This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PR target/21299 (reload accepting invalid asm)

Richard Sandiford <> writes:

> Well sure, but if this is in reply to what I've been saying, I think
> it's a strawman.  It almost sounds like you think I was moaning about
> the inline asm implementaion.  I wasn't.  I was genuinely trying to the
> same thing as you; an incremental fix that I thought would be better.

Hmmm, I'm not really proposing an incremental fix at all.  I just
suggested a way to turn the ICE into an error message, which I think
is the minimal required change.

I understood the incremental fix you were proposing to be: have the
compiler check inline assembler code to make sure that it obeys
certain restrictions.  I believe that any incremental fix along the
lines of explicitly pre-rejecting inline assembler code (as opposed to
letting it run and then rejecting it only if it fails) would be
harmful, for the reasons I sketched out.  I accept that you disagree.

Sorry for proposing a strawman.  I was trying to be a little bit
careful since there seemed to be some misunderstanding somewhere.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]