Your GAS change caused a testsuite regression for cris-elf

[Hans-Peter Nilsson]

(Sorry for this duplicate, Alan, I see I missed CC:ing binutils@.)

Well, actually the test is a.out+stabs-specific (the ELF variant
in undef3.d
doesn't fail):
...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-cris/cris.exp ...
FAIL: ld-cris/undef2
...

ld.log:
./ld-new  -L/h/hp/binutils/cvs_latest/src/ld/testsuite/ld-cris  -mcrisaout -o tmpdir/dump tmpdir/dump0.o tmpdir/dump1.o
failed with: <tmpdir/dump1.o:tmpdir/dump1.o:96: undefined reference to `globsym1'>, expected: <.o:/blah/foo.c:96: undefined reference to `globsym1'$>
tmpdir/dump1.o:tmpdir/dump1.o:96: undefined reference to `globsym1'
failed with: <tmpdir/dump1.o:tmpdir/dump1.o:96: undefined reference to `globsym1'>, expected: <.o:/blah/foo.c:96: undefined reference to `globsym1'$>
FAIL: ld-cris/undef2

Initially, it looked like a typo somewhere.  Looking at the
changes in the time-frame back to the previous change, I had
trouble believing that this change (src/gas):

2005-09-20  Alan Modra  <amodra@bigpond.net.au>

	* read.c (pseudo_set): Set segment of expression syms to expr_section.

actually caused the regression, but it does, or at least causes
it to be exposed.  I checked by testing with and without (cvs
diff -D yesterday -l | patch -R) that change.  I guess something
strange and wonderful is going on.  Luckily for me, I can just
point at the regression and say "waaah!".

brgds, H-P