Message10275

There's a minor difference between CPython 2.7's sre_compile.py and our patched version:

$ diff Lib/sre_compile.py lib-python/2.7/sre_compile.py
14c14
<
---
> import sre_parse
15a16
> from _sre import MAXREPEAT
27,32d27
< def set(seq):
<     s = {}
<     for elem in seq:
<         s[elem] = 1
<     return s
<
308,310d302
<     # problems with optimization in Jython, forget about it for now
<     return charset
<
353,355c345
<         # change this for Jython from 'I', since that will expand to
<         # long, and cause needless complexity (or so it seems)
<         code = 'i'
---
>         code = 'I'
510d499
<         import sre_parse

The latter change in the diff is not even used - optimization as a whole for Unicode is skipped. I don't see this as an issue.

Because re contains a cache of compile REs, it's possible there is a chance of reentrancy issues. However my simple stress test running 20 threads for 15 minutes didn't see this issue. We have the same re.py as in CPython 2.7, but there's a possibility that we need to update.

Maybe you can explicitly list REs that this problem arises on? Right now, it works for me.