ARM EABI v3
Richard Earnshaw
rearnsha@arm.com
Wed Mar 24 17:05:00 GMT 2004
On Tue, 2004-03-23 at 18:57, Daniel Jacobowitz wrote:
> Yes, I understand that. My concern is that with just this patch, we've
> lost this valuable information (when someone chooses to use the new
> flag, that is). I've seen this discussed on and off for years, and
> there's still no standard way to record this information in EABI
> objects.
>
> I get a lot of bug reports from people trying to link incompatible bits
> of code together. At least now, I can tweak binutils to warn.
While I understand your concern, we can't fix everything at once.
Keeping the existing code, however, is not the way to address this: in
my experience it's wrong just as often as it's right. Take for example
the case of -msoft-float on arm-elf. The arm-elf toolchain does
generate soft-float code by default, but it fails to mark the flags
correctly, so objects produced that way end up being 'incompatible' with
each other.
Finding a new, and better, way of representing these options is a
Quality of Implementation issue. On the other hand, correctly marking
the version of AAELF to which we conform is a *correctness* issue. That
has to come first.
R.
More information about the Binutils
mailing list