[Bug c++/23139] [3.4/4.0/4.1 Regression] -pedantic -ffast-math breaks working code

["mmitchel at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From mmitchel at gcc dot gnu dot org  2005-09-06 22:03 -------
The "floating constant exceeds range" message comes from interpret_float in
c-lex.c, which does test just "pedantic", rather than CPP_PEDANTIC(pfile).  So,
while the preprocessor warning about use of a hexadecimal floating constant does
seem to be like the other PRs, the actual pedwarn/error is coming from the fact
the C++ front end is pre-lexing tokens.

I agree that disabling warnings in expansions of macros from system headers is a
good idea, independently of whether or not any other changes might be
appropriate to deal with other uses of __extension__.  There are certain to be
system headers not using __extension__ in macro definitions, on some systems.  I
guess we could have the preprocessor insert a special token to mark the
start/stop of a macro expansion, with an indicator of the location of the macro
definition, including its system-headerness.  Then, the front end could
clear/reset pedantic when seeing these tokens.  This would also allow us to
issue diagnostics based on the location of the macro, rather than its user, if
we wanted.  A more brutal approach would be to have the preprocessor simulate
#include'ing the file from which the macro came when doing the expansion.  That
would not require updating the front-ends, but would be more confusing to users.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mark at codesourcery dot
                   |                            |com, joseph at codesourcery
                   |                            |dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23139