Re: testsuite: 24 25 64 failed

[Nix]

On 24 Mar 2008, Ralf Wildenhues spake thusly:

> * Nix wrote on Mon, Mar 24, 2008 at 12:39:33PM CET:
>> > * Nix wrote on Fri, Mar 21, 2008 at 08:01:13PM CET:
>> >> 
>> >> The underlying cause of test 24's failure appears to be that LDFLAGS is
>> >> not consistently propagated down: note that a lot of the link steps do
>> >> not pass in my LDFLAGS of -specs=specs.no-as-needed, so get linked with
>> >> --as-needed (which is the default on my system) and subsequently
>> >> fail. (FWIW the only things I've ever seen fail with --as-needed are
>> >> linker-semantics testsuites like this one. :)
>> >
>> > First, you can try
>> >   LDFLAGS=-Wc,-specs=specs.no-as-needed
>> 
>> It seems to me that adding -Wl to the LDFLAGS wouldn't do anything
>> useful: it's the compiler driver that interprets the specs file, not the
>> linker. (And -Wc doesn't exist :) )
>> 
>> Hm, yes, it's not useful:
>
> Ah, sorry for not being precise.  Just try
>
>   make check-local TESTSUITEFLAGS='-v -d -x -k runpath 
> LDFLAGS=-Wc,-specs=specs.no-as-needed'

Ooo, you can pass LDFLAGS inside TESTSUITEFLAGS! Neat.

... nonetheless, that gets transformed into -Wl, which, as stated, is
not useful:

./runpath-in-lalib.at:55: $LIBTOOL --mode=link $CC $CFLAGS $LDFLAGS -o m 
m.$OBJEXT libb.la
+ /usr/packages/libtool/i686-hades/libtool --mode=link gcc -g -O2 
-Wc,-specs=specs.no-as-needed -o m m.o libb.la
stderr:
/usr/bin/ld: unrecognized option '-specs=specs.no-as-needed'
/usr/bin/ld: use the --help option for usage information
collect2: ld returned 1 exit status

A separate test (run by hand in testsuite.dir/24) proves this:

$ /usr/packages/libtool/i686-hades/libtool --mode=link gcc -g -O2 
-Wc,-specs=specs.no-as-needed -o m m.o libb.la
libtool: link: gcc -g -O2 -Wl,-specs=specs.no-as-needed -o .libs/m m.o  
./.libs/libb.so 
/usr/packages/libtool/i686-hades/tests/testsuite.dir/24/.libs/liba.so 
-Wl,-rpath -Wl,/usr/packages/libtool/i686-hades/tests/testsuite.dir/24/inst/lib


-specs=... isn't a linker flag: it's a compiler driver flag.

However, if I try

make check-local TESTSUITEFLAGS='-v -d -x -k runpath 
LDFLAGS=-specs=specs.no-as-needed'

the LDFLAGS never get passed down at all, so we get the other class of
failure.

> (all on one line).  And yes, `-Wc,' exists as both a documented compiler
> flag (for which this particular instance is _not_ useful), and as a

Not in my compiler, nor in any upstream snapshot I've checked
(admittedly my upstream snapshot is a month or so out of date: perhaps
it's been added since then?)

> documented libtool link mode flags.

Aha. I managed to overlook that bit of the docs.

>                                      (For issues such as this one, an
> Automake variable LIBTOOL_LDFLAGS would be useful).

Indeed it would.

-- 
`The rest is a tale of post and counter-post.' --- Ian Rawlings
                                                   describes USENET