Re: [PATCH] t5562: do not reuse output files

[Junio C Hamano <gitster@pobox.com>]

Max Kirillov <max@max630.net> writes:

> On Sat, Nov 24, 2018 at 04:47:19PM +0900, Junio C Hamano wrote:
>> I do agree that forcing the parent to wait, like you described in
>> the comment, would be far more preferrable,
>
> It looks like it can be done as simple as:
>
> --- a/http-backend.c
> +++ b/http-backend.c
> @@ -486,6 +486,8 @@ static void run_service(const char **argv, int buffer_input)
>         if (buffer_input || gzipped_request || req_len >= 0)
>                 cld.in = -1;
>         cld.git_cmd = 1;
> +       cld.clean_on_exit = 1;
> +       cld.wait_after_clean = 1;
>         if (start_command(&cld))
>                 exit(1);
>
> at least according to strate it does what it should.

Sounds sane.

I am offhand not sure what the right value of wait_after_clean for
this codepath be, though.  46df6906 ("execv_dashed_external: wait
for child on signal death", 2017-01-06) made this non-default but
turned it on for dashed externals (especially to help the case where
they spawn a pager), as the parent has nothing other than to wait
for the child to exit in that codepath.  Does the same reasoning
apply here, too?

This is a meta point, but I wonder if there is an easy way to "grep"
for uses of run-command interface that do *not* set clean_on_exit.
The pager that can outlive us long after we exit, kept alive while
the user views our output, is an example cited by afe19ff7
("run-command: optionally kill children on exit", 2012-01-07), and
while I am wondering if the default should hae been to kill the
children instead, such a "grep" would have been very useful to know
what codepaths would be affected if we flipped the default.