emacsclient --tty crashes immediately (regression)

[aaronjensen]

Description

I can no longer connect tty clients to a graphical emacs on osx. I bisected, and
this is the offending commit:

cf3068e

Because of its nature (it is simply re-enabling a hook that was disabled by an
upstream change), reverting the commit is probably not the right thing to do.

Reproduction guide

• Start Emacs
• Run emacsclient --tty something

Observed behaviour:
emacsclient starts and crashes immediately

This error appears in messages:

Error during redisplay: (eval (window-numbering-get-number-string)) signaled (wrong-type-argument hash-table-p nil) [2 times]

Expected behaviour:
emacsclient does not crash

System Info

• OS: darwin
• Emacs: 25.0.93.1
• Spacemacs: 0.105.19
• Spacemacs branch: nil (rev. cf3068e)
• Graphic display: t
• Distribution: spacemacs
• Editing style: hybrid
• Completion: helm
• Layers:

(better-defaults spacemacs-helm emacs-lisp markdown
                 (syntax-checking :variables syntax-checking-enable-tooltips nil)
                 (auto-completion :variables auto-completion-enable-sort-by-usage nil auto-completion-enable-snippets-in-popup t)
                 erlang elixir git dash html clojure
                 (org :variables org-enable-github-support t)
                 colors osx github javascript react deft floobits ruby
                 (shell :variables shell-default-shell 'ansi-term shell-default-height 30 shell-default-position 'bottom)
                 spell-checking ranger version-control rcirc tmux evil-little-word auto-correct frame-geometry cleverparens-lispy)

[aaronjensen]

If this line is commented out:

spacemacs/layers/+spacemacs/spacemacs-layouts/funcs.el

Line 510 in cf3068e

(eyebrowse--load-window-config current-slot))

then everything works fine. I don't understand eyebrowse/persp well enough to go much further.

@syl20bnr @TheBB could I propose that this particular scenario (emacsclient --tty xxxx connecting to a gui emacs) be part of acceptance criteria for pull requests? This has broken at least twice and usually remains broken for a long time. It's a shame to disable this handy scenario.

[TheBB]

Probably prudent to ping @bmag on this one.

[bmag]

@aaronjensen The bug also reproduces on my machine (Ubuntu 16.04). emacsclient starts and crashes immediately, leaving my terminal in a weird state, but I don't see any error message.

The Bug

I don't know why it happens, but I suspect it has something to do with persp-mode using the same nil-perspective for all(?) frames. Because persp-mode uses the same perspective for the terminal frame as for the graphical frame, we try to load a graphical window state (Eyebrowse workspaces are mainly a window state object) into a terminal frame, resulting in error. There is indeed an error if I manually load a graphical window state into a terminal frame, even in stock Emacs.

To sum it up: we can't use the same workspace on different frame types (graphical vs. terminal), but because the same perspectives are used in both frame types, and because we make workspaces perspective-local, the result is that we try to do exactly that. Specifically in @aaronjensen's case, the shared perspective is probably the default perspective.

Possible Solution 1

Make perspective local to a frame, IOW perspectives should not be shared between frames. I don't know if this is possible, and it will certainly make it harder for users who want to share perspectives between frames.

Possible Solution 2

Store two sets of workspaces as perspective parameters - one set for graphical frames, a second set for terminal frames. When loading workspace from perspective, load the correct one according to display-graphic-p. The persp params that we currently use:

(persp-parameter 'eyebrowse-window-configs))
(persp-parameter 'eyebrowse-current-slot))
(persp-parameter 'eyebrowse-last-slot))

will be replaced by:

;; for graphical frames
(persp-parameter 'gui-eyebrowse-window-configs))
(persp-parameter 'gui-eyebrowse-current-slot))
(persp-parameter 'gui-eyebrowse-last-slot))
;; for terminal frames
(persp-parameter 'term-eyebrowse-window-configs))
(persp-parameter 'term-eyebrowse-current-slot))
(persp-parameter 'term-eyebrowse-last-slot))

I think solution 2 is simpler and easier to implement/maintain than the other possible solutions.

Possible Solution 3

Same as solution 2, but instead of using just two sets, we use a distinct set for each frame (graphical or not).

[bmag]

@aaronjensen can you test #6031? Terminal frames and graphical frames should have totally separate workspaces (but not separate layouts).

[aaronjensen]

@bmag that works great, thank you!

[d12frosted]

Fixed with released of Spacemacs v0.200. Let us know if you still have any problems with this issue.