On 08 Nov 2021 16:05, Corinna Vinschen wrote:
> On Nov  8 06:46, Mike Frysinger wrote:
> > On 08 Nov 2021 11:05, Corinna Vinschen wrote:
> > > On Nov  6 20:21, Mike Frysinger wrote:
> > > > i stumbled across _COMPILING_NEWLIB and it seems to be what i want: a symbol
> > > > that indicates the code currently being compiled is newlib itself so that the
> > > > header can change behavior for that environment specifically.  is that what
> > > > it's meant for ?
> > > > 
> > > > if so, why does it seem to be inconsistently defined ?  newlib/configure.host
> > > > will add it for a few random targets, as does the mips-specific
> > > > newlib/libc/machine/mips/Makefile.am, as do a few specific winsup/cygwin/
> > > > files.  it feels like the patch below is what we should have.
> > > > 
> > > > if that's not what this is for, is there a define that has this meaning ?
> > > > in the glibc & gnulib world, the plain _LIBC define indicates this.
> > > 
> > > _COMPILING_NEWLIB might be older than that.  In Cygwin we certainly need
> > > it during build.  Your patch looks good to me, did you test it for some
> > > targets?
> > 
> > yes, i tested it for bfin-elf and with a change that needed it in ctype.h.
> > the ctype.h change didn't work until i updated the build this way.
> 
> uhm... why does a change in a header file depend on the build system?
> That sounds weird.

i'm making a fix to ctype.h that would utilize _COMPILING_NEWLIB as part of
its fix.  i was holding off posting that until the question of this symbol
could be sorted out.

> I tested building on Cygwin, which looks good.
> 
> So, here's a question: The patch is ok, just a change to the commit
> message would be nice.  However, would you like to take the opportunity
> to change _COMPILING_NEWLIB to _LIBC throughout?  That's something we
> should have done long ago, I guess.

let me do it as a series.  _LIBC is already used in newlib, so i'm afraid
it might not be as clean as just renaming the define.  but i'll give it a
shot and see what blows up.
-mike