Bug report

Bug description:

re.Pattern.match() (and prefixmatch(), fullmatch(), scanner().match()) run the matching engine when pos > endpos, although the docs say "If endpos is less than pos, no match will be found". Zero-width patterns then return an empty match, and a MULTILINE $ reads the character at pos, which is one past the end of the buffer when pos == len(string). search() already returns None here.

import mmap, re

print(re.compile("").match("abc", 3, 1))    # <re.Match object; span=(3, 3), match=''>
print(re.compile("").search("abc", 3, 1))   # None

mm = mmap.mmap(-1, 2 * mmap.PAGESIZE)
mm.resize(mmap.PAGESIZE)                    # leaves the next page unmapped
re.compile(rb"$", re.M).match(mm, mmap.PAGESIZE, 0)   # Segmentation fault

Expected None from every call. str and bytes subjects are NUL terminated so the read stays in bounds for them; mmap and array.array subjects are not.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-160316: Make re.Pattern.match() return None when pos > endpos #160317
• [3.15] gh-160316: Make re.Pattern.match() return None when pos > endpos (GH-160317) #161043
• [3.14] gh-160316: Make re.Pattern.match() return None when pos > endpos (GH-160317) #161044