Re: [PATCH] Do not decorate symbols as dllexport on Cygwin

[Corinna Vinschen]

Hi Bruno,

On Feb  5 22:22, Corinna Vinschen wrote:
> On Feb  5 21:41, Bruno Haible wrote:
> > Another option — since we are talking about a single symbol and a single
> > platform — would be if the locking for setlocale_null were not necessary
> > on Cygwin in the first place. I determined that it is necessary by running
> > the unit test gnulib/tests/test-setlocale_null-mt-all.c [3] on Cygwin:
> > without the lock, it crashed within less than 1 second. Could the
> > implementation of setlocale() in Cygwin be changed in such a way that this
> > test does not crash? Then the lock would be necessary.
> 
> Well, we could do that by adding Cygwin-internal locking to setlocale
> calls.  But that would only be available in the next Cygwin version
> of course.
> 
> May I ask what's the idea to provide a thread-safe setlocale?  It was
> never defined as thread-safe and POSIX explicitely mentions that.  Any
> application expecting to call setlocale thread-safe is broken by design.
> It should use the newlocale/duplocale/uselocale/freelocale API instead,
> isn't it?

Ahhh, I finally see what's going on.  The problem is not thread-safety
as such, but thread-safety when reading the value of the LC_ALL category.

Glibc's setlocale isn't entirely thread-safe either, but there's a
difference:

- GLibc creates the global strings returned by setlocale(LC_xxx, NULL)
  at the time the locale data is changed.  All setlocale(LC_xxx, NULL)
  calls only return pointer to strings created earlier.

- Cygwin or, better, newlib, also return a pointer to global strings.
  However, while the global strings for the specific categories are
  created when the locale is changed, the string returned for LC_ALL
  gets created on the fly when setlocale(LC_ALL, NULL) is called.
  That's why test-setlocale_null-mt-all fails almost immediately.

I created a patch to newlib's setlocale to tweak the LC_ALL string
each time the locale is changed, while setlocale(LC_ALL, NULL)
just returns the already prepared string.

https://cygwin.com/git/?p=newlib-cygwin.git;a=commitdiff;h=23e49b18ce39

This patch will be in the next Cygwin release 3.4.6.

I'm just a bit fuzzy what patches will be required for gnulib now...


Thanks,
Corinna