[engine.i] even faster gnc_get_match_commodity_splits

[christopherlam]

gnc_get_match_commodity_splits needs to scan the account splitlists to find suitable splits upto end_date. We can stop scanning each account at end_date because the splitlists are sorted by date. This saves significant time because majority of splits can by bypassed in the early report dates.

My primary book net worth linechart benchmarks -- fastest speeds from 1.05s to 0.54s, now practically unnoticeable

[christopherlam]

no idea why it won't compile on github... all passes well on my pop_os end.

[jralls]

no idea why it won't compile on github... all passes well on my pop_os end.

It's just macOS (so far anyway). That's the problem I posted about on IRC.

[jralls]

What's going wrong with std::lower_bound? It's not really the distro, of course, it's some version of libstdc++ that has a problem. Does it work the way you expect in some other version?

[christopherlam]

What's going wrong with std::lower_bound? It's not really the distro, of course, it's some version of libstdc++ that has a problem. Does it work the way you expect in some other version?

Beats me... Try reverting the fix-up commit and push: The Ubuntu 22.04 ci shows lots of test errors proving that the lower_bound approach is giving incorrect results!

[christopherlam]

https://github.com/Gnucash/gnucash/actions/runs/8865090288/job/24340972868

[jralls]

he Ubuntu 22.04 ci shows lots of test errors proving that the lower_bound approach is giving incorrect results!

You got the comparison backwards in is_after_date. std::lower_bound returns the first iterator where the passed-in function returns false so the comparison needs to be txn_date <= end_date. Alternatively change to std::upper_bound and reverse the parameters in is_after_date. I think the latter better expresses the intent because you're evaluating [begin(), is_after_date_iter); lower_bound would make sense for [is_after_date_iter, end()).

[christopherlam]

he Ubuntu 22.04 ci shows lots of test errors proving that the lower_bound approach is giving incorrect results!

You got the comparison backwards in is_after_date. std::lower_bound returns the first iterator where the passed-in function returns false so the comparison needs to be txn_date <= end_date. Alternatively change to std::upper_bound and reverse the parameters in is_after_date. I think the latter better expresses the intent because you're evaluating [begin(), is_after_date_iter); lower_bound would make sense for [is_after_date_iter, end()).

Ok you're right :-)

[christopherlam]

The next question is why did arch ci pass the tests then...