[fortran,patch] Use new stdint.h-related macros in ISO_C_BINDING

[FX <fxcoudert at gmail dot com>]

Hi all,

The attached patch modifies the definitions of ISO_C_BINDING constants  
for INT*_T, INT_LEAST*_T and INT_FAST*_T so that we use the newly  
introduced stdint.h-related macros introduced by Joseph Myers (thanks  
Joseph)! Currently, we use ad hoc methods for determining INT*_T and  
INT_LEAST*_T, and we used -2 for INT_FAST*_T (which is what the  
standard mandates to mean "unsupported by the companion C compiler").

The longest part of the patch is the function  
get_typenode_from_name(): because target macros define names for the  
types (i.e., INT8_T might be "signed char") and, unlike the C front- 
end, the Fortran front-end doesn't have these type names registered,  
we need to do string comparison to find the appropriate type node.

The patch also adds the following ISO_FORTRAN_ENV named integer  
parameters: INT8, INT16, INT32, INT64 (which correspond to the INT*_T)  
and REAL32, REAL64 and REAL128 (which correspond to integers kinds of  
a given width).

Finally, the testcases are modified to check for the validity of all  
the constants. The tests might come to fail (while they previously  
passed) in some targets that haven't yet registered stdint.h-related  
target macros. For now, this job is done for all glibc and newlib  
targets, as well as cygwin, mingw, HP-UX and AIX [*]. Others beware.


Bootstrapped and regtested on x86_64-linux (both -m32 and -m64). OK to  
commit?

FX

Attachment: stdint.ChangeLog
Description: Binary data

Attachment: stdint.diff
Description: Binary data

[*]  A patch for darwin (by yours truly) has been gathering dust for a  
month and a half while waiting for a review