Re: PATCH: Support LD_SYMBOLIC and LD_SYMBOLIC_FUNCTIONS

[Nick Clifton <nickc at redhat dot com>]

Hi H.J.

> > At the very least H.J. you ought to extend your patch so that the linker 
> > clearly tells the user that it is changing its behaviour because of an 
> > environment variable.
>
> We have
>
> [hjl@gnu-2 ld]$ grep getenv *.c
> ldemul.c:  char *from_outside = getenv (TARGET_ENVIRON);
> ldmain.c:  demangling = getenv ("COLLECT_NO_DEMANGLE") == NULL;
> ldmain.c:  emulation = getenv (EMULATION_ENVIRON);
>
> Linker does support environment variable.

Indeed, but it really should be telling the user about the use of these 
variables as well (if they are defined).

> >   Is it really that much easier to set an environment variable for a 
> > particular package, as opposed to say, altering its Makefile or build 
> > script ?
>
> Can you try to enable -Bsymbolic-functions for all or selective
> packages, e.g. GTK+, Pango, Firefox, Open Office, ...  in a distro,
> like FC6 or FC7?

No.  But then I do not do things like that and I do not have a build 
environment set up to run this kind of test.  I was asking you, since I 
think that you do have this kind of knowledge, if you could tell me why 
using environment variables in this way would be a good thing.  Are the 
alternatives really so bad ?  If this feature is only for user 
convenience, then we should really consider whether it is worth it.  If 
supporting this environment variable will make our 
(binutils-maintaining) lives harder then do we really want to do it ?

Cheers
  Nick