gh-55646: Do not crash IDLE on an invalid key binding

[serhiy-storchaka]

A typo in a custom key binding entered via Advanced Key Binding (which is not validated), such as <Alt-Key-up> instead of <Alt-Key-Up>, crashed IDLE at startup:

  File ".../idlelib/multicall.py", line ..., in bind
    self.__binders[triplet[1]].bind(triplet, func)
  ...
_tkinter.TclError: bad event type or keysym "up"

MultiCall.event_add stores the parsed sequence without validating the keysym, so the TclError is raised later, when a virtual event is bound and MultiCall performs the actual Tk bind. With no window yet, IDLE just exits.

Catch TclError at both binder call sites (MultiCall.bind and MultiCall.event_add) and ignore the invalid binding with a warning naming it. The bind path also drops the bad sequence from the virtual event, which avoids a secondary ValueError if it is later rebound or unbound.

🤖 Generated with Claude Code

• Issue: IDLE crash on startup with typo in config-keys.cfg #55646

[terryjreedy]

The issue is an example of making the good the enemy of the adequate. Unexpected crashes/exits are bad and this is definately better than no fix.

The merge conflict was due to this being the 2nd patch to add tests at the end of test_multicall. The local interface had a button to accept both additions.

I also looked at Serwy's patch: https://bugs.python.org/file24351/issue11437.patch.
His concern about repeated processing is mostly moot for most users. We merged all the IDLE feature extensions into the main code about a decade ago, leaving only a test extension.

Main question: Why catch errors in multicall rather than configparser? Does every sequence get run through multicall? Does this catch more errors?

[serhiy-storchaka]

Note that "Accept both" eats a blank line between fragments. You have to add it after pressing "Accept both" while you are in the online editor. I fixed this.

Yes, every configured key binding goes through MultiCall. MultiCall is the single choke point where a binding string is actually handed to Tk.

Serwy's patch also works. But it pulls a Tk root plus a modal dialog into config code that also runs headless (tests, the subprocess). Naming the bad key and suggesting a reconfigure is a nice touch, which I've added to the warning.