Re: Finalizing fix committed for "[PATCH] sim: make sure to include strsignal prototype"

[Mike Frysinger <vapier@gentoo.org>]

[-- Attachment #1: Type: Text/Plain, Size: 1635 bytes --]

On Saturday 24 March 2012 14:24:14 Mike Frysinger wrote:
> On Saturday 24 March 2012 06:55:28 Hans-Peter Nilsson wrote:
> > I built and and started testing for bfin-elf+unix but got a lot
> > of assembly errors for values out-of-range (HEAD binutils,
> > perhaps the assembler isn't tested on a 32-bit host?)
> 
> should work, but i'll build it up to see

for most tests, it was just old code with too many leading digits (which 
didn't matter as we only cared about the truncated value which got tested).  i 
manually tweaked the tests to use 32bit values only.

there is one test (fact.s) which is arguably not a bug in the test as the 
source code specifies the real answer (which is larger than 32bits).  as the 
test runs and things start to overflow, this is fine as only the high bits get 
truncated away.  so we compare the truncated right answer to the truncated 
computed answer and are good to go.

at any rate, seems to be a bug in common code that only shows up if you are on 
a 32bit system and don't specify --enable-64-bit-bfd.  simple test case which 
hits the same problem regardless of target:
$ cat test.s
.if 6227020800 & ~0x7fff
.endif
$ ./as-new test.s 
test.s: Assembler messages:
test.s:1: Warning: left operand is a bignum; integer 0 assumed

the BFD64 logic in gas' expr.c is fairly light.  i'm not even sure it makes 
sense to key off of BFD64 since this code doesn't use bfd for number 
processing.  it's just deciding whether to automatically upgrade bignum's into 
int64's.  i'll post a patch to just delete the BFD64 logic there and see what 
people have to say.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]