bpo-23689: re module, fix memory leak when a match is terminated by a signal or memory allocation failure #32283
Conversation
_CompileData can store intermediate data.
argument clinic
-_validate_inner(SRE_CODE *code, SRE_CODE *end, Py_ssize_t groups) +_validate_inner(SRE_CODE *code, SRE_CODE *end, PatternObject *self)
LGTM.
I have only one question: how to prove that we need only one SRE_REPEAT structure per the REPEAT code?
| self.assertEqual(get_debug_out(r'(?:ab)*(?:cd)*'), '''\ | ||
| MAX_REPEAT 0 MAXREPEAT | ||
| self.assertEqual(get_debug_out(r'(?:ab)*?(?:cd)*'), '''\ | ||
| MIN_REPEAT 0 MAXREPEAT |
There was a problem hiding this comment.
You just read my mind! I was going to propose such a change, but I thought that I was already bothering you too much.
There was a problem hiding this comment.
I just thought this after posting this PR.
I thought that I was already bothering you too much.
As an inactive contributor, this is not a matter.
I'm not practised, so need continuously improve the patch to get to a good state.
When I think it's good, I can always find its shortcomings afterwards.
I have only one question: how to prove that we need only one SRE_REPEAT structure per the REPEAT code?
I have to think about how to answer your question.
At any time, an
This wouldn't work if re engine could memorize some backtracking states to optimize performance, but then the code of re module would be much more complicated. |
@serhiy-storchaka
This time, I checked several rounds carefully, it should be in good state.
Tested with
VERBOSE/VVERBOSEmacros defined, it builds and runs well.https://bugs.python.org/issue23689