This is the mail archive of the
mailing list for the GCC project.
re. [-no-undefined issues] Re: [PATCH] Fix build on PE-COFF targets and PR40125
On 28 Nov 2010, at 19:17, Ralf Wildenhues wrote:
* IainS wrote on Sun, Nov 28, 2010 at 07:24:42PM CET:
On 28 Nov 2010, at 18:16, Ralf Wildenhues wrote:
Libtool has its undefined flag set to something like this. Why
would it be necessary to try to override it? Also, the libtool
script will eat those arguments not prefixed with -Wl, anyway. To
me it seems the darwin special case is not needed at all.
If it causes problems on Darwin, then it's probably better to fix
libtool.m4 to adjust.
it seems to override libtool's normal choice
Where would that setting be?
... the causality is conjecture, the flag (_lt_dar_allow_undefined) is
present in lib*/configure.
... I'll try and track down where it comes from (that's just what came
out of grepping configure).
this causes a problem for plugins that need -undefined
Ah. But *that* is a problem for plugins only, not for libraries like
libquadmath or libgfortran, right?
yes, that's what I think.
It seems a bit suboptimal to not see a specific build failure,
all details here. Can you please open a PR with all the information
a cut and paste of the build failure, how configure was invoked etc?
The comments here are not in lieu of a proper PR - but it might take a
while to re-construct what happened.
The fail was of a X from i686-darwin9 to cris-elf ...
... the lto-plugin failed to build because of an unresolved symbol
(that needs to come from crt).
[this is not noticed on Darwin native, since we don't use the lto-
it was tracked down to the inclusion of '-no-unresolved' as a flag to
libtool -- which seemed (conjecture) to override libtool's normally
correct choice for "-undefined"
The flag does not seem to be present in current trunk - so I'll have
to look for the patch and cook something up to replicate the issue.