When compiling a regular expression with groups (subpatterns),
circular references are created.
Here is an example to illustrate the problem:

>>> import gc
>>> import re
>>> gc.disable() # disable garbage collector
>>> gc.collect() # make sure we start with 0
0
>>> re.compile('(a|b)') # compile something with groups
re.compile('(a|b)')
>>> gc.collect() # collects x objects depending on the compiled string
11

To fix the issue a weakref object for p is used.

Thank you for your report and patch Jeroen.

Indeed, there is a regression, and your patch fixes it. But I don't like the idea of using weakref. For now sre_parse has very little dependencies, but weakref depends on collections that depends on a number of modules. For now importing weakref works, but it is too easy to create a dependency loop in future.

Here is alternative patch that gets rid of references at all. The subpatterns list was added in the patch for bpo-9179 and is an implementation detail. We can replace it with a list of subpattern widths.

Thanks Serhiy,

I totally agree with your solution. Using a list with subpattern widths is definitely better compared to using weakref.

New changeset 7f4fca8f13a2 by Serhiy Storchaka in branch '3.5':
Issue bpo-25554: Got rid of circular references in regular expression parsing.
https://hg.python.org/cpython/rev/7f4fca8f13a2

New changeset 8621727dd9f7 by Serhiy Storchaka in branch 'default':
Issue bpo-25554: Got rid of circular references in regular expression parsing.
https://hg.python.org/cpython/rev/8621727dd9f7