Re: pg_regress: lookup shellprog in $PATH

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Thu, Aug 25, 2022 at 10:48 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> If we were executing a program that the user needs to have some control
>> over, sure, but what we have here is an implementation detail that I
>> doubt anyone cares about. The fact that we're using a shell at all is
>> only because nobody has cared to manually implement I/O redirection logic
>> in these places; otherwise we'd be execl()'ing the server or psql directly.
>> Maybe the best answer would be to do that, and get out of the business
>> of knowing where the shell is?

> Well that also would not be crazy.

I experimented with this, and it seems like it might not be as awful as
we've always assumed it would be. Attached is an incomplete POC that
converts pg_regress proper to doing things this way. (isolationtester
and pg_regress_ecpg are outright broken by this patch, because they rely
on pg_regress' spawn_process and I didn't fix them yet. But you can run
the core regression tests to see it works.)

The Windows side of this is completely untested and may be broken; also,
perhaps Windows has something more nearly equivalent to execvp() that we
could use instead of reconstructing a command line? It's annoying that
the patch removes all shell-quoting hazards on the Unix side but they
are still there on the Windows side.

regards, tom lane