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]

Re: [PATCH] style fix for aix51.h and rs6000/mach.h


At 12:52 03.05.2001, David O'Brien wrote:
>On Thu, May 03, 2001 at 12:41:37PM +0200, Franz Sirl wrote:
> > You ask on the lists on what is intended?
>
>I don't need to for what I was doing -- a __STYLE__ cleanup.

Yes, I know. Nevertheless it seems like mentioning potential problems you 
run across during your style cleanup would be a very good thing and _I_ 
would certainly expect that.


> > You ask me to clean up stuff caused by you not following some simple rules?
>
>Ok, then I'm asking you to now go thru gcc/config/*/ and make it so none
>of the platform headers include any other header.  This is the simple
>rule, and you'll see it isn't followed by large extent.[*]

Well, if you look at the changlogs you'll see we certainly have the same 
goal here. I guess the rs6000 dir is currently the cleanest in this respect.

>In the rs6000 directory lynx.h needs to be unwound.  config/linux.h also
>needs to be unwound (stop including svr4).  arm/ needs a lot of cleanup
>too.

Fixing rs6000/lynx.h will be messy and I don't know who can test it, that's 
why I left it as-is. You have to preprocess old and new header style and 
compare the resulting #define's one by one.

config/linux.h is really a config/linuxx86.h, so fixing it means fixing x86 
at the same time. A good start is usually to remove nearly all #undef's 
from config/elfos.h and config/svr4.h, that brings up a lot of problems 
quickly. Maybe it should be policy that all #undefs in the header files 
below config/ need to be reasonably documented. I just see that a lot of 
new #undef's entered elfos.h recently, that shouldn't happen anymore. I'll 
prepare a patch that adds a comment at the top of elfos.h that there should 
be now new #undefs in there, cause they most probably point to an inclsuion 
order violation.

I can say nothing about the ARM dir.

On aix51.h, I guess thats just an oversight, as it went in shortly after my 
removal of nearly all #include directives in rs6000/ back in January.

David, can you comment if this double inclusion is really necessary? It 
looks like a lot of the undefs in aix51.h are solely necessary because of 
the double inclusion...

Franz.


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