When `write` is called on a `StreamTransport`, the current sequence of
operations is:
* copy data to queue
* register for "write" event notification
* return unfinished future to `write` caller
* wait for "write" notification (in `poll`)
* perform one `send`
* wait for notification again if there's more data to write
* complete the future
In this PR, we introduce a fast path for writing:
* If the queue is empty, try to send as much data as possible
* If all data is sent, return completed future without `poll` round
* If there's more data to send than can be sent in one go, add the
rest to queue
* If the queue is not empty, enqueue as above
* When notified that write is possible, keep writing until OS buffer is
full before waiting for event again
The fast path provides significant performance benefits when there are
many small writes, such as when sending gossip to many peers, by
avoiding the poll loop and data copy on each send.
Also fixes an issue where the socket would not be removed from the
writer set if there were pending writes on close.
* 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
Eugene Kabanov
Jacek Sieka
Miran
Tanguy
Mamy Ratsimbazafy