On Mon, 7 Mar 2005, Mark Mitchell wrote:
Roger Sayle wrote:
I truly hope you're not trying to suggest that it was me that introduced the concept of MIN_EXPR and MAX_EXPR as lvalues into the C++ front-end:
I thought you were the person who introduced changes to fold that caused it to generate these expressions when the GNU source extension did not appear. Anyhow, who did what to whom isn't important. :-)
I do appreciate the smiley, but I still feel unfairly maligned by such
accusations.
Am I incorrect about the fact that you introduced the changes to fold? If so, I humbly retract all previous maligning, and acknowledge the unfairness thereof. If not, then that change probably introduced this regression. But, I'm still not meaning to malign; until I've introduced zero regressions in my last thousand patches, who am I to talk?
If I seem frustruated, it's simply because we seem to be running into this issue again, after attempting to fix it before. That's my fault; I wasn't comfortable with the fix you came up with for 7503, but instead of pushing on that issue, I went along with build_modify_expr change to reconsitute COND_EXPRs after the fact.
The way I think about this is that G++ has long supported the GNU min/max expression extension -- and it's long been broken. Over the years, I've fielded several bug reports about that extension, and we've gradually cleaned it up, but mostly it's just been neglected. I think that you probably got fooled by the fact that the code is there into thinking that it works. :-) (I wouldn't be surprised if it works well today, but I'd also not be surprised if it doesn't work. It certainly didn't work reliably at the time of the original CVS import.)
I think that the simplest, most reliable, short-term change is to disable this particular optimization in fold, only for C++. That's not an ideal solution. If you can come up with a better approach that doesn't involve disabling the optimization, that's great -- but I think we need to convince ourselves that it handles all the cases. (I actually think that your MIN_EXPR-is-a-canonical-form-of-COND_EXPR argument can be made to work -- but you need to push the build_modify_expr logic into more places.) In the long term, moving fold until after the front end has done its thing, is the right way to go.
-- Mark Mitchell CodeSourcery, LLC [EMAIL PROTECTED] (916) 791-8304
