Re: don't add -isysroot to compiler flags if already in $CC

[Ralf Wildenhues]

Hi Peter,

* Peter O'Gorman wrote on Thu, Dec 01, 2005 at 04:59:32PM CET:
> Ralf Wildenhues wrote:
> >Also, I wonder why you leave out CFLAGS and such, so why not check
> >against $compile_command?

> You are right, of course. The original patch, by checking for CC and then 
> removing the flag broke building of executables.

Ah!  Heck, don't blame me, I didn't even see *that* bug.
Most of my comments were bogus itself.

> When we build shared libraries we use CC, not compile_command.
> compile_command is okay and never gets duplicates, but if a flag is in
> CC and we recognise it and put it in compiler_flags, there is the
> possibility of it being duplicated on the link line.

OK.  But I don't see where we make sure $compile_command never gets
duplicates.. anyway, I guess this part is fine, you'd have seen it.

> Is this version okay & forward port? A space and a tab check is quite 
> satisfactory for me, I don't see the need to go overboard here.

Any reason against using [      ] for this?  I mean, other than the fact
that the forward-port will need another set of []?

> Oops looks like I need to fix indentation too. Please pretend that that is 
> okay for now :)

Sure.

No, the patch is not ok.  You remove prev=darwin_framework, i.e.,
effectively the following argument will not be looped over, thus
it won't be quoted and end up in $libtool_args, thus $relink_command
will be wrong, thus relinking will fail, if only because there'll be
a `-framework' without argument in it.  Right?

If the argument to each of these commands ever need escaping,
you should also consider using $qarg instead of $arg in the
  darwin_framework)
matching part.  I think HEAD does something like that, too
(but the state is `xcompiler' there).

Cheers,
Ralf

> Index: ChangeLog
> 2005-12-01  Peter O'Gorman  <address@hidden>
> 
>       * ltmain.in [darwin]: Don't add -framework. -arch, -isysroot
>       to $compiler_flags if already in $CC with same arguments.