> > Are we sure it ever stabilizes? But yes, I had something like this in mind > (just do one iteration always) in case we need to improve hashing.
Not necesarily - imagine two identical type variants, no matter how hard we hash&rehash, they will be same as they are identical. We can just keep iterating until number of equivalence classes is dropping. If we won't get all hashes unique, we can debug output and start chasing those duplicated trees away. (not sure how much effort it is - I did some experiments on type variants and indeed we do manage to produce identical variants from time to time) > > Patch is OK if you keep the sorting part of the comment, which is the > important part if all this - hash values have to be independent of SCC entry > chosen. Ah yep, will put it back. Thanks! Honza
