gh-102778: IDLE - make sys.last_exc available in Shell after traceback

[iritkatriel]

Fixes #102778.

• Issue: Add sys.last_exc, deprecate sys.last_type, sys.last_value and sys.last_traceback #102778

[iritkatriel]

Is this documented anywhere in idlelib docs?

[terryjreedy]

Yes and no. The 'yes' part is that "Running user code" says that IDLE generally tries to run user code so that it gives the same output (objects) one would get if one runs the same code in python itself, in either batch or interactive mode. Some known differences are then listed in this section and the next, "Displaying output in Shell".

For this issue, user-code sys.last_xyz expressions should have the same result as with python.exe in the same circumstances. In user code, sys.last_exc must continue raising an attribute error in 3.11 like it does in REPL. I considered whether to use last_exc 'internally' in 3.11, but decided to leave 3.11 alone. (And you never suggested touching it.)

In 3.12, last_exc must be set after the the first uncaught-by-user-code 'interactive' exception. Since IDLE intends to simulate running code from an editor with python -i, all user exceptions caught by IDLE are treated as interactive. Hence the two new lines in run.py that unconditionally set last_exc along with the deprecated but still supported attributes. For users, these 2 lines are the only new code relevant to 'Support sys.last_exc in IDLE'. They should have been in the last alpha and should be added, I think, now.

The immediate manual test is that IDLE imitate the REPL as below instead of raising AttributeError as it does now.

>>>import sys; 1/0
Traceback ...
>>> sys.last_exc
ZeroDivisionError('division by zero')

Fixing test_run to check this may or may not be simple.

This change should get a line in What's New 3.12. Check the IDLE entry in What's New 3.11. This could be done later.

The no part is that the rest of the patch, including that to pyshell.py, only concerns stackviewer. This is internal code of no direct concern to users. It need not be done now, or even in 3.12, as it should continue to work. When revised, the support for 2.x string exceptions and the use of the old attributes (which were for support of string exceptions) should be removed. The only effect on users would be that the way to disable stackviewer will change from del sys.last_traceback (undocumented except the the error message ;-) to del sys.last_exc. So revert the current stackview changes and draft a separate PR or wait for me to do so in maybe a week.

[terryjreedy]

Reviewing further, I think that the second sys.last_exc = value may not be needed for the user code issue. The test would be to make only the first addition and manually test. The second may be redundant with the first or it may only be used for internal communication. The exception ultimately comes from sys.exc_info (or from a test) and IDLE does not have to use sys as a cache. It should only set sys.last_xyz as needed for non-IDLE code.

In any case, I would like to be able to pass the exception object to stackviewer and its tree item so that the module has no need of sys. Such a change could be attached to #102832 and backported to 3.11. #68978 has a list of possible improvements for stackviewer and if I get familiar with the code again I might want to do some and backport to 3.11.

Closing was an accident. Time to sleep.

[terryjreedy]

I verified that the setting of sys.last_exc in print_exception should be sufficient.