Re: 'make check' on Win 2k

[Kevin Ryde]

Sisyphus <address@hidden> writes:
>
> FAIL: t-bswap.exe
> /bin/sh: ./t-bswap.exe: No such file or directory

Hmm.  Automake and libtool are fighting.  I wonder why nobody reported
this before.  Maybe nobody with mingw ever did a make check! :-)
(I've done a make check, but only in wine.)

The problem is that libtool makes a t-bswap shell script, which runs
.libs/t-bswap.exe.  But automake rewrites the check_PROGRAMS list to
say t-bswap.exe and tries to run that instead of of plain t-bswap.

Not sure why this isn't a problem on djgpp.  Maybe bash there quietly
looks for plain foo if it can't find foo.exe.

Maybe for mingw it depends on the shell too (ie. /bin/sh replacement).

In any case this should be getting better with libtool 1.5 in the next
release, where it uses a ".exe" wrapper (as well as a shell script).

> 'make check'
> proceeds to build the 5 mpn test executables, again in the wrong
> directory ('tests/mpn/.libs').

Re-building of the .exe's is a known problem (Windows DLL test
programs).  Annoying, but not harmful in itself.

> t-fib_ui.c:63: GNU MP assertion failed: (__gmp_fib_table[(i)+1]) != 0
> FAIL: t-fib_ui.exe

Thanks, lack of __GMP_DECLSPEC on the declaration of __gmp_fib_table
in gmp-impl.h.  You can add that there if you like.  The actual
mpz_fib_ui function should be fine, it's just the test program making
direct access to a non-exported variable.

-- 
Note: All followups to address@hidden only, please.