gnulib-tool.py and gnulib-tool.sh disagree on AC_REQUIRE([AM_PROG_CC_C_O])

[Collin Funk]

Hi Bruno,

Could you please help me find the correct solution to this issue?

In a few of the test cases there is the following diff in
gnulib-comp.m4:

$ diff -u ./test-oath-toolkit-2.result/liboath/gl/m4/gnulib-comp.m4 
tmp700002-result/liboath/gl/m4/gnulib-comp.m4
--- ./test-oath-toolkit-2.result/liboath/gl/m4/gnulib-comp.m4   2024-03-28 
12:16:32.375344488 -0700
+++ tmp700002-result/liboath/gl/m4/gnulib-comp.m4       2024-03-28 
12:18:09.135468095 -0700
@@ -42,6 +42,7 @@
   AC_REQUIRE([gl_USE_SYSTEM_EXTENSIONS])
   AC_REQUIRE([gl_PROG_AR_RANLIB])
 
+  AC_REQUIRE([AM_PROG_CC_C_O])
   # Code from module absolute-header:
   # Code from module alignasof:
   # Code from module alloca-opt:

This is output in line 6030 of gnulib-tool.sh:

    if ! $gnu_make && test -n "$uses_subdirs"; then
      echo "  AC_REQUIRE([AM_PROG_CC_C_O])"
    fi

However, since gnulib-comp.m4 is emitted before lib/Makefile.am and
tests/Makefile.am, I don't think it would be possible for
$uses_subdirs to be set. Therefore this line would never be printed.
This grep shows it only shows up in the tmp* directories from
gnulib-tool.py, which seems to back up my interpretation:

    $ grep -r 'AC_REQUIRE(\[AM_PROG_CC_C_O\])' *
    tmp699959-result/m4/gnulib-comp.m4:  AC_REQUIRE([AM_PROG_CC_C_O])
    tmp700002-result/liboath/gl/m4/gnulib-comp.m4:  AC_REQUIRE([AM_PROG_CC_C_O])
    tmp700100-result/m4/gnulib-comp.m4:  AC_REQUIRE([AM_PROG_CC_C_O])

In GLImport.gnulib_comp() we have:

        for module in moduletable['main']:
            # Test whether there are some source files in subdirectories.
            for file in module.getFiles():
                if (file.startswith('lib/') and file.endswith('.c')
                        and file.count('/') > 1):
                    uses_subdirs = True
                    break
        if not gnu_make and uses_subdirs:
            emit += '  AC_REQUIRE([AM_PROG_CC_C_O])\n'

which is why it is printed.

I don't think we *need* this macro but there might be a compatibility
reason we want it that I am not aware of [1].

[1] https://git.savannah.gnu.org/cgit/automake.git/tree/NEWS#n496

Collin