Patches attached to make libpq thread-safe, now uses strerror_r(),
gethostbyname_r(), getpwuid_r() and crypt_r() where available. Where
strtok() was previously used strchr() is now used.
Thanks, Lee.
Lee Kindness writes:
> Tom Lane writes:
> > Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > > We have definatly had requests for improved thread-safeness for libpq
> > > and ecpg in the past, so whatever you can do would be a help. We say
> > > libpq is thread-safe, but specifically mention the non-threadsafe calls
> > > in the libpq documentation, or at least we should.
> > We do:
> > http://www.ca.postgresql.org/users-lounge/docs/7.3/postgres/libpq-threading.html
> > But Lee's point about depending on possibly-unsafe libc routines is a
> > good one. I don't think anyone's gone through the code with an eye to
> > that.
>
> Right, so a reasonable angle for me to take is to go through the libpq
> source looking for potential problem areas and use of "known bad"
> functions. I can add autoconf checks for the replacement *_r()
> functions, and use these in place of the traditional ones where
> available.
>
> If any function is found to be not thread-safe and cannot be made so
> using standard library calls then it needs to be documented as such
> both in the source and the aforementioned documentation.
>
> This approach avoids any thread library dependencies and documents the
> current state of play WRT thread safety (i.e it's a good, and needed,
> basis for any later work).
>
> ECPG is a separate issue, and best handled as such (it will need
> thread calls). I'll post a patch for it at a later date so the changes
> are available to anyone who wants to play with ECPG and threads.
>
> Ta, Lee.