bug#18744: Problem with am__is_gnu_make: Arg list too long

[Stefano Lattarini]

Reference: http://debbugs.gnu.org/18744

On 10/16/2014 04:57 AM, Daniel Richard G. wrote:
> Hello gnus,
>
> I was hacking on libarchive for compatibility with an older Tru64
> system, in conjuction with Automake 1.14.1, when I encountered a problem
> during "gmake check":
>
> [...]
> gmake  check-TESTS
> gmake[2]: Entering directory `/tmp/libarchive-3.1.2/_build'
> gmake[3]: Entering directory `/tmp/libarchive-3.1.2/_build'
> ../build/autoconf/test-driver: line 107: 23320 Segmentation fault      "$@" >$log_file 
> 2>&1
> FAIL: libarchive_test
> FAIL: bsdtar_test
> FAIL: bsdcpio_test
> FAIL: bsdcat_test
> gmake[3]: execvp: bash: Arg list too long
> gmake[3]: *** [test-suite.log] Error 127
> gmake[3]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake[2]: *** [check-TESTS] Error 2
> gmake[2]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake[1]: *** [check-am] Error 2
> gmake[1]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake: *** [check] Error 2
>
>
> Where was that "Arg list too long" error coming from? I removed some
> leading "@" specifiers from some makefile recipe lines, and got this:
>
> [...]
> FAIL: libarchive_test
> FAIL: bsdtar_test
> FAIL: bsdcpio_test
> FAIL: bsdcat_test
> bases='libarchive_test.log bsdtar_test.log bsdcpio_test.log bsdcat_test.log'; 
> bases=`for i in $bases; do echo $i; done | sed 's/\.log$//'`; bases=`echo 
> $bases`; \
>         am__f_ok () { test -f "$1" && test -r "$1"; }; \
>         redo_bases=`for i in $bases; do \
>                       am__f_ok $i.trs && am__f_ok $i.log || echo $i; \
>                     done`; \
>         if test -n "$redo_bases"; then \
>           redo_logs=`for i in $redo_bases; do echo $i.log; done`; \
>           redo_results=`for i in $redo_bases; do echo $i.trs; done`; \
>           if (target_option=n; case ${target_option-} in ?) ;; *) echo "am__make_running_with_option: 
> internal error: invalid" "target option '${target_option-}' specified" >&2; exit 1;; 
> esac; has_opt=no; sane_makeflags=$MAKEFLAGS; if test -n ' Makefile cat/.deps/bsdcat-bsdcat.Po 
> cat/.deps/bsdcat-cmdline.Po cat/test/.deps/bsdcat_test-main.Po cat/test/.deps/bsdcat_test-test_0.Po 
> cat/test/.deps/bsdcat_test-test_empty_gz.Po cat/test/.deps/bsdcat_test-test_empty_lz4.Po 
> cat/test/.deps/bsdcat_test-test_empty_xz.Po cat/test/.deps/bsdcat_test-test_error.Po 
> cat/test/.deps/bsdcat_test-test_error_mixed.Po cat/test/.deps/bsdcat_test-test_expand_Z.Po 
> cat/test/.deps/bsdcat_test-test_expand_bz2.Po cat/test/.deps/bsdcat_test-test_expand_gz.Po 
> cat/test/.deps/bsdcat_test-test_expand_lz4.Po cat/test/.deps/bsdcat_test-test_expand_mixed.Po 
> cat/test/.deps/bsdcat_test-test_expand_plain.Po cat/test/.deps/bsdcat_test-test_expand_xz.Po 
> cat/test/.deps/bsdcat_test-test_help.Po cat/test/.deps/bsdcat_test-test_version.
Po cpio/.deps/bsdcpio-cmdline.Po cpio/.deps/bsdcpio-cpio.Po 
cpio/.deps/bsdcpio-cpio_windows.Po cpio/.deps/bsdcpio_test-cmdline.Po 
cpio/test/.deps/bsdcpio_test-main.Po cpio/test/.deps/bsdcpio_test-test_0.Po 
cpio/test/.deps/bsdcpio_test-test_basic.Po [...]
> That last "if test -n" statement includes what appears to be every *.Po
> and *.Plo file in the entire project---not helped, perhaps, by the fact
> that this project has a bazillion source files and only a single top-
> level makefile. Not only that, the output showed *two* full instances of
> this list in the same recipe line, presumably resulting from two
> expansions of the same problematic construct.
>
> This appeared to be $(am__make_dryrun) at first, but nested within was
> $(am__make_running_with_option), and in turn $(am__is_gnu_make):
>
>      am__is_gnu_make = test -n '$(MAKEFILE_LIST)' && test -n '$(MAKELEVEL)'
>
> To confirm that this was the problem, I added the following rule to the
> makefile...
>
>      print-makefile-list:
>              : $(MAKEFILE_LIST)
>
> ...and then tried
>
>      $ gmake print-makefile-list
>      :  Makefile cat/.deps/bsdcat-bsdcat.Po cat/.deps/bsdcat-cmdline.Po 
> cat/test/.deps/bsdcat_test-main.Po cat/test/.deps/bsdcat_test-test_0.Po 
> cat/test/.deps/bsdcat_test-test_empty_gz.Po [...]
>
> Yep, those *.Po/*.Plo files are makefile fragments that get include'd,
> so it makes sense that we end up with a huge MAKEFILE_LIST. If I change
> am__is_gnu_make to...
>
>      am__is_gnu_make = test -n '$(MAKE_VERSION)' && test -n '$(MAKELEVEL)'
>
> ...given that MAKE_VERSION appears to be a shorter example of a GNU-Make-
> only variable, then things work as they should:
>
> [...]
> gmake[5]: Nothing to be done for `all-am'.
> gmake[5]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake[4]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> ============================================================================
> Testsuite summary for libarchive 3.1.2
> ============================================================================
> # TOTAL: 4
> # PASS:  0
> # SKIP:  0
> # XFAIL: 0
> # FAIL:  4
> # XPASS: 0
> # ERROR: 0
> ============================================================================
> See ./test-suite.log
> Please report to address@hidden
> ============================================================================
> gmake[3]: *** [test-suite.log] Error 1
> gmake[3]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake[2]: *** [check-TESTS] Error 2
> gmake[2]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake[1]: *** [check-am] Error 2
> gmake[1]: Leaving directory `/tmp/libarchive-3.1.2/_build'
> gmake: *** [check] Error 2
>
>
> (At least as far as the test infrastructure is concerned!)
>
>
> --Daniel
Thanks for the report.  I agree with the spirit of your fix, but
unfortunately the $(MAKE_VERSION) macro seem present also in non-GNU
implementations of make (e.g., FreeBSD's).  Let me look for a better
variable...