triggering a transient menu while in eca-chat can leave emacs unavigable with transient bindings active in all windows

[stsquad]

I'm not sure if the bug is in eca (see editor-code-assistant/eca-emacs#114) or transient but it certainly an interaction between the two which triggers the failure mode. Usually I'm in eca-chat and I trigger magit or another transient action and try to quit but something fails and I'm unable to navigate emacs because the keybindings are still in effect. However C-g or C-q don't quite the transient. Usually at this point I have to kill and restart emacs. The messages log shows transient commands getting run while I try to quit:

Error while exiting transient: (error "Window #<window 43 on  *transient*> has not same side right as its parent #<window 42>") [2 times]
Error in pre-command-hook (transient--pre-command): (error "Window #<window 43 on  *transient*> has not same side right as its parent #<window 42>")
ECA :: Shutting down...
ECA :: process has exited (killed)
Back to top level
When done with this frame, type C-x 5 0
magit-blame--pre-blame-assert: Buffer isn’t visiting a file
magit-file-stage: Not visiting a file
magit-blame--pre-blame-assert: Buffer isn’t visiting a file
Symbol’s value as variable is void: \’overriding-terminal-local-map
When done with this frame, type C-x 5 0
transient-backward-button: Wrong type argument: window-valid-p, #<window 43>
transient-forward-button: Wrong type argument: window-valid-p, #<window 43>
C-c <up> is undefined
Invalid read syntax: ")"
magit-diff-buffer-file: Buffer isn’t visiting a file
Type q in help window to delete it.
Use this command? (no, yes, SPC (space bar) only once, ! use and enable all): 

Invalid choice.  Use this command? (no, yes, SPC (space bar) only once, ! use and enable all): 
Quit
Back to top level
When done with this frame, type C-x 5 0
Completion of local vars is disabled in *scratch* (untrusted content)
magit-blame--pre-blame-assert: Buffer isn’t visiting a file [4 times]
nil
Symbol’s function definition is void: helpful
Wrong number of arguments: help, 1
magit-blame--pre-blame-assert: Buffer isn’t visiting a file
Symbol’s function definition is void: counsel-describe-binding
Type q in help window to delete it.
magit-blame--pre-blame-assert: Buffer isn’t visiting a file
transient-forward-button: Wrong type argument: window-valid-p, #<window 43> [2 times]
Type q in help window to restore previous buffer.
magit-blame--pre-blame-assert: Buffer isn’t visiting a file [3 times]
transient-backward-button: Wrong type argument: window-valid-p, #<window 43>
transient-forward-button: Wrong type argument: window-valid-p, #<window 43>

transient-backward-button: Wrong type argument: window-valid-p, #<window 43>
[mu4e] Update process is already running
Invalid read syntax: ")"
[mu4e] Update process returned with non-zero exit code
a is undefined
magit-blame--pre-blame-assert: Buffer isn’t visiting a file
transient-forward-button: Wrong type argument: window-valid-p, #<window 43>
transient-backward-button: Wrong type argument: window-valid-p, #<window 43>
next-line: End of buffer [3 times]

I can issue elisp via emacsclient so I would appreciate any hints on how find out why transient failed and didn't quit cleanly.

[tarsius]

This is a long list of seemingly unrelated errors. Some of them may be a consequence of transient erroring and then entering an inconsistent state. Other errors however, such as invalid read syntax, for example, are likely completely unrelated. Since eca was presumingly in large parts written by a LLM, you would first look there for defects.

Again, its possible that there's something wrong in transient that needs fixing, but I don't see how I can help you with this from a distance. I recommend producing and studying a backtrace. M-x transient-toggle-debug may also help.

Also, does transient still enter an inconsistent state, if you use a version after ecb2005?

[stsquad]

To answer the last question first - yes. I track MELPA on my main machine and I've just hit the case again. It's certainly having eca enabled that gets me into the state but it always seems to be transient that gets stuck.

It seems executing:

emacsclient -a '' -e '(transient-quit-all)'

from my console isn't enough to unwedge things.

I'll try and get a better backtrace.

[stsquad]

OK this is a cleaner log with transient debug enabled:

ECA :: Downloading eca server from https://github.com/editor-code-assistant/eca/releases/download/0.122.0/eca-native-static-linux-amd64.zip to /home/alex/.emacs.d/eca/eca.zip...
ECA :: Downloaded eca, unzipping it...
(Shell command succeeded with no output)
ECA :: Installed eca successfully!
ECA :: Starting process '/home/alex/.emacs.d/eca/eca server'
ECA :: Started with workspaces: /home/alex/lsrc/qemu.git/
[mu4e] Update process is already running
[mu4e] Update process returned with non-zero exit code
my-set-mu4e-index-params: idle:237 lazy:t cleanup:nil
next-line: End of buffer
[mu4e] Update process returned with non-zero exit code [2 times]
my-set-mu4e-index-params: idle:1442 lazy:nil cleanup:t
[mu4e] Update process returned with non-zero exit code [9 times]
Mark set [2 times]
-- setup                  (cmd: hubi, event: "C-c c", exit: nil)
-- stack-zap              (cmd: hubi, event: "C-c c", exit: nil)
-- setup-transient        (cmd: hubi, event: "C-c c", exit: nil)
     push transient--transient-map
     push transient--redisplay-map
-- post-command           (cmd: hubi, event: "C-c c", exit: nil)
-- pre-command            (cmd: hubi-get-dir, event: "D", exit: nil)
-- premature-post-command (cmd: hubi-get-dir, event: "", exit: nil)
-- suspend-override       (cmd: hubi-get-dir, event: "", exit: nil)
     pop  transient--transient-map
     pop  transient--redisplay-map
-- unwind-command         (cmd: vertico-exit, event: "D", exit: nil)
-- resume-override        (cmd: vertico-exit, event: "D", exit: nil)
     push transient--transient-map
     push transient--redisplay-map
-- post-command           (cmd: vertico-exit, event: "D", exit: nil)
-- pre-command            (cmd: transient-update, event: "C-c", exit: nil)
     pop  transient--redisplay-map
-- post-command           (cmd: transient-update, event: "C-c", exit: nil)
     pop  transient--redisplay-map
     push transient--redisplay-map
-- pre-command            (cmd: hubi--do-compile, event: "C-c C-c", exit: nil)
-- stack-zap              (cmd: hubi--do-compile, event: "C-c C-c", exit: nil)
-- pre-exit               (cmd: hubi--do-compile, event: "C-c C-c", exit: t)
     pop  transient--transient-map
     pop  transient--redisplay-map
Saving file /home/alex/lsrc/qemu.git/.dir-locals-2.el...
Wrote /home/alex/lsrc/qemu.git/.dir-locals-2.el
(Saved .dir-locals-2.el)
-- post-command           (cmd: hubi--do-compile, event: "C-c C-c", exit: t)
-- post-exit              (cmd: hubi--do-compile, event: "C-c C-c", exit: t)
[mu4e] Found 1000 matching messages; 0 hidden; search: 550.9 ms (0.55 ms/msg); render: 160.8 ms (0.16 ms/msg)
Quit
(No changes need to be saved)
Compilation finished
Mark set [2 times]
-- setup                  (cmd: magit-commit, event: "c", exit: nil)
-- stack-zap              (cmd: magit-commit, event: "c", exit: nil)
-- setup-transient        (cmd: magit-commit, event: "c", exit: nil)
     push transient--transient-map
     push transient--redisplay-map
-- post-command           (cmd: magit-commit, event: "c", exit: nil)
-- pre-command            (cmd: magit-commit-create, event: "c", exit: nil)
-- stack-zap              (cmd: magit-commit-create, event: "c", exit: nil)
-- pre-exit               (cmd: magit-commit-create, event: "c", exit: t)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- emergency-exit         (cmd: magit-commit-create, event: "c", exit: t, from: pre-command)
-- pre-exit               (cmd: magit-commit-create, event: "c", exit: t)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
Error in pre-command-hook (transient--pre-command): (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- post-command           (cmd: magit-commit-create, event: "c", exit: t)
-- post-exit              (cmd: magit-commit-create, event: "c", exit: t)
Only whitespace and/or comments; message not saved
Diffing changes to be committed (C-g to abort diffing)
When done with a buffer, type C-c C-c
Type C-c C-c to finish, C-c C-k to cancel, and M-p and M-n to recover older messages
-- setup                  (cmd: magit-dispatch, event: "h", exit: nil)
     pop  transient--transient-map
     pop  transient--redisplay-map
   autoload magit-ediff-dwim
   autoload magit-gitignore
   autoload magit-subtree
   autoload magit-patch
-- setup-transient        (cmd: magit-dispatch, event: "h", exit: nil)
     push transient--transient-map
     push transient--redisplay-map
-- post-command           (cmd: magit-dispatch, event: "h", exit: nil)
-- pre-command            (cmd: magit-am, event: "w", exit: nil)
-- stack-push             (cmd: magit-am, event: "w", exit: nil)
-- pre-exit               (cmd: magit-am, event: "w", exit: replace)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- emergency-exit         (cmd: magit-am, event: "w", exit: replace, from: pre-command)
-- pre-exit               (cmd: magit-am, event: "w", exit: t)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
Error in pre-command-hook (transient--pre-command): (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- setup                  (cmd: magit-am, event: "w", exit: t)
     pop  transient--transient-map
     pop  transient--redisplay-map
-- setup-transient        (cmd: magit-am, event: "w", exit: t)
     push transient--transient-map
     push transient--redisplay-map
-- post-command           (cmd: magit-am, event: "w", exit: replace)
-- post-exit              (cmd: magit-am, event: "w", exit: replace)
-- pre-command            (cmd: transient-quit-one, event: "C-g", exit: nil)
-- pre-exit               (cmd: transient-quit-one, event: "C-g", exit: t)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- emergency-exit         (cmd: transient-quit-one, event: "C-g", exit: t, from: pre-command)
-- pre-exit               (cmd: transient-quit-one, event: "C-g", exit: t)
Error while exiting transient: (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
Error in pre-command-hook (transient--pre-command): (error "Window #<window 49 on  *transient*> has not same side right as its parent #<window 48>")
-- post-command           (cmd: transient-quit-one, event: "C-g", exit: t)
-- post-exit              (cmd: transient-quit-one, event: "C-g", exit: t)
Back to top level
When done with this frame, type C-x 5 0

[tarsius]

The errors come from window--sides-check-failed. It appears something about your window management configuration is broken. I recommend you remove all of that, and then re-enable it piece by piece, until you know which part is problematic.

From the above output I cannot tell where this started to divert from what you expected. I am guessing you successfully used hubi and after that you then intentionally invoked magit-commit, which then resulted in the window issue.

[stsquad]

Yes the hubi invocation was fine, the magit-commit got stuck and couldn't quit. I'll have a look at what windows--sides-check-failed is looking for and see what might break it.

[stsquad]

OK I shall take this issue back to editor-code-assistant/eca-emacs#114 as I see ECA jumps through a number of hoops to preserve window configuration.

[tarsius]

8b14203 should help make this issue less fatal. The transient keymaps should now be deactivated in the face of such a failure. The transient window and buffer may remain visible, but it should be possible to delete them manually.