#twisted-dev

Tickets pending review: https://tm.tl/#9299, #9305, #9210, #9337, #9340, #8966 (the0id), #9100, #9118 (the0id), #4964 (jameshilliard), #9138, #9176, #7495

In looking at a solution to 9333, I'm wondering is there value in a public format method that appends tracebacks (note that formatEvent does not)?

Or should such a method remain private for internal use. A related question is how often the raw formatters are used instead of observers

Tickets pending review: https://tm.tl/#9299, #9305, #9210, #9337, #9340, #8966 (the0id), #9100, #9118 (the0id), #4964 (jameshilliard), #9138, #9176, #7495

Mark Williams <mrw@enotuniq.org>; closed <https://tm.tl/#7033>; - Use PyOpenSSL's ECDHE API

markrwilliams reviewed <https://tm.tl/#9210>; - TLS handshake failure with OpenSSL 1.1.0f (unassigned)

markrwilliams submitted <https://tm.tl/#9210>; - TLS handshake failure with OpenSSL 1.1.0f (unassigned) for review

Craig Rodrigues <rodrigc@FreeBSD.org>; closed <https://tm.tl/#9303>; - Port twisted.python._shellcomp to Python 3

new core enhancement https://tm.tl/#9343 by rodrigc: Add a toBytes() helper function

rodrigc submitted <https://tm.tl/#9343>; - Add a toBytes() helper function (unassigned) for review

Tickets pending review: https://tm.tl/#9299, #9305, #9337, #9340, #9210, #9343, #8966 (the0id), #9100, #9118 (the0id), #4964 (jameshilliard), #9138, #9176, #7495

Tickets pending review: https://tm.tl/#9299, #9305, #9337, #9340, #9210, #9343, #8966 (the0id), #9100, #9118 (the0id), #4964 (jameshilliard), #9138, #9176, #7495

why does Twisted need to find out if the read side of a ProcessWriter pipe has been closed?

it looks like it's done to determine when a process has exited, but SIGCHLD should handle that

it seems dubious to me that twisted actually supports non-windows platforms that don't support SIGCHLD

SIGCHLD doesn't tell you which process has exited (mostly).

Also ... You want to know if a process has closed its stdin so you can notify application code of that event.

Closed stdin is not the same event as process exited.

exarkun: yes, SIGCHLD doesn't tell you what process has exited, but the traditional approach of calling waitpid(-1, &wstatus, WNOHANG) in a loop addresses that, right?

exarkun: also it's certainly true that stdin (or any of a process' readable file descriptors) is meaningfully different than a process exit, but the tradition there is to discover that the read side is closed by handling EPIPE on the next write

was the idea with ProcessWriter that twisted could do better with pipes than it can with sockets?

runciter: waiting for EPIPE definitely sucks

I don't see how it's better for pipes than for sockets

It's the same

Get the event when it happens, not after an arbitrary delay?

You don't have to wait for a bum write() on a socket

(I think I don't understand what you mean)

exarkun: you're right, a socket will come up as readable, but you still have to be prepared to handle EPIPE

exarkun: reading ProcessWriter's source and The Linux Programming Interface's table on pipe behaviors, though, the write side of a pipe does not necessarily become readable when its read side is closed

Indeed. Isn't that why there's a function for detecting that behavior?

The bottom line is that child process management is a charlie foxtrot on Linux and you probably can't do better than a collection of hacks

We should try to keep the hacks in the reactor and out of application code

bbl

exarkun: yes, there is a function that detects that. i am wondering if we can get rid of it, and in doing so eliminate the need for a new hack to handle https://twistedmatrix.com/trac/ticket/9323

exarkun: thanks for the input

i don't think we can given that everything in the last ~12 years expects childConnectionLost to get called as soon as a pipe is closed

Tickets pending review: https://tm.tl/#9299, #9305, #9337, #9340, #9210, #9343, #8966 (the0id), #9100, #9118 (the0id), #4964 (jameshilliard), #9138, #9176, #7495

runciter: maybe relevant, https://github.com/twisted/twisted/commit/94d8f...

foom: oh, thanks!

I hated this code a decade ago. Good luck. =)

foom: unix never forgets

Well, this is really self-inflicted wound in twisted, with polling for write before calling processEnded

Polling on the write pipe, that is.