gh-127555: Document that sys.tracebacklimit can be set to None

[srittau]

Add versionchanged marker

• Issue: Setting sys.tracebacklimit to None is not documented #127555

---

📚 Documentation preview 📚: https://cpython-previews--127556.org.readthedocs.build/

[ZeroIntensity]

I don't think this is the intended behavior. The source is here:

cpython/Python/traceback.c

Line 723 in 979bf24

if (limitv && PyLong_Check(limitv)) {

Setting to any type that isn't an integer will result in this happening. To me, seeing sys.tracebacklimit = None looks like "no limit," not "default limit." I'm not sure we should document that.

[srittau]

Considering that this behavior is explicitly called out in the NEWS entries for 3.6.4 and 3.7.0a3, it sounds deliberate to me.

[ZeroIntensity]

If we really do want None as a special case, we should emit a warning (or possibly raise an exception) for other types. cc @serhiy-storchaka, who wrote the code.

[serhiy-storchaka]

Please remember the issue wnd the changelog corresponding to that NEWS entry. I do no remember such issue.

I most likely did not invent that behavior, but simply fixed crashes or error handling in corner cases, or/and synchronized the C and the Python implementations. L9k also at the Python code.

[srittau]

For reference, here is the original issue: #76130 and the PR: #4289. The PR introduced an explicit test for None:

cpython/Lib/test/test_sys.py

Line 1219 in 8ba9f5b

check(None, traceback)

[serhiy-storchaka]

Thanks, @srittau. What does the corresponding Python code do? There should be a code using tracebacklimit in the traceback module.

[srittau]

The only instance is this:

cpython/Lib/traceback.py

Line 458 in 8ba9f5b

limit = getattr(sys, 'tracebacklimit', None)

So, None is supported, but that might be incidental. The following code basically skips over any frame limiting if limit is None, although I'm not sure what that means exactly, as I'm not particular familiar with the traceback module.

[rruuaanng]

The only instance is this:

cpython/Lib/traceback.py

Line 458 in 8ba9f5b

limit = getattr(sys, 'tracebacklimit', None)

So, None is supported, but that might be incidental. The following code basically skips over any frame limiting if limit is None, although I'm not sure what that means exactly, as I'm not particular familiar with the traceback module.

Can support None, and this is necessary, this's not a incidental. After the getattr(sys, 'tracebacklimit', None) statement is executed, it always returns None, because the tracebacklimit attribute does not exist and it never appears in sys module.

Running Release|x64 interpreter...
Python 3.14.0a0 (heads/master-dirty:ff4e499, Nov 16 2024, 20:59:33) [MSC v.1941 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.tracebacklimit
Traceback (most recent call last):
  File "<python-input-1>", line 1, in <module>
    sys.tracebacklimit
AttributeError: module 'sys' has no attribute 'tracebacklimit'

(It's fix appears in another PR of mine.)

[ZeroIntensity]

Yeah, I manually checked locally, and setting tracebacklimit to None does the same for the Python implementation (it still might be incidental, but it doesn't matter anyway). Anyways, intentional or not, this seems to be the proper behavior for both implementations of traceback, so this LGTM.

[iritkatriel]

Thanks, @srittau. What does the corresponding Python code do? There should be a code using tracebacklimit in the traceback module.

@serhiy-storchaka we no longer have C and Python implementations of the traceback formatting logic. It's all in Python now.

[iritkatriel]

Are there tests for setting it to None?

[serhiy-storchaka]

There is still C code in _PyTraceBack_Print().

There are tests for setting sys.tracebacklimit to None in test_sys. But the Python implementation in the traceback module is not tested for this.

And it seems to me that setting sys.tracebacklimit to None has different effect on C and Python implementations.

[ZeroIntensity]

I think it's vice-versa: we're testing the Python implementation, not the C-level _PyTraceBack_Print. Again though, they seem to do the same thing for None, at least if my test was extensive enough.