Re: [PATCH 0/4] On Windows, limit which file handles are inherited by spawned child processes

From: Junio C Hamano <gitster@pobox.com>
To: "Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
Cc: git@vger.kernel.org, Johannes Schindelin <johannes.schindelin@gmx.de>
Subject: Re: [PATCH 0/4] On Windows, limit which file handles are inherited by spawned child processes
Date: Mon, 25 Nov 2019 14:42:02 +0900	[thread overview]
Message-ID: <xmqqimn8y0k5.fsf@gitster-ct.c.googlers.com> (raw)
In-Reply-To: <pull.670.git.git.1574433665.gitgitgadget@gmail.com> (Johannes Schindelin via GitGitGadget's message of "Fri, 22 Nov 2019 14:41:01 +0000")

"Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
writes:

> The problem to be solved: files cannot be deleted on Windows when even one
> process has an open file handle to it. So when a process opens a temporary
> file, then spawns a child process that inherits that file handle by mistake,
> and then the parent process tries to delete the temporary file while the
> child process is still running, the deletion will fail. (This description is
> slightly simplified, see the commit message "spawned processes need to
> inherit only standard handles" for more detail.)

Makes me wonder if we should be marking these fds with FD_CLOEXEC to
solve the issue in a way that is platform agnostic.  It may turn out
that we'd be better off to make it an explicit choice by the parent
when it leaves a FD open while spawning a child process (and by that
spawned child when it runs another executable---but I undertand that
it is a single-step operation, not a two-step one, on Windows).

In any case, synchronizing the differences in compat/ between our
trees is good.  Queued.