|
##
|
|
|
|
## Immediately after this procedure is called, ``future.cancelled()`` will
|
|
|
|
## not return ``true`` (unless the Future was already cancelled).
|
|
|
|
if future.finished():
|
|
|
|
return false
|
|
|
|
|
|
|
|
if not(isNil(future.child)):
|
|
|
|
if cancel(future.child, getSrcLocation()):
|
|
|
|
return true
|
|
|
|
else:
|
|
|
|
if not(isNil(future.cancelcb)):
|
|
|
|
future.cancelcb(cast[pointer](future))
|
|
|
|
future.cancelcb = nil
|
|
|
|
cancelAndSchedule(future, getSrcLocation())
|
|
|
|
|
|
|
|
future.mustCancel = true
|
|
|
|
return true
|
2019-06-20 20:30:41 +00:00
|
|
|
|
|
|
|
template cancel*[T](future: Future[T]) =
|
|
|
|
## Cancel ``future``.
|
2020-11-13 12:22:58 +00:00
|
|
|
discard cancel(FutureBase(future), getSrcLocation())
|
2019-06-20 20:30:41 +00:00
|
|
|
|
2018-05-16 08:22:34 +00:00
|
|
|
proc clearCallbacks(future: FutureBase) =
|
2020-09-15 07:55:43 +00:00
|
|
|
future.callbacks = default(seq[AsyncCallback])
|
2018-05-16 08:22:34 +00:00
|
|
|
|
|
|
|
proc addCallback*(future: FutureBase, cb: CallbackFunc, udata: pointer = nil) =
|
|
|
|
## Adds the callbacks proc to be called when the future completes.
|
|
|
|
##
|
|
|
|
## If future has already completed then ``cb`` will be called immediately.
|
2019-03-15 00:43:51 +00:00
|
|
|
doAssert(not isNil(cb))
|
2019-06-20 20:30:41 +00:00
|
|
|
if future.finished():
|
2018-05-16 08:22:34 +00:00
|
|
|
callSoon(cb, udata)
|
|
|
|
else:
|
|
|
|
let acb = AsyncCallback(function: cb, udata: udata)
|
|
|
|
future.callbacks.add acb
|
|
|
|
|
|
|
|
proc addCallback*[T](future: Future[T], cb: CallbackFunc) =
|
|
|
|
## Adds the callbacks proc to be called when the future completes.
|
|
|
|
##
|
|
|
|
## If future has already completed then ``cb`` will be called immediately.
|
2018-05-17 08:45:18 +00:00
|
|
|
future.addCallback(cb, cast[pointer](future))
|
2018-05-16 08:22:34 +00:00
|
|
|
|
|
|
|
proc removeCallback*(future: FutureBase, cb: CallbackFunc,
|
|
|
|
udata: pointer = nil) =
|
2020-09-15 07:55:43 +00:00
|
|
|
## Remove future from list of callbacks - this operation may be slow if there
|
|
|
|
## are many registered callbacks!
|
2019-03-15 00:43:51 +00:00
|
|
|
doAssert(not isNil(cb))
|
2020-09-15 07:55:43 +00:00
|
|
|
# Make sure to release memory associated with callback, or reference chains
|
|
|
|
# may be created!
|
|
|
|
future.callbacks.keepItIf:
|
|
|
|
it.function != cb or it.udata != udata
|
2018-05-16 08:22:34 +00:00
|
|
|
|
|
|
|
proc removeCallback*[T](future: Future[T], cb: CallbackFunc) =
|
2018-05-17 08:45:18 +00:00
|
|
|
future.removeCallback(cb, cast[pointer](future))
|
2018-05-16 08:22:34 +00:00
|
|
|
|
|
|
|
proc `callback=`*(future: FutureBase, cb: CallbackFunc, udata: pointer = nil) =
|
|
|
|
## Clears the list of callbacks and sets the callback proc to be called when
|
|
|
|
## the future completes.
|
|
|
|
##
|
|
|
|
## If future has already completed then ``cb`` will be called immediately.
|
|
|
|
##
|
|
|
|
## It's recommended to use ``addCallback`` or ``then`` instead.
|
2018-05-29 18:04:11 +00:00
|
|
|
# ZAH: how about `setLen(1); callbacks[0] = cb`
|
2018-05-16 08:22:34 +00:00
|
|
|
future.clearCallbacks
|
|
|
|
future.addCallback(cb, udata)
|
|
|
|
|
|
|
|
proc `callback=`*[T](future: Future[T], cb: CallbackFunc) =
|
|
|
|
## Sets the callback proc to be called when the future completes.
|
|
|
|
##
|
|
|
|
## If future has already completed then ``cb`` will be called immediately.
|
|
|
|
`callback=`(future, cb, cast[pointer](future))
|
|
|
|
|
2019-06-20 20:30:41 +00:00
|
|
|
proc `cancelCallback=`*[T](future: Future[T], cb: CallbackFunc) =
|
|
|
|
## Sets the callback procedure to be called when the future is cancelled.
|
|
|
|
##
|
|
|
|
## This callback will be called immediately as ``future.cancel()`` invoked.
|
|
|
|
future.cancelcb = cb
|
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
template getFilenameProcname(entry: StackTraceEntry): (string, string) =
|
|
|
|
when compiles(entry.filenameStr) and compiles(entry.procnameStr):
|
|
|
|
# We can't rely on "entry.filename" and "entry.procname" still being valid
|
|
|
|
# cstring pointers, because the "string.data" buffers they pointed to might
|
|
|
|
# be already garbage collected (this entry being a non-shallow copy,
|
|
|
|
# "entry.filename" no longer points to "entry.filenameStr.data", but to the
|
|
|
|
# buffer of the original object).
|
|
|
|
(entry.filenameStr, entry.procnameStr)
|
|
|
|
else:
|
|
|
|
($entry.filename, $entry.procname)
|
|
|
|
|
2018-05-16 08:22:34 +00:00
|
|
|
proc getHint(entry: StackTraceEntry): string =
|
|
|
|
## We try to provide some hints about stack trace entries that the user
|
|
|
|
## may not be familiar with, in particular calls inside the stdlib.
|
2021-05-06 07:49:55 +00:00
|
|
|
|
|
|
|
let (filename, procname) = getFilenameProcname(entry)
|
|
|
|
|
|
|
|
if procname == "processPendingCallbacks":
|
|
|
|
if cmpIgnoreStyle(filename, "asyncdispatch.nim") == 0:
|
2018-05-16 08:22:34 +00:00
|
|
|
return "Executes pending callbacks"
|
2021-05-06 07:49:55 +00:00
|
|
|
elif procname == "poll":
|
|
|
|
if cmpIgnoreStyle(filename, "asyncdispatch.nim") == 0:
|
2018-05-16 08:22:34 +00:00
|
|
|
return "Processes asynchronous completion events"
|
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
if procname.endsWith("_continue"):
|
|
|
|
if cmpIgnoreStyle(filename, "asyncmacro.nim") == 0:
|
2018-05-16 08:22:34 +00:00
|
|
|
return "Resumes an async procedure"
|
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
proc `$`(stackTraceEntries: seq[StackTraceEntry]): string =
|
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
|
|
|
try:
|
2021-05-06 07:49:55 +00:00
|
|
|
when defined(nimStackTraceOverride) and declared(addDebuggingInfo):
|
|
|
|
let entries = addDebuggingInfo(stackTraceEntries)
|
|
|
|
else:
|
|
|
|
let entries = stackTraceEntries
|
|
|
|
|
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
|
|
|
# Find longest filename & line number combo for alignment purposes.
|
|
|
|
var longestLeft = 0
|
|
|
|
for entry in entries:
|
2021-05-06 07:49:55 +00:00
|
|
|
let (filename, procname) = getFilenameProcname(entry)
|
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
* `$` should never raise
* remove unnecessary gcsafe annotations
* fix exceptions leaking out of timer waits
* fix some imports
* functions must signal raising the union of all exceptions across all
platforms to enable cross-platform code
* switch to unittest2
* add `selectors2` which supercedes the std library version and fixes
several exception handling issues in there
* fixes
* docs, platform-independent eh specifiers for some functions
* add feature flag for strict exception mode
also bump version to 3.0.0 - _most_ existing code should be compatible
with this version of exception handling but some things might need
fixing - callbacks, existing raises specifications etc.
* fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
if procname == "": continue
|
|
|
|
|
|
|
|
let leftLen = filename.len + len($entry.line)
|
|
|
|
if leftLen > longestLeft:
|
|
|
|
longestLeft = leftLen
|
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 indent = 2
|
|
|
|
# Format the entries.
|
|
|
|
for entry in entries:
|
2021-05-06 07:49:55 +00:00
|
|
|
let (filename, procname) = getFilenameProcname(entry)
|
|
|
|
|
|
|
|
if procname == "":
|
|
|
|
if entry.line == reraisedFromBegin:
|
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
|
|
|
result.add(spaces(indent) & "#[\n")
|
|
|
|
indent.inc(2)
|
2021-05-06 07:49:55 +00:00
|
|
|
elif entry.line == reraisedFromEnd:
|
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
* `$` should never raise
* remove unnecessary gcsafe annotations
* fix exceptions leaking out of timer waits
* fix some imports
* functions must signal raising the union of all exceptions across all
platforms to enable cross-platform code
* switch to unittest2
* add `selectors2` which supercedes the std library version and fixes
several exception handling issues in there
* fixes
* docs, platform-independent eh specifiers for some functions
* add feature flag for strict exception mode
also bump version to 3.0.0 - _most_ existing code should be compatible
with this version of exception handling but some things might need
fixing - callbacks, existing raises specifications etc.
* fix AsyncCheck for non-void T
2021-03-24 09:08:33 +00:00
|
|
|
indent.dec(2)
|
|
|
|
result.add(spaces(indent) & "]#\n")
|
|
|
|
continue
|
|
|
|
|
2021-05-06 07:49:55 +00:00
|
|
|
let left = "$#($#)" % [filename, $entry.line]
|
exception tracking (#166)
* exception tracking
This PR adds minimal exception tracking to chronos, moving the goalpost
one step further.
In particular, it becomes invalid to raise exceptions from `callSoon`
callbacks: this is critical for writing correct error handling because
there's no reasonable way that a user of chronos can possibly _reason_
about exceptions coming out of there: the event loop will be in an
indeterminite state when the loop is executing an _random_ callback.
As expected, there are several issues in the error handling of chronos:
in particular, it will end up in an inconsistent internal state whenever
the selector loop operations fail, because the internal state update
functions are not written in an exception-safe way. This PR turns this
into a Defect, which probably is not the optimal way of handling things
- expect more work to be done here.
Some API have no way of reporting back errors to callers - for example,
when something fails in the accept loop, there's not much it can do, and
no way to report it back to the user of the API - this has been fixed
with the new accept flow - the old one should be deprecated.
Finally, there is information loss in the API: in composite operations
like `poll` and `waitFor` there's no way to differentiate internal
errors from user-level errors originating from callbacks.
* store `CatchableError` in future
* annotate proc's with correct raises information
* `selectors2` to avoid non-CatchableError IOSelectorsException
* `$` 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
|
|
|
result.add((spaces(indent) & "$#$# $#\n") % [
|
|
|
|
left,
|
|
|
|
spaces(longestLeft - left.len + 2),
|
2021-05-06 07:49:55 +00:00
|
|
|
procname
|
exception tracking (#166)
* exception tracking
This PR adds minimal exception tracking to chronos, moving the goalpost
one step further.
In particular, it becomes invalid to raise exceptions from `callSoon`
callbacks: this is critical for writing correct error handling because
there's no reasonable way that a user of chronos can possibly _reason_
about exceptions coming out of there: the event loop will be in an
indeterminite state when the loop is executing an _random_ callback.
As expected, there are several issues in the error handling of chronos:
in particular, it will end up in an inconsistent internal state whenever
the selector loop operations fail, because the internal state update
functions are not written in an exception-safe way. This PR turns this
into a Defect, which probably is not the optimal way of handling things
- expect more work to be done here.
Some API have no way of reporting back errors to callers - for example,
when something fails in the accept loop, there's not much it can do, and
no way to report it back to the user of the API - this has been fixed
with the new accept flow - the old one should be deprecated.
Finally, there is information loss in the API: in composite operations
like `poll` and `waitFor` there's no way to differentiate internal
errors from user-level errors originating from callbacks.
* store `CatchableError` in future
* annotate proc's with correct raises information
* `selectors2` to avoid non-CatchableError IOSelectorsException
* `$` 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 hint = getHint(entry)
|
|
|
|
if hint.len > 0:
|
|
|
|
result.add(spaces(indent+2) & "## " & hint & "\n")
|
|
|
|
except ValueError as exc:
|
2021-05-07 20:52:24 +00:00
|
|
|
return exc.msg # Shouldn't actually happen since we set the formatting
|
|
|
|
# string
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
proc injectStacktrace(future: FutureBase) =
|
|
|
|
const header = "\nAsync traceback:\n"
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
var exceptionMsg = future.error.msg
|
|
|
|
if header in exceptionMsg:
|
|
|
|
# This is messy: extract the original exception message from the msg
|
|
|
|
# containing the async traceback.
|
|
|
|
let start = exceptionMsg.find(header)
|
|
|
|
exceptionMsg = exceptionMsg[0..<start]
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
var newMsg = exceptionMsg & header
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
let entries = getStackTraceEntries(future.error)
|
|
|
|
newMsg.add($entries)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
newMsg.add("Exception message: " & exceptionMsg & "\n")
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-07-08 16:48:01 +00:00
|
|
|
# # For debugging purposes
|
2021-05-06 07:49:55 +00:00
|
|
|
# newMsg.add("Exception type:")
|
2020-07-08 16:48:01 +00:00
|
|
|
# for entry in getStackTraceEntries(future.error):
|
|
|
|
# newMsg.add "\n" & $entry
|
|
|
|
future.error.msg = newMsg
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2021-05-07 20:52:24 +00:00
|
|
|
proc internalCheckComplete*(fut: FutureBase) {.
|
|
|
|
raises: [Defect, CatchableError].} =
|
2019-08-15 13:32:46 +00:00
|
|
|
# For internal use only. Used in asyncmacro
|
|
|
|
if not(isNil(fut.error)):
|
2020-07-08 16:48:01 +00:00
|
|
|
when defined(chronosStackTrace):
|
|
|
|
injectStacktrace(fut)
|
2019-08-15 13:32:46 +00:00
|
|
|
raise fut.error
|
|
|
|
|
|
|
|
proc internalRead*[T](fut: Future[T] | FutureVar[T]): T {.inline.} =
|
|
|
|
# For internal use only. Used in asyncmacro
|
|
|
|
when T isnot void:
|
|
|
|
return fut.value
|
|
|
|
|
2021-05-07 20:52:24 +00:00
|
|
|
proc read*[T](future: Future[T] | FutureVar[T]): T {.
|
|
|
|
raises: [Defect, CatchableError].} =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Retrieves the value of ``future``. Future must be finished otherwise
|
2019-06-25 07:50:56 +00:00
|
|
|
## this function will fail with a ``ValueError`` exception.
|
2018-05-16 08:22:34 +00:00
|
|
|
##
|
|
|
|
## If the result of the future is an error then that error will be raised.
|
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 future.finished():
|
2019-08-15 13:32:46 +00:00
|
|
|
internalCheckComplete(future)
|
|
|
|
internalRead(future)
|
2018-05-16 08:22:34 +00:00
|
|
|
else:
|
|
|
|
# TODO: Make a custom exception type for this?
|
2019-06-25 07:50:56 +00:00
|
|
|
raise newException(ValueError, "Future still in progress.")
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2021-05-07 20:52:24 +00:00
|
|
|
proc readError*[T](future: Future[T]): ref CatchableError {.
|
|
|
|
raises: [Defect, ValueError].} =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Retrieves the exception stored in ``future``.
|
|
|
|
##
|
2019-06-25 07:50:56 +00:00
|
|
|
## An ``ValueError`` exception will be thrown if no exception exists
|
2018-05-16 08:22:34 +00:00
|
|
|
## in the specified Future.
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(isNil(future.error)):
|
|
|
|
return future.error
|
2018-05-16 08:22:34 +00:00
|
|
|
else:
|
2019-06-25 07:50:56 +00:00
|
|
|
# TODO: Make a custom exception type for this?
|
|
|
|
raise newException(ValueError, "No error in future.")
|
2018-05-16 08:22:34 +00:00
|
|
|
|
|
|
|
proc mget*[T](future: FutureVar[T]): var T =
|
|
|
|
## Returns a mutable value stored in ``future``.
|
|
|
|
##
|
|
|
|
## Unlike ``read``, this function will not raise an exception if the
|
|
|
|
## Future has not been finished.
|
|
|
|
result = Future[T](future).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
|
|
|
template taskFutureLocation(future: FutureBase): string =
|
|
|
|
let loc = future.location[0]
|
|
|
|
"[" & (
|
|
|
|
if len(loc.procedure) == 0: "[unspecified]" else: $loc.procedure & "()"
|
|
|
|
) & " at " & $loc.file & ":" & $(loc.line) & "]"
|
|
|
|
|
|
|
|
template taskErrorMessage(future: FutureBase): string =
|
|
|
|
"Asynchronous task " & taskFutureLocation(future) &
|
2021-05-06 07:46:58 +00:00
|
|
|
" finished with an exception \"" & $future.error.name &
|
|
|
|
"\"!\nMessage: " & future.error.msg &
|
|
|
|
"\nStack trace: " & future.error.getStackTrace()
|
exception tracking (#166)
* exception tracking
This PR adds minimal exception tracking to chronos, moving the goalpost
one step further.
In particular, it becomes invalid to raise exceptions from `callSoon`
callbacks: this is critical for writing correct error handling because
there's no reasonable way that a user of chronos can possibly _reason_
about exceptions coming out of there: the event loop will be in an
indeterminite state when the loop is executing an _random_ callback.
As expected, there are several issues in the error handling of chronos:
in particular, it will end up in an inconsistent internal state whenever
the selector loop operations fail, because the internal state update
functions are not written in an exception-safe way. This PR turns this
into a Defect, which probably is not the optimal way of handling things
- expect more work to be done here.
Some API have no way of reporting back errors to callers - for example,
when something fails in the accept loop, there's not much it can do, and
no way to report it back to the user of the API - this has been fixed
with the new accept flow - the old one should be deprecated.
Finally, there is information loss in the API: in composite operations
like `poll` and `waitFor` there's no way to differentiate internal
errors from user-level errors originating from callbacks.
* store `CatchableError` in future
* annotate proc's with correct raises information
* `selectors2` to avoid non-CatchableError IOSelectorsException
* `$` 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
|
|
|
template taskCancelMessage(future: FutureBase): string =
|
|
|
|
"Asynchronous task " & taskFutureLocation(future) & " was cancelled!"
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2020-09-01 18:41:18 +00:00
|
|
|
proc asyncSpawn*(future: Future[void]) =
|
|
|
|
## Spawns a new concurrent async task.
|
|
|
|
##
|
|
|
|
## Tasks may not raise exceptions or be cancelled - a ``Defect`` will be
|
|
|
|
## raised when this happens.
|
|
|
|
##
|
|
|
|
## This should be used instead of ``discard`` and ``asyncCheck`` when calling
|
|
|
|
## an ``async`` procedure without ``await``, to ensure exceptions in the
|
|
|
|
## returned future are not silently discarded.
|
|
|
|
##
|
|
|
|
## Note, that if passed ``future`` is already finished, it will be checked
|
|
|
|
## and processed immediately.
|
|
|
|
doAssert(not isNil(future), "Future is nil")
|
|
|
|
|
|
|
|
proc cb(data: pointer) =
|
|
|
|
if future.failed():
|
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
|
|
|
raise newException(FutureDefect, taskErrorMessage(future))
|
2020-09-01 18:41:18 +00:00
|
|
|
elif future.cancelled():
|
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
|
|
|
raise newException(FutureDefect, taskCancelMessage(future))
|
2020-09-01 18:41:18 +00:00
|
|
|
|
|
|
|
if not(future.finished()):
|
|
|
|
# We adding completion callback only if ``future`` is not finished yet.
|
|
|
|
future.addCallback(cb)
|
|
|
|
else:
|
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
|
|
|
cb(nil)
|
|
|
|
|
|
|
|
proc asyncCheck*[T](future: Future[T]) {.
|
2021-05-07 20:52:24 +00:00
|
|
|
deprecated: "Raises Defect on future failure, fix your code and use" &
|
|
|
|
" asyncSpawn!".} =
|
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
|
|
|
## This function used to raise an exception through the `poll` call if
|
|
|
|
## the given future failed - there's no way to handle such exceptions so this
|
|
|
|
## function is now an alias for `asyncSpawn`
|
|
|
|
##
|
|
|
|
when T is void:
|
|
|
|
asyncSpawn(future)
|
|
|
|
else:
|
|
|
|
proc cb(data: pointer) =
|
|
|
|
if future.failed():
|
|
|
|
raise newException(FutureDefect, taskErrorMessage(future))
|
|
|
|
elif future.cancelled():
|
|
|
|
raise newException(FutureDefect, taskCancelMessage(future))
|
|
|
|
|
|
|
|
if not(future.finished()):
|
|
|
|
# We adding completion callback only if ``future`` is not finished yet.
|
|
|
|
future.addCallback(cb)
|
|
|
|
else:
|
|
|
|
cb(nil)
|
2020-09-01 18:41:18 +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 asyncDiscard*[T](future: Future[T]) {.
|
|
|
|
deprecated: "Use asyncSpawn or `discard await`".} = discard
|
|
|
|
## `asyncDiscard` will discard the outcome of the operation - unlike `discard`
|
|
|
|
## it also throws away exceptions! Use `asyncSpawn` if you're sure your
|
|
|
|
## code doesn't raise exceptions, or `discard await` to ignore successful
|
|
|
|
## outcomes
|
2019-03-15 00:43:51 +00:00
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
proc `and`*[T, Y](fut1: Future[T], fut2: Future[Y]): Future[void] {.
|
|
|
|
deprecated: "Use allFutures[T](varargs[Future[T]])".} =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Returns a future which will complete once both ``fut1`` and ``fut2``
|
|
|
|
## complete.
|
2019-06-20 20:30:41 +00:00
|
|
|
##
|
2019-07-04 12:04:59 +00:00
|
|
|
## If cancelled, ``fut1`` and ``fut2`` futures WILL NOT BE cancelled.
|
2019-06-04 16:51:35 +00:00
|
|
|
var retFuture = newFuture[void]("chronos.`and`")
|
2018-05-16 08:22:34 +00:00
|
|
|
proc cb(data: pointer) =
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
if fut1.finished() and fut2.finished():
|
2018-05-16 08:22:34 +00:00
|
|
|
if cast[pointer](fut1) == data:
|
2019-06-20 20:30:41 +00:00
|
|
|
if fut1.failed():
|
|
|
|
retFuture.fail(fut1.error)
|
|
|
|
else:
|
|
|
|
retFuture.complete()
|
2018-05-16 08:22:34 +00:00
|
|
|
else:
|
2019-06-20 20:30:41 +00:00
|
|
|
if fut2.failed():
|
|
|
|
retFuture.fail(fut2.error)
|
|
|
|
else:
|
|
|
|
retFuture.complete()
|
2018-05-16 08:22:34 +00:00
|
|
|
fut1.callback = cb
|
|
|
|
fut2.callback = cb
|
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 cancellation(udata: pointer) =
|
2019-07-04 12:04:59 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
2020-11-13 12:22:58 +00:00
|
|
|
if not(fut1.finished()):
|
|
|
|
fut1.removeCallback(cb)
|
|
|
|
if not(fut2.finished()):
|
|
|
|
fut2.removeCallback(cb)
|
2019-06-20 20:30:41 +00:00
|
|
|
|
2020-01-27 18:28:44 +00:00
|
|
|
retFuture.cancelCallback = cancellation
|
2018-05-16 08:22:34 +00:00
|
|
|
return retFuture
|
|
|
|
|
2020-04-21 04:07:49 +00:00
|
|
|
proc `or`*[T, Y](fut1: Future[T], fut2: Future[Y]): Future[void] =
|
2018-05-16 08:22:34 +00:00
|
|
|
## Returns a future which will complete once either ``fut1`` or ``fut2``
|
|
|
|
## complete.
|
2019-07-04 12:04:59 +00:00
|
|
|
##
|
2020-04-21 04:33:31 +00:00
|
|
|
## If ``fut1`` or ``fut2`` future is failed, the result future will also be
|
|
|
|
## failed with an error stored in ``fut1`` or ``fut2`` respectively.
|
|
|
|
##
|
|
|
|
## If both ``fut1`` and ``fut2`` future are completed or failed, the result
|
|
|
|
## future will depend on the state of ``fut1`` future. So if ``fut1`` future
|
|
|
|
## is failed, the result future will also be failed, if ``fut1`` future is
|
|
|
|
## completed, the result future will also be completed.
|
2020-04-21 04:07:49 +00:00
|
|
|
##
|
2019-07-04 12:04:59 +00:00
|
|
|
## If cancelled, ``fut1`` and ``fut2`` futures WILL NOT BE cancelled.
|
2020-04-21 04:07:49 +00:00
|
|
|
var retFuture = newFuture[void]("chronos.or")
|
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 cb: proc(udata: pointer) {.gcsafe, raises: [Defect].}
|
2021-04-03 08:05:08 +00:00
|
|
|
cb = proc(udata: pointer) {.gcsafe, raises: [Defect].} =
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
var fut = cast[FutureBase](udata)
|
|
|
|
if cast[pointer](fut1) == udata:
|
2018-05-16 08:22:34 +00:00
|
|
|
fut2.removeCallback(cb)
|
|
|
|
else:
|
|
|
|
fut1.removeCallback(cb)
|
2020-04-21 04:07:49 +00:00
|
|
|
if fut.failed():
|
|
|
|
retFuture.fail(fut.error)
|
|
|
|
else:
|
|
|
|
retFuture.complete()
|
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 cancellation(udata: pointer) =
|
2019-07-04 12:04:59 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
2020-11-13 12:22:58 +00:00
|
|
|
if not(fut1.finished()):
|
|
|
|
fut1.removeCallback(cb)
|
|
|
|
if not(fut2.finished()):
|
|
|
|
fut2.removeCallback(cb)
|
2019-06-20 20:30:41 +00:00
|
|
|
|
2020-04-21 04:07:49 +00:00
|
|
|
if fut1.finished():
|
|
|
|
if fut1.failed():
|
|
|
|
retFuture.fail(fut1.error)
|
|
|
|
else:
|
|
|
|
retFuture.complete()
|
|
|
|
return retFuture
|
|
|
|
|
|
|
|
if fut2.finished():
|
|
|
|
if fut2.failed():
|
|
|
|
retFuture.fail(fut2.error)
|
|
|
|
else:
|
|
|
|
retFuture.complete()
|
|
|
|
return retFuture
|
|
|
|
|
|
|
|
fut1.addCallback(cb)
|
|
|
|
fut2.addCallback(cb)
|
|
|
|
|
2020-01-27 18:28:44 +00:00
|
|
|
retFuture.cancelCallback = cancellation
|
2018-05-16 08:22:34 +00:00
|
|
|
return retFuture
|
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
proc all*[T](futs: varargs[Future[T]]): auto {.
|
|
|
|
deprecated: "Use allFutures(varargs[Future[T]])".} =
|
2019-03-15 00:43:51 +00:00
|
|
|
## Returns a future which will complete once all futures in ``futs`` complete.
|
2018-05-16 08:22:34 +00:00
|
|
|
## If the argument is empty, the returned future completes immediately.
|
|
|
|
##
|
|
|
|
## If the awaited futures are not ``Future[void]``, the returned future
|
|
|
|
## will hold the values of all awaited futures in a sequence.
|
|
|
|
##
|
2019-03-15 00:43:51 +00:00
|
|
|
## If the awaited futures *are* ``Future[void]``, this proc returns
|
|
|
|
## ``Future[void]``.
|
|
|
|
##
|
|
|
|
## Note, that if one of the futures in ``futs`` will fail, result of ``all()``
|
|
|
|
## will also be failed with error from failed future.
|
2019-06-20 20:30:41 +00:00
|
|
|
##
|
|
|
|
## TODO: This procedure has bug on handling cancelled futures from ``futs``.
|
|
|
|
## So if future from ``futs`` list become cancelled, what must be returned?
|
|
|
|
## You can't cancel result ``retFuture`` because in such way infinite
|
|
|
|
## recursion will happen.
|
2019-03-15 00:43:51 +00:00
|
|
|
let totalFutures = len(futs)
|
|
|
|
var completedFutures = 0
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-03-15 00:43:51 +00:00
|
|
|
# Because we can't capture varargs[T] in closures we need to create copy.
|
|
|
|
var nfuts = @futs
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-03-15 00:43:51 +00:00
|
|
|
when T is void:
|
2019-06-04 16:51:35 +00:00
|
|
|
var retFuture = newFuture[void]("chronos.all(void)")
|
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 cb(udata: pointer) =
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
2018-05-16 08:22:34 +00:00
|
|
|
inc(completedFutures)
|
2019-06-20 20:30:41 +00:00
|
|
|
if completedFutures == totalFutures:
|
|
|
|
for nfut in nfuts:
|
|
|
|
if nfut.failed():
|
|
|
|
retFuture.fail(nfut.error)
|
|
|
|
break
|
|
|
|
if not(retFuture.failed()):
|
|
|
|
retFuture.complete()
|
|
|
|
|
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb)
|
2018-05-16 08:22:34 +00:00
|
|
|
|
2019-03-15 00:43:51 +00:00
|
|
|
if len(nfuts) == 0:
|
2018-05-16 08:22:34 +00:00
|
|
|
retFuture.complete()
|
|
|
|
|
|
|
|
return retFuture
|
|
|
|
else:
|
2019-06-04 16:51:35 +00:00
|
|
|
var retFuture = newFuture[seq[T]]("chronos.all(T)")
|
2019-03-15 00:43:51 +00:00
|
|
|
var retValues = newSeq[T](totalFutures)
|
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 cb(udata: pointer) =
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
2019-03-15 00:43:51 +00:00
|
|
|
inc(completedFutures)
|
2019-06-20 20:30:41 +00:00
|
|
|
if completedFutures == totalFutures:
|
|
|
|
for k, nfut in nfuts:
|
|
|
|
if nfut.failed():
|
|
|
|
retFuture.fail(nfut.error)
|
|
|
|
break
|
|
|
|
else:
|
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
|
|
|
retValues[k] = nfut.value
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.failed()):
|
|
|
|
retFuture.complete(retValues)
|
|
|
|
|
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb)
|
2019-03-15 00:43:51 +00:00
|
|
|
|
|
|
|
if len(nfuts) == 0:
|
2018-05-16 08:22:34 +00:00
|
|
|
retFuture.complete(retValues)
|
|
|
|
|
|
|
|
return retFuture
|
2019-06-04 16:51:35 +00:00
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
proc oneIndex*[T](futs: varargs[Future[T]]): Future[int] {.
|
|
|
|
deprecated: "Use one[T](varargs[Future[T]])".} =
|
2019-06-04 16:51:35 +00:00
|
|
|
## Returns a future which will complete once one of the futures in ``futs``
|
|
|
|
## complete.
|
|
|
|
##
|
|
|
|
## If the argument is empty, the returned future FAILS immediately.
|
|
|
|
##
|
|
|
|
## Returned future will hold index of completed/failed future in ``futs``
|
|
|
|
## argument.
|
|
|
|
var nfuts = @futs
|
|
|
|
var retFuture = newFuture[int]("chronos.oneIndex(T)")
|
|
|
|
|
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 cb(udata: pointer) =
|
2019-06-04 16:51:35 +00:00
|
|
|
var res = -1
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
var rfut = cast[FutureBase](udata)
|
2019-06-04 16:51:35 +00:00
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if cast[FutureBase](nfuts[i]) != rfut:
|
|
|
|
nfuts[i].removeCallback(cb)
|
|
|
|
else:
|
|
|
|
res = i
|
|
|
|
retFuture.complete(res)
|
|
|
|
|
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb)
|
|
|
|
|
|
|
|
if len(nfuts) == 0:
|
2019-06-25 07:50:56 +00:00
|
|
|
retFuture.fail(newException(ValueError, "Empty Future[T] list"))
|
2019-06-04 16:51:35 +00:00
|
|
|
|
|
|
|
return retFuture
|
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
proc oneValue*[T](futs: varargs[Future[T]]): Future[T] {.
|
|
|
|
deprecated: "Use one[T](varargs[Future[T]])".} =
|
2019-06-04 16:51:35 +00:00
|
|
|
## Returns a future which will complete once one of the futures in ``futs``
|
|
|
|
## complete.
|
|
|
|
##
|
|
|
|
## If the argument is empty, returned future FAILS immediately.
|
|
|
|
##
|
|
|
|
## Returned future will hold value of completed ``futs`` future, or error
|
|
|
|
## if future was failed.
|
|
|
|
var nfuts = @futs
|
|
|
|
var retFuture = newFuture[T]("chronos.oneValue(T)")
|
|
|
|
|
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 cb(udata: pointer) =
|
2019-06-04 16:51:35 +00:00
|
|
|
var resFut: Future[T]
|
2019-06-20 20:30:41 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
var rfut = cast[FutureBase](udata)
|
2019-06-04 16:51:35 +00:00
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if cast[FutureBase](nfuts[i]) != rfut:
|
|
|
|
nfuts[i].removeCallback(cb)
|
|
|
|
else:
|
|
|
|
resFut = nfuts[i]
|
2019-06-20 20:30:41 +00:00
|
|
|
if resFut.failed():
|
2019-06-04 16:51:35 +00:00
|
|
|
retFuture.fail(resFut.error)
|
|
|
|
else:
|
|
|
|
when T is void:
|
|
|
|
retFuture.complete()
|
|
|
|
else:
|
|
|
|
retFuture.complete(resFut.read())
|
|
|
|
|
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb)
|
|
|
|
|
|
|
|
if len(nfuts) == 0:
|
2019-06-25 07:50:56 +00:00
|
|
|
retFuture.fail(newException(ValueError, "Empty Future[T] list"))
|
2019-06-20 20:30:41 +00:00
|
|
|
|
|
|
|
return retFuture
|
|
|
|
|
2020-07-30 00:47:11 +00:00
|
|
|
proc cancelAndWait*[T](fut: Future[T]): Future[void] =
|
2020-11-13 12:22:58 +00:00
|
|
|
## Initiate cancellation process for Future ``fut`` and wait until ``fut`` is
|
|
|
|
## done e.g. changes its state (become completed, failed or cancelled).
|
|
|
|
##
|
|
|
|
## If ``fut`` is already finished (completed, failed or cancelled) result
|
|
|
|
## Future[void] object will be returned complete.
|
2020-07-30 00:47:11 +00:00
|
|
|
var retFuture = newFuture[void]("chronos.cancelAndWait(T)")
|
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()):
|
|
|
|
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 cancellation(udata: pointer) =
|
2020-11-13 12:22:58 +00:00
|
|
|
if not(fut.finished()):
|
|
|
|
fut.removeCallback(continuation)
|
|
|
|
if fut.finished():
|
|
|
|
retFuture.complete()
|
|
|
|
else:
|
|
|
|
fut.addCallback(continuation)
|
2020-11-17 10:04:19 +00:00
|
|
|
retFuture.cancelCallback = cancellation
|
2020-11-13 12:22:58 +00:00
|
|
|
# Initiate cancellation process.
|
|
|
|
fut.cancel()
|
2019-06-04 16:51:35 +00:00
|
|
|
return retFuture
|
2019-07-04 12:04:59 +00:00
|
|
|
|
|
|
|
proc allFutures*[T](futs: varargs[Future[T]]): Future[void] =
|
|
|
|
## Returns a future which will complete only when all futures in ``futs``
|
|
|
|
## will be completed, failed or canceled.
|
|
|
|
##
|
|
|
|
## If the argument is empty, the returned future COMPLETES immediately.
|
|
|
|
##
|
|
|
|
## On cancel all the awaited futures ``futs`` WILL NOT BE cancelled.
|
|
|
|
var retFuture = newFuture[void]("chronos.allFutures()")
|
|
|
|
let totalFutures = len(futs)
|
|
|
|
var completedFutures = 0
|
|
|
|
|
|
|
|
# Because we can't capture varargs[T] in closures we need to create copy.
|
|
|
|
var nfuts = @futs
|
|
|
|
|
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 cb(udata: pointer) =
|
2019-07-04 12:04:59 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
inc(completedFutures)
|
|
|
|
if completedFutures == totalFutures:
|
|
|
|
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 cancellation(udata: pointer) =
|
2019-07-04 12:04:59 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
2020-01-27 18:28:44 +00:00
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if not(nfuts[i].finished()):
|
|
|
|
nfuts[i].removeCallback(cb)
|
2019-07-04 12:04:59 +00:00
|
|
|
|
|
|
|
for fut in nfuts:
|
2020-04-06 10:56:24 +00:00
|
|
|
if not(fut.finished()):
|
|
|
|
fut.addCallback(cb)
|
|
|
|
else:
|
|
|
|
inc(completedFutures)
|
2019-07-04 12:04:59 +00:00
|
|
|
|
2020-01-27 18:28:44 +00:00
|
|
|
retFuture.cancelCallback = cancellation
|
2020-04-06 10:56:24 +00:00
|
|
|
if len(nfuts) == 0 or len(nfuts) == completedFutures:
|
2019-07-04 12:04:59 +00:00
|
|
|
retFuture.complete()
|
|
|
|
|
|
|
|
return retFuture
|
|
|
|
|
2020-04-01 09:10:56 +00:00
|
|
|
proc allFinished*[T](futs: varargs[Future[T]]): Future[seq[Future[T]]] =
|
|
|
|
## Returns a future which will complete only when all futures in ``futs``
|
|
|
|
## will be completed, failed or canceled.
|
|
|
|
##
|
|
|
|
## Returned sequence will hold all the Future[T] objects passed to
|
|
|
|
## ``allCompleted`` with the order preserved.
|
|
|
|
##
|
|
|
|
## If the argument is empty, the returned future COMPLETES immediately.
|
|
|
|
##
|
|
|
|
## On cancel all the awaited futures ``futs`` WILL NOT BE cancelled.
|
|
|
|
var retFuture = newFuture[seq[Future[T]]]("chronos.allFinished()")
|
|
|
|
let totalFutures = len(futs)
|
|
|
|
var completedFutures = 0
|
|
|
|
|
|
|
|
var nfuts = @futs
|
|
|
|
|
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 cb(udata: pointer) =
|
2020-04-01 09:10:56 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
inc(completedFutures)
|
|
|
|
if completedFutures == totalFutures:
|
|
|
|
retFuture.complete(nfuts)
|
|
|
|
|
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) =
|
2020-04-01 09:10:56 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
|
|
|
for fut in nfuts.mitems():
|
|
|
|
if not(fut.finished()):
|
|
|
|
fut.removeCallback(cb)
|
|
|
|
|
|
|
|
for fut in nfuts:
|
2020-04-06 10:56:24 +00:00
|
|
|
if not(fut.finished()):
|
|
|
|
fut.addCallback(cb)
|
|
|
|
else:
|
|
|
|
inc(completedFutures)
|
2020-04-01 09:10:56 +00:00
|
|
|
|
|
|
|
retFuture.cancelCallback = cancellation
|
2020-04-06 10:56:24 +00:00
|
|
|
if len(nfuts) == 0 or len(nfuts) == completedFutures:
|
2020-04-01 09:10:56 +00:00
|
|
|
retFuture.complete(nfuts)
|
|
|
|
|
|
|
|
return retFuture
|
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
proc one*[T](futs: varargs[Future[T]]): Future[Future[T]] =
|
|
|
|
## Returns a future which will complete and return completed Future[T] inside,
|
|
|
|
## when one of the futures in ``futs`` will be completed, failed or canceled.
|
|
|
|
##
|
|
|
|
## If the argument is empty, the returned future FAILS immediately.
|
|
|
|
##
|
2020-04-06 10:56:24 +00:00
|
|
|
## On success returned Future will hold finished Future[T].
|
2019-07-04 12:04:59 +00:00
|
|
|
##
|
|
|
|
## On cancel futures in ``futs`` WILL NOT BE cancelled.
|
|
|
|
var retFuture = newFuture[Future[T]]("chronos.one()")
|
|
|
|
|
|
|
|
# Because we can't capture varargs[T] in closures we need to create copy.
|
|
|
|
var nfuts = @futs
|
|
|
|
|
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 cb: proc(udata: pointer) {.gcsafe, raises: [Defect].}
|
2021-04-03 08:05:08 +00:00
|
|
|
cb = proc(udata: pointer) {.gcsafe, raises: [Defect].} =
|
2019-07-04 12:04:59 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
var res: Future[T]
|
|
|
|
var rfut = cast[FutureBase](udata)
|
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if cast[FutureBase](nfuts[i]) != rfut:
|
|
|
|
nfuts[i].removeCallback(cb)
|
|
|
|
else:
|
|
|
|
res = nfuts[i]
|
|
|
|
retFuture.complete(res)
|
|
|
|
|
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) =
|
2019-07-04 12:04:59 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
2020-01-27 18:28:44 +00:00
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if not(nfuts[i].finished()):
|
|
|
|
nfuts[i].removeCallback(cb)
|
2019-07-04 12:04:59 +00:00
|
|
|
|
2020-04-06 10:56:24 +00:00
|
|
|
# If one of the Future[T] already finished we return it as result
|
|
|
|
for fut in nfuts:
|
|
|
|
if fut.finished():
|
|
|
|
retFuture.complete(fut)
|
|
|
|
return retFuture
|
|
|
|
|
2019-07-04 12:04:59 +00:00
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb)
|
|
|
|
|
|
|
|
if len(nfuts) == 0:
|
|
|
|
retFuture.fail(newException(ValueError, "Empty Future[T] list"))
|
|
|
|
|
2020-01-27 18:28:44 +00:00
|
|
|
retFuture.cancelCallback = cancellation
|
2019-07-04 12:04:59 +00:00
|
|
|
return retFuture
|
2020-11-26 22:50:55 +00:00
|
|
|
|
|
|
|
proc race*(futs: varargs[FutureBase]): Future[FutureBase] =
|
|
|
|
## Returns a future which will complete and return completed FutureBase,
|
|
|
|
## when one of the futures in ``futs`` will be completed, failed or canceled.
|
|
|
|
##
|
|
|
|
## If the argument is empty, the returned future FAILS immediately.
|
|
|
|
##
|
|
|
|
## On success returned Future will hold finished FutureBase.
|
|
|
|
##
|
|
|
|
## On cancel futures in ``futs`` WILL NOT BE cancelled.
|
|
|
|
var retFuture = newFuture[FutureBase]("chronos.race()")
|
|
|
|
|
|
|
|
# Because we can't capture varargs[T] in closures we need to create copy.
|
|
|
|
var nfuts = @futs
|
|
|
|
|
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 cb: proc(udata: pointer) {.gcsafe, raises: [Defect].}
|
2021-04-03 08:05:08 +00:00
|
|
|
cb = proc(udata: pointer) {.gcsafe, raises: [Defect].} =
|
2020-11-26 22:50:55 +00:00
|
|
|
if not(retFuture.finished()):
|
|
|
|
var res: FutureBase
|
|
|
|
var rfut = cast[FutureBase](udata)
|
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if nfuts[i] != rfut:
|
|
|
|
nfuts[i].removeCallback(cb)
|
|
|
|
else:
|
|
|
|
res = nfuts[i]
|
|
|
|
retFuture.complete(res)
|
|
|
|
|
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) =
|
2020-11-26 22:50:55 +00:00
|
|
|
# On cancel we remove all our callbacks only.
|
|
|
|
for i in 0..<len(nfuts):
|
|
|
|
if not(nfuts[i].finished()):
|
|
|
|
nfuts[i].removeCallback(cb)
|
|
|
|
|
|
|
|
# If one of the Future[T] already finished we return it as result
|
|
|
|
for fut in nfuts:
|
|
|
|
if fut.finished():
|
|
|
|
retFuture.complete(fut)
|
|
|
|
return retFuture
|
|
|
|
|
|
|
|
for fut in nfuts:
|
|
|
|
fut.addCallback(cb, cast[pointer](fut))
|
|
|
|
|
|
|
|
if len(nfuts) == 0:
|
|
|
|
retFuture.fail(newException(ValueError, "Empty Future[T] list"))
|
|
|
|
|
|
|
|
retFuture.cancelCallback = cancellation
|
|
|
|
return retFuture
|