bug#33018: 26.1.50; thread starvation with async processes and accept-process-output

["Basil L. Contovounesios" <contovob@tcd.ie>]

[-- Attachment #1: Type: text/plain, Size: 775 bytes --]

Michael Albinus <michael.albinus@gmx.de> writes:

> "Basil L. Contovounesios" <contovob@tcd.ie> writes:
>
>> I attach a sample program test.el whose central function, test-slave,
>> invokes wget asynchronously before waiting for the process to exit.
>>
>> The issue I'm facing is that running test-slave twice in succession,
>> each time in a new thread, causes accept-process-output to hang with no
>> output (unless a timeout argument is given, in which case the function
>> returns nil) the second time around.
>
> If you want a process to communicate in a given thread, you must call
> `set-process-thread'. See the elisp manual.

Thanks, this is the first thing I tried when earlier experiments started
to hang.  I tried both the following redundant but explicit call:


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: test.diff --]
[-- Type: text/x-diff, Size: 616 bytes --]

diff -u --label /tmp/test.el --label \#\<buffer\ /tmp/test.el\> /tmp/test.el /tmp/buffer-content-hY0EDf
--- /tmp/test.el
+++ #<buffer /tmp/test.el>
@@ -24,6 +24,7 @@
                :command '("wget" "-qO-" "https://en.wikipedia.org/wiki/Emacs")
                :connection-type 'pipe
                :sentinel #'test-sentinel)))
+    (set-process-thread proc (current-thread))
     (while (eq (process-status proc) 'run)
       (test-debug proc 'accept-output (accept-process-output proc 5)))
     (test-debug proc 'exit (process-status proc) (process-exit-status proc))))

Diff finished.  Sun Oct 14 16:04:48 2018

[-- Attachment #3: Type: text/plain, Size: 295 bytes --]


as well as replacing (current-thread) with nil, to unlock the process.

Adding test-debug calls before and after accept-process-output revealed
nothing out of the ordinary, and explicitly un/locking the process
didn't fix the hang.  Should I be calling set-process-thread elsewhere?

-- 
Basil