Re: remaining tasks before Autoconf release

Paul Eggert wrote:
> > 3) The hint
> >
> >     Did you mean to build a 64-bit binary? (E.g.,
> >     'CC="gcc -m64"'.)
> >
> > should not occur on a 32-bit OS. It should only occur on bi-arch systems
> > (64-bit OS, 32-bit $CC).
> 
> If we could come up with a reliable way to distinguish between the two, 
> I suppose we could use the abovementioned message on bi-arch platforms 
> and some other message (e.g., "Wouldn't it be better to build for a 
> platform that won't stop working in 2038?") on 32-bit time_t only 
> platforms.

The code in lib/autoconf/specific.m4 already makes this distinction,
and I think it is reliable:

  * The test in AC_SYS_YEAR2038 (specific.m4 lines 211..219) detects
    a bi-arch system in which a 64-bit binary is year-2038-safe
    whereas a 32-bit binary is not.
    This is exactly the case in which the suggestion "Did you mean
    to build a 64-bit binary?" makes sense.

  * In the other cases
      - 32-bit OS, or
      - bi-arch system in which neither a 32-bit nor a 64-bit binary are
        year-2038-safe,
    AC_SYS_YEAR2038_RECOMMENDED (specific.m4 lines 228..233) the
    suggestion "Did you mean to build a 64-bit binary?" is pointless.

Due to the AC_REQUIRE in line 227, the AC_SYS_YEAR2038 test is always
executed before the AC_SYS_YEAR2038_RECOMMENDED test.

> > -       Did you mean to build a 64-bit binary? (E.g., 'CC="${CC} -m64"'.)
> 
> Doesn't removing this line suppress the suggestion even for bi-arch 
> systems where the suggestion is useful? That seems like a step backwards.

No. As you can see, either from the specific.m4 code or from the testing
that I reported, there are two AC_MSG_FAILURE invocations. Removing this
line from the invocation in AC_SYS_YEAR2038_RECOMMENDED does not remove it
from the invocation in AC_SYS_YEAR2038.

So, I continue to suggest applying the patch from
https://lists.gnu.org/archive/html/autoconf/2023-05/msg00002.html .

Bruno