This is the mail archive of the gcc-patches@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]

Re: [PATCH] testsuite: Add new effective target keywords


Am Montag, 18. April 2005 16:22 schrieb Joseph S. Myers:
> On Mon, 18 Apr 2005, [utf-8] BjÃrn Haase wrote:
> > . I do not know for shure where the board description files are supposed
> > to reside normally (my own setup for simulator tests for avr is a bit
> > particular) but I think that these files are supposed to reside somewhere
> > in your dejagnu installation directory.
>
> A solution involving every user on the platform needing to edit their
> unix.exp file manually, or to install an upgraded version of dejagnu, is
> not particularly useful in this regard.  The question isn't one of how an
> individual installation can be customised, but one of how GCC can do the
> right thing for a given platform out of the box without needing to modify
> the dejagnu installation or do anything more than GCC's standard "make -k
> check".
While I think that such kind of information logically should belong to the 
target board description, I agree that it would be better if one had a 
solution that is independent of dejagnu.
I'd like to suggest to use a method similar to the case of profiling. There, 
e.g. already is such a hard-coded list of targets in "target-supports.exp" so 
one might consider to extend the check_effective_target_support_trampolines 
procedure my patch is proposing so that it first looks at the board_info 
variable and in case that it does not find anything there, looks at a 
hand-coded selection list. ?

> As for whether int is more than 16 bits, the testsuite harness should be
> able to determine that for itself (for each multilib) by testing the
> compiler just as it does e.g. to test for availability of -fpic or iconv
> or ilp32 or lp64. 

Of course that would be desireable. It would be even better, if there were a 
reliable define symbol that could be used for compile switches, so that some 
tests need not to be skipped but could be adapted.

Yours,

BjÃrn


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