This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: 'broken words' misbehavior?
On Thu, 22 Feb 2007, Jan Beulich wrote:
> >>> Hans-Peter Nilsson <hp@bitrange.com> 22.02.07 13:40 >>>
> A feature that causes valid code to break... And yes, I looked (should I say
> stared) at this code (in dis-belief).
For context, you'd have to try and fit a difference in
(historically 32-bit, though that's not tested) labels in a
16-bit quantity, and fail to make it fit in 16-bit signed
for the feature to trig.
> Yes, x86 (in fact I did say that).
Sorry, my bad.
> I have no idea why it doesn't define this,
> and am hardly in the position (not knowing why it's the way it is) to request
> it being changed.
I don't see why not. You're maintainer of (part of) the x86
port, power-user enough to trig this *by accident*, and
supposedly have enough experience to know whether you want this
feature or not, and (just guessing) with enough user base
experience to know whether *they* want it or not.
> >If you instead decide to hack^Wfix this feature, there are
> >test-cases in the cris testsuite, cris-axis-elf. The feature is
> >not ready for removal, yet.
>
> I didn't want to suggest removing the feature either, but wanted to get
> an understanding what it's supposed to deal with and in turn whether the
> breakage is fixable.
Constraining the case where it trigs is certainly possible (see
read.c:emit_expr, for example to force one label known at the
time and in a code section), but I'd recommend x86 maintainers
to consider whether now is the time to #define WORKING_DOT_WORD.
brgds, H-P