AddressSanitizer issue on multihreaded insert about parallel-hashmap HOT 7 CLOSED

Hi, just to let you know I finally found the issue. In documentation, you're using typename HT::hasher hasher and then hashval = hasher(key), whereas I found in bench.cc that you use hashval = hash.hash(key). When I changed to the latter, insert works fine.

from parallel-hashmap.

@chybz where did you see that reference to typename HT::hasher in the documentation?
But you are right the original version should work as well, sorry for the issue. I'll look into fixing this.

from parallel-hashmap.

Thanks @greg7mdp for your update.

It's there in the code block right after "and here is the code for the multi-threaded insert".

I've another question, though. Could you please add an insert(...) method when map subkey is already known, similar to what find(key, subkey) does ? My use case is that I have a hierarchical and parallel aggregation on many groups, merged in parallel and when I merge, I already computed map subkeys and would like to reuse them. Note that I cannot insert directly a map in another map because I must know what entries are new.

Thanks for any help.

from parallel-hashmap.

Sounds good @chybz I'll try to get this done this week-end. Thanks for pointing out the location of the code!

from parallel-hashmap.

Hi @chybz I tried adding the insert() with the hash value provided, but in my tests it doesn't make any speed difference, so I am not sure it is worth it.

from parallel-hashmap.

Hi @greg7mdp, thank you very much for the time you spent on the subject !

from parallel-hashmap.

Hi @chybz pas de problème, merci d'utiliser parallel_hashmap!

from parallel-hashmap.