Re: a bunch of MIPS GAS testsuite failures.

[cgd at sibyte dot com]

ica2_ts@csv.ica.uni-stuttgart.de ("Thiemo Seufer") writes:
> cgd@sibyte.com wrote:
> > So, more in the way of figuring out the state of the world.  I've now
> > expanded my builds to:
> > 
> > 	mips-elf mips64-elf
> > 	mipsel-elf mips64el-elf
> > 	mips-linux mipsel-linux
> > 	mips64-linux mips64el-linux
> > 	mips-ecoff mipsel-ecoff
> 
> First of all, thanks for doing this boring work.

Heh.  You're welcome.  It's not so bad when you've got a nice MP
server to do the builds quickly, and have done a little bit of
automation.  8-)

(I tend to do this stuff when doing a merge into my local source
trees...  But really given the number of regressions that seem to
happen on a fairly regular basis, I wonder if maybe it makes sense to
just go the rest of the way with the automation and run it regularly.)


> > 	These (and the XPASSes for "MIPS64 reloc 2")
> > 	are due to incorrect marking of little-endian MIPS as
> > 	"traditional mips."  I dunno why that was done, but it seems
> > 	to have been done intentionally.
> 
> The non-traditional mips is SGI'ish, they don't make little endian
> MIPS machines. So treating little endian as traditional is a wrong
> but half-working approach.

(This was also briefely discussed in
http://sources.redhat.com/ml/binutils/2001-07/msg00379.html, which
started off-list.)

To recap, I have two issues here:

(1) regardless of what should be done for little-endian MIPS in terms
of "traditional" vs not, right now the testsuite and the configury do
not agree as to what little-endian MIPS actually use.

(2) historically, little-endian MIPS (e.g. mipsel-elf, mips64el-elf,
etc.) have _not_ used "traditional MIPS."  People have, however, been
using them for embedded development for ... who knows how long.  It
would seem ... unwise to change them to use the "traditional MIPS" abi
at this point.



cgd