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]

Re: Compile performance of Linux kernels in mainline gcc


On Sat, Oct 30, 2004 at 01:13:39PM -0700, Zack Weinberg wrote:
> That may not be the whole story.  When I trigger the same diagnostic
> from mainline GCC in a terminal that doesn't support Unicode, with
> LANG set to en_US.UTF-8, I get this:
> 
> test.c:5: warning: pointer targets in passing argument 1 of ###foo###
> differ in signedness
> 
> where each # is a different garbage character.  In other words,
> there's garbage around the function name, but the diagnostic is still
> readable.  I want to know why Andi is seeing nothing but garbage after
> the "of".

It's an standard xterm from XFree86 4.2, logged in over ssh 
to another machine with the UTF-8 locale. ssh doesn't attempt
to negotiate locales.

> 
> There may not be much we can do about this in any case - there is no
> programmatic way to probe whether the terminal supports Unicode that I
> know of; we are stuck trusting the locale.

Well, I already posted a patch that eliminates the problem ;-)

-Andi

P.S.: I traced down why the 4.0 kernel didn't boot now. The early
boot code missed a -fno-strict-aliasing flag (now fixed) 
and another function with a loop gets miscompiled at -O2, works at -O0.
I haven't a runnable test case yet unfortunately.


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