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]

Re: IRIX 6 configuration sets MD_EXEC_PREFIX


>>>>> "Jim" == Jim Wilson <wilson@cygnus.com> writes:

    Jim> I don't object to a similar scheme that works at configure
    Jim> time instead of compile time.  However, how are you going to
    Jim> do the configure time check?  Are you proposing testing to
    Jim> see if the first ld on the user's path works?

No.  We don't check to see if `rm' works at configure-time.  Instead,
we expect that the `rm' in the builder's path removes files.  I
suggest just looking for `ld' at configure-time, falling back on
expected defaults (/usr/ccs/bin/ld, etc.) if no `ld' is found.  If the
builder specifies a bad `ld', things will break.  But, that's at
configure/build time, not at run-time, so I doubt you'll see nearly so
many false bug reports.  Note that this scheme, like the
MD_EXEC_PREFIX bit, avoids searching the path at run-time, which is
what leads to the confusion.  All I'm suggesting is to calculate
MD_EXEC_PREFIX, rather than hard-wiring it into a configuration file.
I guess I would propose the following search order:
 
  o ld in --prefix/bin, if any
  o ld found at configure-time, if any
  o ld in MD_EXEC_PREFIX/bin, if any
  o ld in PATH

The first bullet handles the one-tree build case, and the typical
case for installing GNU binutils.  Then, we fall back on the `ld' the
original builder of GCC had in her path at build-time.  Next, the `ld'
that we expect the system to have.  Finally, as a last resort, we
search the PATH.

    Jim> This seems like an awful lot of trouble to me.  Adding a
    Jim> MD_EXEC_PREFIX_1 macro would be easier.

I'll leave the decision in your capable hands.  I think I'm out of my
depth in terms of all of the cross-compilation issues.  But, I think
the important issue is that GCC should behave in as much the same way
as possible on all targets.  So, if GCC searches the path to find `ld'
on Linux, it should do so on Irix.  Certainly, if it does so on a
system like Solaris (where you can use the non-GNU linker), it should
do so on Irix.  But, I really don't see why whether or not the system
linker is the GNU linker should make a difference.

Of course, we need some way to fall back on an expected location for
these tools.  On Linux, they are probably in /usr/bin, on Irix in
/opt/MIPSpro/bin or /usr/bin, and on Solaris in /usr/ccs/bin, if
memory serves.  I'm not suggesting that users should have to put these
directories in their path.

Right now the behavior is inconsistent acrros platforms.  That's
confusing.  Adding MD_EXEC_PREFIX_1 won't undo that inconsistency.

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com


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