I've been adding graal support to a classifier type project naively based on applying a bunch of regexes to an input, and while Graal works the regex application is quite slow: it's about 4x slower than cpython, while using 4 times the CPU.

Here's a repro script and attending data (basically a cut down version of the naive classifier implementation): script.zip

timings:

> python3.12 --version
Python 3.12.6
> time python3.12 run.py
75158 lines in 11.7s
	156.1 us/line
python3.12 run.py  11.77s user 0.05s system 97% cpu 12.172 total
> graalpy --version
GraalPy 3.11.7 (Oracle GraalVM Native 24.1.0)
> time graalpy run.py
75158 lines in 48.2s
	640.7 us/line
graalpy run.py  192.00s user 1.00s system 394% cpu 48.976 total

This is on a 10-core M1 Pro. Using cpusampler I confirmed that essentially all the "user" time is in _sre:

 _search   ||  43580ms  98.6% ||   43580ms  98.6% || <frozen graalpy._sre>~1:0