This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Autoconf 2.59 breaks libstdc++ checks


A change was made to autoconf sometime between 2.13 and 2.59
that breaks the efficacy of some of the checks in libstdc++,
or anything else for that matter. It has to do with AC_TRY_LINK.
The comment says that the change was to get around the fact
HP-UX CC under certain optimizing conditions (-O3) and the
AIX linker that does somethiong wrong too. Look at the comment
above AC_LANG_FUNC_LINK_TRY in c.m4.

The problem is the new code there does a #undef $1. This is
what is breaking things, and its wrong. It is a perfectly
valid thing for an OS to use #define to implement some function
in terms of another. For example, to implement isinf() in terms
of a call to fpclassify().

I will file a bug against this for autoconf but if they fix it,
would it be acceptable to require autoconf 2.60, and if they
dont fix it, how can we (I) best get a fix for this into
GCC? Add a modified check to each acinclude.m4?

Kean


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]