2011/1/11 Colin Guthrie <gmane at colin.guthr.ie>: > 'Twas brillig, and Maarten Bosmans at 06/01/11 01:39 did gyre and gimble: >> @@ -100,12 +100,12 @@ BINLDFLAGS = -static >> ?endif >> >> ?if OS_IS_WIN32 >> -AM_LDFLAGS+=-Wl,--export-all-symbols >> +AM_LDFLAGS+=-Wl,--export-all-symbols,--enable-auto-import -no-undefined >> ?WINSOCK_LIBS=-lwsock32 -lws2_32 -lwininet >> ?endif > > I've got no objection to the other bits in this patch (although I'd have > thought the auto tools stuff would use AM_LDFLAGS in any LDFLAGS section > automagically...) but I'm not sure we should be mandating -no-undefined > here... This is something we add in at the distro level when appropriate > normally. I thought so too, but then some apparantly random stuff didn't link successfully and it turned out that those bits had missing AM_LDFLAGS in their LDFLAGS. > But that said, I'm not really an expert on this stuff so perhaps there > is good reason to include this here (I appreciate this change will > require your previous patch)? What do people think generally? What I understand is that win32 DLLs require this flag, because if there are undefined symbols the whole library can't be build. Furthermore the flag should be given when building the library, but not when the configure script is testing for features, see http://www.mail-archive.com/libtool at gnu.org/msg09482.html Maarten
