GeezerVison

Verification screenshots

Built CPython from source and ran the REPL to verify before/after behavior.

Before (Python 3.13, no cross-language hints)

After (this PR)

All 447 tests in test_traceback pass (including 28 new cross-language tests).

I will make a review of this PR when I have time (by the end of the week), so fellow core devs, please hold off any merge, TiA!

Thanks for taking this on! From a language perspective, I think these can be shorter. Instead of:

AttributeError: 'list' object has no attribute 'push'. Did you mean '.append' instead of '.push'?

I think it is enough to say:

AttributeError: 'list' object has no attribute 'push'. Did you mean '.append' instead?

or even:

AttributeError: 'list' object has no attribute 'push'. Did you mean '.append'?

Good call - the error message already says "has no attribute 'push'" so repeating it in the hint is noise. Shortened to Did you mean '.append'? in 6d58cdc.

I kept the existing Levenshtein format as-is (that's a separate code path and probably a separate discussion), so the cross-language hints now have their own slighty shorter style.

@ZeroIntensity re: suggesting language constructs - added both in 6d58cdc. They use the custom hint format since they suggest constructs rather than method equivalents:

>>> {}.put("a", 1)
AttributeError: 'dict' object has no attribute 'put'. Use dict[key] = value for item assignment

>>> [].contains(1)
AttributeError: 'list' object has no attribute 'contains'. Use 'x in list' to check membership

The existing architecture already handles this - entries with a space in the hint string are rendered as-is rather than wrapped in "Did you mean" format.

I will make a review of this PR when I have time (by the end of the week), so fellow core devs, please hold off any merge, TiA!

appreciat you digging in when you're ready! ::anxiously waits::

Oh, you also need to add an entry to "What's New in Python 3.15".

Added in 579d037 - covers the basic hint format and the language-construct variant (dict.put, list.contains).

I applied my suggestion to replace types.MappingProxyType with frozendict.

@ZeroIntensity @picnixz: Are you ok with this change?

Thanks @picnixz for the review and approval! Happy to wait for @serhiy-storchaka's feedback.

Is the nit you mentioned something I should address now, or was it from the earlier round (the Levenshtein point and the raw hint formatting)? Those should be fixed in the latest commits but let me know if I missed something.

The nit is just adding a newline before the "1.".

Added the blank line in 0089761. Waiting on @serhiy-storchaka whenever he has time.

See my comments on the issue.

First, it should work for subclasses. We should also fix suggestions for float.__or__ and other wrong suggestions. Add suggestions for tuple.append, frozenset.add, frozendict.update, etc. Maybe add some suggestions for None.

Thanks @serhiy-storchaka for the thorough feedback! Addressed all your points in e2c12ec:

Subclass support: Switched from type(obj) exact match to isinstance() check. OrderedDict().entries() now suggests .items(), custom list subclasses get hints too. Table restructured to index by method name for efficient isinstance iteration.

Mutable on immutable: Added tuple.append/extend/insert/remove (suggests list), frozenset.add/discard/remove/update (suggests set), frozendict.update (suggests dict).

None suggestions: Added hints for common methods tried on None: None.keys() suggests dict, None.upper() suggests str, None.sort() suggests list, etc.

Float bitwise: Added float.__or__/__and__/__xor__/__lshift__/__rshift__ suggesting int. Cross-language hints now take priority over Levenshtein (more specific), so the misleading __dir__ suggestion is replaced. Levenshtein still fires as fallback when no table match exists.

Demo (built from source, Python REPL):

All 433 tests in test_traceback pass.

Addressed all three in e1eb6f4:

1. type(None) → types.NoneType - Done.

2. frozendict wrapper restored - The frozendict() wrapper was lost during the isinstance() refactor in e2c12ec. Restored it, and converted inner lists to tuples for consistency with the immutable container.

3. None doctest - Good point, None.keys() was contrived. Replaced with your example:

>>> lst = [3, 1, 2].sort()
>>> lst.pop()

This shows the real scenario (method returned None instead of the expected object). Also updated the prose to explain why.

(Also fixed the CI failure from the previous push in 564ca3a - three cross-language tests were missing @force_not_colorized.)

Documentation build overview

📚 cpython-previews | 🛠️ Build #32533375 | 📁 Comparing 5ad1a90 against main (de66149)

  🔍 Preview build  

Merged, thanks for your contribution to Python.

@vstinner was a great experience. Thank you and team

Awesome to see! But shouldn't all the mutable methods be included? The first thing I tried was .clear(), which was missing from frozendict, frozenset, and tuple. I think it would be best to check all the missing methods and add all that make sense, it's a small list for each:

>>> set(dir(dict)) - set(dir(frozendict))
{'clear', 'pop', 'update', '__ior__', 'setdefault', 'popitem', '__setitem__', '__delitem__'}
>>> set(dir(list)) - set(dir(tuple))
{'__imul__', 'sort', '__reversed__', 'clear', 'pop', 'insert', 'copy', 'reverse', '__iadd__', 'extend', '__setitem__', 'append', '__delitem__', 'remove'}
>>> set(dir(set)) - set(dir(frozenset))
{'__isub__', '__iand__', 'clear', 'pop', 'update', '__ior__', 'difference_update', 'intersection_update', 'symmetric_difference_update', 'add', '__ixor__', 'discard', 'remove'}

I'm not sure that it's worth to add suggestions for all methods, but adding clear() sounds reasonable.

@mvanhorn: Do you want to propose a PR adding clear() suggestions?

@vstinner sure, happy to. I'll open a follow-up PR adding clear() to the cross-language hints table for tuple, frozenset, and frozendict (each suggesting the mutable counterpart, same shape as the existing append/discard/update entries). Aiming for this week. Thanks @henryiii for the catch.