Closed joankaradimov closed 5 months ago
@joankaradimov I guess this commit causes this problem. I'll take a look at it some time this weekend. Thanks for catching the serious problem!
In that case I think it might be this line. There are multiple std::string
constructors being called within a while loop. This line can easily be moved outside the loop. And then the additional std::string
allocations can be avoided altogether in the non-case-insensitive case.
This improves the situation a little bit, but it does not resolve it completely. The performance is still worse than 1.8.5.
A much better solution would be to get rid of the ignore_case_
checks in match
and tweak the dictionary in the trie...
std::map<std::string, Info, std::less<>> dic_;
... to use case-insensitive comparisons instead of std::less
when necessary.
@joankaradimov could you check performance with the latest peglib.h in the master branch? If you confirm that it fixes the performance regression, I'll bump the version to 1.8.8. Thanks!
@yhirose I can confirm the problem is fixed.
From ~10 seconds previously down to below 50 ms now.
It's now included in v1.8.8. Thanks for your fine contribution!
Thank you for the quick responses and fixes!
@yhirose - Thanks for this. Are you responsible for the vcpkg? That's still sitting at 1.8.6.
@joankaradimov no. I don't know who are working on vcpkg, or other package managers like conan.
I'm running this grammar: https://github.com/joankaradimov/Magnetar/blob/master/MagnetarCraft/src/iscript_parser.cpp#L71 ... with this input file: https://raw.githubusercontent.com/neivv/aice/master/test_scripts/vanilla.txt
On v1.8.5 I'm observing parse times of below 1 second. On 1.8.6 it's above 10 seconds.
Similarly -- on the web playgound I was running identical examples for under 150 ms the previous week. I'm assuming the version over there was updated these last couple of days, since the execution times are suddenly 1-2 orders of magnitude worse.