Re: What's cooking in git.git (Jul 2025, #02; Mon, 7)

From: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Date: Jul 9, 2025, 10:41 UTC
Message-ID: <vl7ahhcqgci3xemqhtdugdhar24ewl7mu4wqwxnc3jag5blpoo@l7b24mahadyi>
In-Reply-To: <xmqqplebzgm7.fsf@gitster.g>

On Mon, Jul 07, 2025 at 05:11:12PM -0800, Junio C Hamano wrote:

Show 12 quoted lines

> 
> * cb/daemon-reap-children (2025-06-26) 4 commits
>  - daemon: explicitly allow EINTR during poll()
>  - daemon: use sigaction() to install child_handler()
>  - compat/mingw: allow sigaction(SIGCHLD)
>  - compat/posix.h: track SA_RESTART fallback
> 
>  Futz with SIGCHLD handling in "git daemon".
> 
>  Stalled?
>  cf. <dba9ae0d-1e43-4345-a7ec-b57a07d45a07@gmail.com>
>  source: <pull.2002.v3.git.git.1750927988.gitgitgadget@gmail.com>

I don't think so, but definitely missing reviews (specially for patch 2) and slightly controversial.

I wasn't planning any further changes, and while I have some for the "related" patchset which enhances the notifications by using a self pipe was holding them (as well as the related feedback) to allow for this to mature on its own.

On the rationale on why this is independently useful, note that we are currently relying in at least 2 "unspecified" behaviours:

1) using signal() means that it is implementation defined if SA_RESTART
will be enabled or not for that signal, and
2) if SA_RESTART is enabled, it is implementation defined if we might get
interrupted in poll().

both are resolved by using sigaction() instead, and it is up to us to decide if SA_RESTART is enabled or not (which might make patch 4 obsolete if we decide against).

Carlo

CC: Johannes Sixt on feedback for SIGCHLD in mingw which uses 17 instead
    of 22 (cygwin and others) and therefore seem to cause signal to err
    without setting errno as it should.