Tk Source Code

Example code:

  package require Tk
  entry .e
  .e insert end "\u4e4e"

Running this on macOS Ventura with an X11 build using XQuartz, it triggers:

X Error of failed request:  BadValue (integer parameter out of range for operation)
  Major opcode of failed request:  45 (X_OpenFont)
  Value in failed request:  0x200011
  Serial number of failed request:  107
  Current serial number in output stream:  108

then execution stops.

The Tk source code does not call X_OpenFont. However I have traced this down to GetScreenFont(). The following call in tkUnixFont.c:2808 generates the message and aborts:

	fontStructPtr = XLoadQueryFont(display, buf);

The error message and execution stop happen before this call returns. The solution is therefore not about checking the fontStructPtr return value for NULL.

When the problem shows up, the value in buf is -apple-geneva-medium-r-normal--12-*-*-*-*-*-iso10646-1

This looks odd because this font definition seems to be provided by XListFonts(), i.e. the X11 server itself provided it.

Note that the example code above is a trimmed down version of test entry-3.24, which crashes on macOS Ventura when run in an X11 build with XQuartz.

The workaround appears to work quite well, I now merged this to all branches.

Closing.

Sounds like a [7f20f59927a212d2|good idea]

Another way of working around this XQuartz problem without putting more and more constraints on tests is [b128b1a3]. Next time I merge branch less_tests_constraints into core-8-6-branch this will get propagated.

I have observed that this issue triggers or doesn't trigger depending on the mac used to run the tests. It may trigger on the CI runner but not on a real mac, or vice-versa. This is not a stable observation however, I did't manage to dentify what exactly makes the problem disappear or re-appear.

Instead of flagging numerous tests with 'failsOnXQuarz' one could perhaps try to add more fonts in the mac-build.yml workflow file, at least until the problem is fixed in XQuartz itself.

For the record: there is a branch bug-36e379c01b dealing with this ticket.

It looks like this XQuartz problem popped up on GITHUB:

https://github.com/tcltk/tk/actions/runs/9739591285

Can we just disable - for now - those testcases which trigger the problem?

Hmmm... can't reproduce anymore. XQuartz 2.8.5 (xorg-server 21.1.6).

The XQuartz bug is still open.

Still happening with the latest XQuartz 2.8.4 (xorg-server 21.1.6) dated 2022-12-24.

Thanks for the pointer.

I can again reproduce this with 8.6 on macOS Ventura with --disable-aqua --disable-xft by running:

  package require Tk
  pack [entry .e]
  .e insert end 乎

The Xquartz about box says: XQuartz 2.8.3 (xorg-server 21.1.4).

This seems to be an issue in Xquartz, see https://github.com/XQuartz/XQuartz/issues/216

Hmm, this does not happen anymore after fossil updating Tcl and Tk to core-8-6-branches (which should not have any link to the issue).

XQuartz about box says: XQuartz 2.82. (xorg-server 1.20.14). Maybe this got upgraded since I reported the crash, I'm unsure.

What XQuartz distribution/release version was used, in case that is relevant?