Re: gnulib and glibc sources

[Bruno Haible]

Roland,

I'm CCing the gnulib list.

While discussing argp, Roland McGrath wrote:
> gnulib copies should not deviate in behavior, API,
> and ABI from the canonical functions in glibc; they should get only
> portability changes, and everything should be coordinated with the
> master sources in glibc.

The other gnulib participants can comment on this.

My 2 cents:

- gnulib attempts not to deviate from glibc, regarding the API.

- gnulib attempts not to deviate from glibc, regarding the behaviour of the
  functions, except where glibc violates POSIX or is clearly broken by our
  common judgement (examples: "[Bug libc/4586] printf crashes on some
  'long double' values", or regex each time Paul finds a regex bug in glibc).

- gnulib does not make ABI guarantees. A function called 'foobar' in glibc
  can very well be called 'rpl_foobar' in gnulib.

- Coordination between gnulib is glibc is hampered each time that the glibc
  maintainers refuse to apply a portability patch (examples: libc bug entries
  1054, 1220, 1241, 1281, 1285).
  Citation from one of these bugs, from one of the glibc maintainers:
     "I have no interest to waste my time on unsupported paltforms ..."

Bruno