On Thursday 20 October 2011, Peter Rosin wrote: > Stefano Lattarini skrev 2011-10-20 17:11: > > > Micro-nit: this would be clearer IMHO: > > > > # Interactions between `portability' and `extra-portability' > > # warning categories: > > # 1. `-Wextra-portability' must imply `-Wportability'. > > # 2. `-Wno-portability' must imply `-Wno-portability'. > > fine with me (but with -Wno-extra-portability at the end of 2). > Right, sorry. > >> +AUTOMAKE_fails -Wnone -Wextra-portability > >> +# The expected diagnostic is > >> +# Makefile.am:2: compiling `foo.c' with per-target flags requires > >> `AM_PROG_CC_C_O' in `configure.in' > >> +# .../lib/am/library.am: `libfoo.a': linking libraries using a > >> non-POSIX > >> +# .../lib/am/library.am: archiver requires `AM_PROG_AR' in > >> `configure.in' > >> +# Makefile.am:1: while processing library `libfoo.a' > >> +grep '^Makefile.am:2:.*requires.*AM_PROG_CC_C_O' stderr > >> +grep '/library.am:.*requires.*AM_PROG_AR' stderr > >> + > > No need to also grep the filenames here. Similarly for other usages > > throughout the file. > > Do you mean /library.am only, or do you also mean Makefile.am:2:? > Both.
> If you do not think Makefile.am:2: is needed, then why did you want me > to grep `FILE:LINENO' for the other patch, but not here? > Because here we are only interested in checking the relationship between the `portability' and `extra-portability' warning classes, not to ensure the competelness of the error messages; these are orthogonal concerns IMHO, and should better pertain to distinct tests (unless keeping them separate would become to cumbersome, but I think this is not the case here). Do you agree? > >> diff --git a/tests/extra-portability2.test b/tests/extra-portability2.test > >> new file mode 100755 > >> index 0000000..9f4948b > >> --- /dev/null > >> +++ b/tests/extra-portability2.test > >> @@ -0,0 +1,57 @@ > >> +#! /bin/sh > >> +# Copyright (C) 2011 Free Software Foundation, Inc. > >> +# > >> +# Make sure that extra-portability is not enabled by --gnits, --gnu > >> + > >> +cat >>configure.in <<END > >> +AC_PROG_CC > >> +AC_PROG_RANLIB > >> +AC_OUTPUT > >> +END > >> + > >> +cat >Makefile.am <<END > >> +EXTRA_LIBRARIES = libfoo.a > >> +libfoo_a_SOURCES = foo.c > >> +END > >> + > >> +$ACLOCAL > >> + > >> +# Make sure the test is useful. > >> +AUTOMAKE_fails > >> > > How is it that automake is failing here, without explicilty > > using `-Wextra-portability'? > > -Wall implies *all* warnings, remember? > Right, sorry *blush*. But then ... > >> +$AUTOMAKE --foreign > >> +$AUTOMAKE --gnu > >> +$AUTOMAKE --gnits > >> + ... why is not automake failing here? $AUTOMAKE still contains the `-Wall' option ... Oh, got it, you are relying on the fact that a strictness option also reset the warnings to its corresponding defaults. But this is was a long-standing bug of automake, which has been fixed in the master branch. So, once msvc is merged back, this test will porbably start to fail ... oh well, I guess we can cross the bridge when we come to it. Thanks, Stefano
