Re: orangutan seizes up during isolation-check

On Sat, Oct 11, 2014 at 09:07:46AM -0400, Peter Eisentraut wrote:
> On 10/11/14 1:41 AM, Noah Misch wrote:
> > Good question. It would be nice to make the change there, for the benefit of
> > other consumers. The patch's setlocale_native_forked() assumes it never runs
> > in a multithreaded process, but libintl_setlocale() must not assume that. I
> > see a few ways libintl/gnulib might proceed:
>
> Yeah, it's difficult to see how they might proceed if they keep calling
> into Core Foundation, which might do anything, now or in the future.
>
> I went ahead and submitted a bug report to gettext:
> https://savannah.gnu.org/bugs/index.php?43404
>
> (They way I understand it is that the files concerned originate in
> gettext and are copied to gnulib.)
>
> Let's see what they say.

The gettext maintainer was open to implementing the setlocale_native_forked()
technique in gettext, though the last visible progress was in October. In any
event, PostgreSQL builds will see older gettext for several years. If
setlocale-darwin-fork-v1.patch is not wanted, I suggest making the postmaster
check during startup whether it has become multithreaded. If multithreaded:

FATAL: postmaster became multithreaded during startup
HINT: Set the LC_ALL environment variable to a valid locale.

I wondered whether to downgrade FATAL to LOG in back branches. Introducing a
new reason to block startup is disruptive for a minor release, but having the
postmaster deadlock at an unpredictable later time is even more disruptive. I
am inclined to halt startup that way in all branches.

> I like the idea of calling pthread_is_threaded_np() as a verification.
> This appears to be a OS X-specific function at the moment. If other
> platforms start adding it, then we'll run into the usual problems of how
> to link binaries that use pthread functions. Maybe that's not a
> realistic concern.

True. As written, "configure" will report the function unavailable if it
requires threading libraries. For a measure that's just a backstop against
other bugs, that may be just right.

I would like to go ahead and commit setlocale-main-harden-v1.patch, which is a
good thing to have regardless of what happens with gettext.

Thanks,
nm