This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v2] Remove defines to builtins in string2.h
Joseph Myers wrote:
> Unless you also change all calls to __stpcpy to call stpcpy directly, you
> are potentially losing built-in function optimizations by removing the
> __stpcpy macro. glibc contains calls to __stpcpy as well as direct ones
> to stpcpy. Because of the other redirections, it's safe to change
> __stpcpy calls to stpcpy calls in glibc as a preparatory patch. Likewise
> __mempcpy.
Right, it does seem there are a few cases where __stpcpy may be optimized.
So for this patch I'll keep the #define __stpcpy until existing calls to __stpcpy
have been changed.
> I think you need to separate the case of
>
> #define foo(x) __builtin_foo (x)
>
> from all the other cases, where a more complicated macro expansion is
> involved (in which case we need to see if GCC has relevant optimizations
> or have a reason why they are appropriate to discard), or where the macro
> is __foo instead of foo.
This patch doesn't remove any optimizations - as I mentioned in the description,
the strncat macro is unused as _USE_STRING_ARCH_strchr is always false
(and in the next version I'm removing that define as well).
> Also, patches like this are the sort where it would be useful to make sure
> that installed stripped shared libraries and executables are unchanged by
> the patch, or to have a reason why it results in any changes to the
> generated code.
Yes my goal was no diff, and that's true for the new version that also removes
all unused string defines.
Wilco