Re: No declaration for isblank

[Ludovic Courtès]

Hi,

Neil Jerram <address@hidden> writes:

> The problem was lines like the following in /usr/bin/gnulib-tool.
>
>     if grep AC_GNU_SOURCE "$destdir"/$m4base/*.m4 >/dev/null 2>/dev/null; then
>       echo "  AC_REQUIRE([AC_GNU_SOURCE])"
>
> So, if you create a file m4/workaround.m4, containing just
> "AC_GNU_SOURCE", and then rerun autogen.sh, you get a definition of
> gl_EARLY that includes AC_GNU_SOURCE, and the _GNU_SOURCE stuff in
> config.h.in.

Hmm, strange.  Here I have `extensions.m4' and `gnulib-comp.m4', both of
which require `AC_GNU_SOURCE'.  I believe `gnulib-tool --update' is
deterministic when starting with an `m4' directory only containing
`gnulib-cache.m4', so I don't understand how we can get different
results.

Did you try clearing the `m4' dir (expect for `gnulib-cache.m4') and
re-running `gnulib-tool --update'?  Assuming we both use the same Gnulib
snapshot, `m4' should contain the same files on both machines.  With
Debian's `gnulib-20071001+dfsg-1', I get this:

  $ sha1sum m4/*m4
  817660373a28ce877ae7efcce4c06ed0bfa93052  m4/alloca.m4
  4229cdd16c7512e1c03e1131a3441882bd7ced42  m4/extensions.m4
  fa99215ca5959f83eeb093ada23a82622abb194e  m4/gnulib-cache.m4
  e02774d532c8700d4a722b3f6189d487f0df721e  m4/gnulib-common.m4
  94693473bd8aa9eddd6e30af2213ebdd5ed31852  m4/gnulib-comp.m4
  4d3cae2790230ada32cb11340e797b42bd71e25c  m4/gnulib-tool.m4
  bf7b7750e03c02566c190514b96a2309daf0c1c1  m4/include_next.m4
  f3793585703e62a1c9058f81eb897b08b3342954  m4/onceonly_2_57.m4
  14aea4fa9d1cf27c71ac052ed00e4573e437f48f  m4/strcase.m4
  20a6d0a961ae4f4913ab2162fca088db24bc0b4e  m4/string_h.m4

  $ grep -n GNU_SOURCE m4/*m4
  m4/extensions.m4:45:#ifndef _GNU_SOURCE
  m4/extensions.m4:46:# undef _GNU_SOURCE
  m4/extensions.m4:72:  AC_DEFINE([_GNU_SOURCE])
  m4/gnulib-comp.m4:28:  AC_REQUIRE([AC_GNU_SOURCE])

(Note that once this is done, you also need to run `autoheader'.)

> That seems excessively magic to me, but I suppose we could ensure
> correct future behaviour by checking in such a file to CVS.  What do
> you think?

Normally, only `gnulib-cache.m4' needs to be versioned, since the rest
is installed based on what this file contains [0].  Nevertheless, we
could adjust this if it turns out to be necessary.

Thanks,
Ludovic.

[0] http://www.gnu.org/software/gnulib/manual/html_node/CVS-Issues.html