simplify matchplus and matchstar

kokke / tiny-regex-c

Small portable regex in C

The Unlicense

1.24k stars 174 forks source link

simplify matchplus and matchstar #63

Open marler8997 opened 3 years ago

marler8997 commented 3 years ago

matchplus can be simplified by only modifying matchlength once the complete match is successful. This means it doesn't have to rewind matchlength as it iterates through each possible matchpattern. This also means it keeps matchlength unmodified if it doesn't return a match. Because of this last part, this also means that matchstar can leverage matchplus which reduces it to single line of code return matchplus(...) || matchpattern(..).

marler8997 commented 3 years ago

Very interestingly, this change has noticeably improved performance. Before this change, running time ./tests/test2 3 times in a row gave these times:

real    0m4.047s
real    0m5.525s
real    0m3.952s

After this change I'm consistently getting below 3.6:

real    0m3.577s
real    0m3.571s
real    0m3.587s

The run times are much more consistent and it's around a 25% improvement for that test. My guess is that this performance may be due to less code needed to be swapped in and out of the icache? Not sure, but interesting.