On Saturday 07 August 2010 18:19:30 Samuli Suominen wrote: > On 08/07/2010 05:58 PM, Alexis Ballier wrote: > > On Saturday 07 August 2010 17:45:02 Samuli Suominen wrote: > >> Could use eyepair or two to doublecheck the example code in: > >> > >> [1] http://bugs.gentoo.org/attachment.cgi?id=241779 > >> [2] http://bugs.gentoo.org/show_bug.cgi?id=331527 > > > > You should use AC_TRY_LINK or something similar to check for toolchain > > support. Hardcoding a given set of strings matching the version is > > generally not a good way of writing a check; if you check something, > > check it for real. > > > > Also if you pass -Wl,--as-needed to the compiler that's what you should > > check rather than if the supposed linker supports --as-needed; I suppose > > AC_TRY_LINK will do that. > > I tend to agree but have one concern: > > What if some non-GNU ld's are accepting the syntax of -Wl,--as-needed > but is in fact doing something entirely different? > > I just remember one of our prefix devs, not long ago, saying that > happens. I think it was on this ML on different thread but sorry, can't > find the post now... > > Note that flag-o-matic.eclass's check in function no-as-needed() is > checking for GNU as well.
you can check for both if you prefer, the main problem I see in the original check is that its not checking at all if as-needed works and appends a flag to gcc while it checks only if ld accepts the flag A.
