| What exactly is this fixing? It's easier to look for lossage in that case.

This fixes some weird lossage caused by passing structures by
value.  It seems to be caused when the compiler tries to figure
out what registers to stuff the structure into.

|
| James
|
| >
| >
| >I am in the process of backporting a C++ compiler fix from
| >gcc-current to our in-tree gcc and need some help testing.
| >Since gcc has a single back-end for both 32-bit and 64-bit
| >sparc, any changes could break either port.
| >
| >I'm looking for volunteers to test the accompanying patch
| >on 32-bit sparc.  I suppose the best way would be to run
| >the gcc testsuite with the current compiler first, check
| >for breakage, then do the same with this fix and see if
| >anything has changed.
| >
| >To run the testsuite you need to:
| >
| >1) Get a gcc source tree
| >
| >2) Get deja-gnu from:
| >
| >ftp://gcc.gnu.org/pub/gcc/infrastructure/dejagnu-20010126.tar.gz
| >
| >3) Build and install that version of dejagnu
| >
| >4) Configure gcc (preferably with a separate build directory)
| >
| >5) do a `make check' in the top-level gcc build directory

Unfortunately, this does not use the installed compiler.  Sigh.
So you need to stick the testsuite tree on the end of
src/gnu/dist/toolchain/gcc/ and use that instead.

This really should be automated.

Eduardo