On Dec 6, 2016, at 8:59 AM, Eric Botcazou <ebotca...@adacore.com> wrote: > >> what you got there, looks more or less exactly like what I tried >> to fix with that patch a few months ago, but unfortunately >> we were unable to get a consensus on that approach: > > It's indeed the same underlying issue, but restricted to a specific case > where > reload and other passes do the right thing, so LRA needs to be on par IMO.
So, the only use of X I make during development (other than match_scratch), would be to debug and develop. In that context, I don't want any ICEs, and the semantics I want are no processing of the operand, and allow me to print it out (but not for the assembler, just a comment to read and see what it was). Doesn't strike me as unreasonable and I can't think of any better or sane semantics. Would be nice to allow those semantics as an ICE isn't as useful. So, from that perspective, I don't think patches to fix ICEs should be rejected on that basis alone.
