Closed mrhaandi closed 2 months ago
I'm amazed that it's only two occurrences that take up that long!
FTR I have had firstorder on my radar for a while. It's very inefficient because it strongly normalizes all terms from the context. (It's a bit sad that these instances will disappear for benching purposes, btw.)
Would that be reason enough to keep them though? Could we keep them as a special CI that only runs for benchmarks?
It does not seem reasonable to me to not fix it so that the firstorder issue stays in the CI. @ppedrot It should be easy to make a specific test-case for that no?
Thanks @mrhaandi !
Note that coq/coq#19126 has probably made this PR much less usefull in practice.
Replaced slow
firstorder
calls by appropriateconstructor
,left
,right
calls. This improves performance from5m49.68s | 2644580 ko | PCUICSR.vo
to0m49.37s | 1443968 ko | PCUICSR.vo
, also improving overall project compilation speed.