Repository navigation
SequenceMatcher.get_grouped_opcodes() changes the result of later get_opcodes() calls #161973
Description
Bug report
SequenceMatcher.get_opcodes() caches its result and returns the same list on every call. get_grouped_opcodes() takes that list and overwrites its first and last entries to trim the context, so after one call the matcher's opcodes no longer cover the whole input, and a later get_grouped_opcodes() with a larger n shows less context than it should.
This was mentioned in gh-54194 back in 2010, but that issue was closed after only the get_matching_blocks() part was fixed (gh-65834); the get_grouped_opcodes() part is still there.
import difflib
old = ["line %d\n" % i for i in range(10)]
new = old[:]
new[5] = "changed\n"
sm = difflib.SequenceMatcher(None, old, new)
for group in sm.get_grouped_opcodes(1):
pass # show the hunks with one line of context
print(sm.get_opcodes())Actual:
[('equal', 4, 5, 4, 5), ('replace', 5, 6, 5, 6), ('equal', 6, 7, 6, 7)]
Expected (what get_opcodes() returns before the get_grouped_opcodes() call):
[('equal', 0, 5, 0, 5), ('replace', 5, 6, 5, 6), ('equal', 6, 10, 6, 10)]
unified_diff() and context_diff() are not affected because they make a new matcher each time.
CPython versions tested on:
3.15, CPython main branch
Operating systems tested on:
Linux
Found while building a Lean 4 model of CPython (https://broyojo.com/python-formalization/). AI tools (Claude) helped investigate the findings and prepare this report. The examples were rerun on CPython 3.15.2 and main, and the report received technical review. Record: textlibs-difflib-grouped-opcodes-mutates-cache
Activity
brunnojob commentedon Nov 14, 2026
I confirmed the alias in current
Lib/difflib.py:get_opcodes()returnsself.opcodesdirectly, then theget_grouped_opcodes()generator bindscodesto that same list and replacescodes[0]/codes[-1]while trimming equal spans. Copying before those fixups (codes = self.get_opcodes().copy()) should isolate the mutation; the opcode tuples themselves are immutable.For regression coverage, consume
list(sm.get_grouped_opcodes(1)), then assertsm.get_opcodes()still equals the full original opcodes. On the same matcher, compareget_grouped_opcodes(3)with a fresh matcher so a later request for wider context cannot inherit the earlier trimming.tanveer-arch commentedon Nov 14, 2026
I'd like to take this on. The bug is clear from the report: get_opcodes() returns the cached self.opcodes list, and get_grouped_opcodes() mutates its first and last entries while trimming context, so a later get_opcodes() or a wider n sees the trimmed list instead of the full one. I'm working on a fix that copies the list before trimming, plus a regression test.
Broyojo commentedon Nov 14, 2026
Thanks for confirming! I had a fix with a regression test ready, so I've opened #162012.