| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
Three audit defects in one file. cj/open-with-info-mode used cl-return-from inside a plain defun, so declining the save prompt threw "No catch for tag" instead of cancelling. The decision is now a pure cj/--info-open-plan and the command routes through it. A dead :hook (info-mode . info-persist-history-mode, which names a non-existent mode on the wrong hook) and an empty :preface are gone. The auto-mode-alist entry that mapped .info to that interactive, buffer-killing command is dropped, so find-file-noselect of a .info no longer destroys buffers. cj/open-with-info-mode stays an M-x command and C-h i still browses info files.
|
| |
|
|
| |
Three audit fixes. Single-file zip named the archive after the input file (<<fne>>.<<e>>), making invalid archives and a "foo." name for directories. It now builds <fne>.zip. The dated backup single-quoted $(date ...), so the stamp sat literal in the filename. Now the timestamp is interpolated in Elisp with format-time-string. The dired menu used "M-S-d", which Meta-Shift-d never emits (it sends M-D), so the menu was unreachable in plain dired. It now binds M-D, matching the dirvish sibling. Both command strings moved to top-level builders so they're unit-testable without loading the dwim-shell-command package.
|
| |
|
|
| |
F2 ran markdown-mode's own markdown-preview instead of the custom one, so the impatient-mode strapdown preview was dead. I renamed the custom command to cj/markdown-preview and rebound F2 so markdown-mode no longer shadows it. The server guard now uses httpd-running-p (httpd-process isn't a variable), and a dead (setq imp-set-user-filter 'markdown-html) that named neither a real variable nor a real filter is gone.
|
| |
|
|
| |
C-; b d now runs cj/diff-buffer-with-file (the op I hit most, comparing a buffer against the saved file) and C-; b D runs cj/delete-buffer-and-file. The destructive command sat on the easy lowercase key and diff on the capital. A keymap-lookup test guards the swap.
|
| |
|
|
|
|
|
|
|
| |
Two bugs from the 2026-06 config audit.
- C-s C-s never repeated a search. cj/consult-line-or-repeat calls vertico-repeat on the second press, but vertico-repeat-save was never on minibuffer-setup-hook, so it always signalled "No Vertico session". I hooked vertico-repeat-save next to the vertico block. It's autoloaded, so the load defers to the first minibuffer.
- reconcile-open-repos skipped any repo with a dot in its name. cj/find-git-repos filtered children through "^[^.]+$", which matches only dot-free names, so mcp.el, capture.el, and google-contacts.el were never reconciled while M-P still reported "Complete". It now filters through directory-files-no-dot-files-regexp plus a hidden-dir guard, so dotted repos pass and .git stays out.
Each fix is test-first: a failing assertion (hook membership, three dotted repos found) precedes the change.
|
| |
|
|
|
|
|
|
|
|
| |
Lock screen: slock is X11-only and never grabbed the Wayland session, so C-; ! l silently did nothing. On Wayland the locker now runs loginctl lock-session, which logind turns into a Lock signal that hypridle handles by running hyprlock, the same path idle and before-sleep locking already use. X11 keeps slock. system-commands.el now also requires host-environment, which it used at load time but never declared.
Confirmation tier: the global (fset 'yes-or-no-p 'y-or-n-p) plus use-short-answers t both flattened yes-or-no-p to a single keystroke, so the deliberate strong-confirm tier for irreversible actions was dead. A stray space could power off the machine or destroy files. I added cj/confirm-strong, which binds use-short-answers nil for one call to force a typed "yes", and routed the six irreversible sites through it (shutdown/reboot, permanent file destruction, file overwrites). I dropped the redundant fset and kept use-short-answers t so ordinary prompts stay single-key.
Mail folders: the cmail context set no trash folder, so D fell back to a nonexistent /trash, and no context set a refile folder, so r targeted a nonexistent /archive everywhere. Accepting mu4e's offer to create the maildir stranded mail where mbsync never syncs it. cmail now trashes to /cmail/Trash. Refile is computed per message rather than per context, because mu4e context :vars are sticky and a per-context refile would leak one account's archive folder into another. cmail archives to /cmail/Archive. The Gmail-backed accounts have no synced archive maildir, so they signal rather than move mail into an unsynced folder.
Lock and confirm-tier need a daemon restart to fully take effect. The mail changes apply on next mu4e open.
|
| |
|
|
|
|
|
|
| |
Most of the yasnippet keys start with "<" (<cj, <for, <main, ...), mirroring org-tempo. electric-pair-mode pairs "<" into "<>" wherever the mode's syntax table gives "<" paren syntax, which org and the pairing-enabled prog modes both do. So typing "<cj" lands as "<cj>", and expanding the "<cj" key strands the ">" after the snippet. The cj-comment block came out with a "#+end_src>" close fence, which breaks the cj-scan fence parser and made every respond-to-cj-comments pass hand-parse around it.
I set electric-pair-inhibit-predicate globally to inhibit the open angle bracket and defer to the default for every other character. That keeps the "<"-prefixed snippet convention intact across all 14 of them, at the cost of "<>" auto-pairing where it might otherwise be wanted (C++ templates). The snippet convention is universal, so it wins.
Surfaced from the smoke project, which kept hitting the malformed fence in its todo.org.
|
| |
|
|
|
|
|
|
| |
4a1ecf64 rebound eww/elfeed/calibredb from M-S-e/r/b to the uppercase Meta events M-E/M-R/M-B, on the theory that the keyboard emits the uppercase event and the lowercase-shift spec never matched. It was wrong. keyboard-compat.el installs a key-translation-map entry (M-E -> M-S-e, etc.) in GUI frames, so the original M-S- bindings did reach the launchers. Binding M-E directly while that translation stayed in place rewrote the keypress to the now-unbound M-S-e, breaking all three in GUI on the next restart.
The audit and the review missed it because they checked key-binding, which ignores key-translation-map, and the running daemon still held the pre-fix bindings as stale state.
This restores M-S-e/r/b, deletes the key-binding-only test that certified the broken config, and reclassifies the task as not-a-bug. The real problem, that these chords are dead outside GUI, is the subject of the keybinding-console-safety spec.
|
| |
|
|
|
|
|
|
|
|
| |
The Hyprland Super+Shift+N popup opens a floating emacsclient frame named "org-capture" and runs a capture in it. Three things were wrong. The *Org Select* menu and the CAPTURE buffer split the small frame instead of filling it, so the daemon's last buffer leaked in beside two tmux-like modelines. The menu offered every template, including ones that can't work off the desktop: the pdf templates error outright, and the link and mu4e templates pull a link to whatever file the daemon last had open. And the frame only closed on a completed capture, so aborting at the menu or hitting an erroring template orphaned it.
I added cj/quick-capture as the launcher's entry point instead of org-capture. It offers only Task, Bug, and Event, with Task and Bug retargeted to the global inbox since a desktop capture has no real project context, and it closes the popup frame on every exit path: finalize, abort, or error. A frame-scoped display-buffer-alist entry forces the menu and capture buffer into the frame's sole window, and an org-mks advice drops the Customize entry. Both gate on the "org-capture" frame name, so in-Emacs captures keep their windows.
cj/quick-capture selects the "org-capture" frame by name before capturing rather than trusting the selected frame. The launcher runs before Hyprland settles focus on the new float, so the selected frame is still the main one and the capture would otherwise land there.
Raised from the archsetup project.
|
| |
|
|
| |
An out-of-range numeric prefix made (nth (1- arg) ...) return nil, and find-file on nil signaled a wrong-type-argument. Guard the index and raise a user-error naming how many killed files are available. A new test covers the out-of-range path.
|
| |
|
|
| |
The prior fix listed drill files with a raw directory-files call, bypassing cj/--drill-files-or-error, the shared validated entry point the other drill commands use. That skipped the missing/unreadable-dir user-error, fell through silently on an empty dir, and included leading-dot .org files the module otherwise excludes. Route through cj/--drill-files-or-error + expand-file-name, keeping the let binding so the session-wide org-refile-targets still survives. The test is rewritten into three: validated-helper targets, no global clobber, and a user-error on a missing drill dir.
|
| |
|
|
| |
dashboard-config used setq on recentf-exclude, discarding the five exclusions system-defaults adds earlier in init order (bookmarks, elpa, recentf, ElfeedDB, airootfs). Extract the EMMS exclusion into cj/--dashboard-exclude-emms-from-recentf (the :config side-effect was not reachable for a test) and use add-to-list so prior entries survive. Two ERT tests cover preservation and the added pattern.
|
| |
|
|
| |
cj/drill-refile used setq, permanently replacing the session-wide org-refile-targets so every refile anywhere offered only drill targets until restart; and its (drill-dir :maxlevel . 1) entry named a bound variable, which org reads as a directory string rather than a file list, so the drill side yielded nothing. Let-bind org-refile-targets and supply (directory-files drill-dir t "\\.org$") as the file list. The stale test (which asserted the buggy drill-dir spec) is rewritten into two: file-list targets and no global clobber.
|
| |
|
|
| |
Meta+Shift+<letter> emits the uppercase event (M-E/M-R/M-B), so the M-S-e/M-S-r/M-S-b :bind specs on lowercase letters were never reached by the keychord and the three launchers were dead. Rebind them to M-E/M-R/M-B. Three ERT tests assert each chord resolves to eww, cj/elfeed-open, and calibredb respectively.
|
| |
|
|
| |
The "d" dailies head ran FILETAGS and TITLE together with no newline, so every C-c n d daily was malformed: Org never parsed #+TITLE and the FILETAGS value swallowed the rest of the line. Extracted the head into the cj/--org-roam-dailies-head defconst so it is unit-testable (the value was unreachable inside the use-package :custom form) and gave it real newlines. Two ERT tests assert the FILETAGS/TITLE line separation and the trailing newline.
|
| |
|
|
| |
(interactive "p") forces arg >= 1, so the old (if arg (nth arg ...) (car ...)) always took the nth branch and plain M-S-z re-opened the second-most-recently-killed file. Index with (nth (1- arg) ...) so a numeric prefix is 1-based and no prefix re-opens the most recent. The two undo-kill tests now exercise the real no-prefix path (arg=1 reopens the first) and a 1-based numeric prefix; both failed against the bug and pass after. Filed a follow-up for a separate delq-vs-delete skip-visited defect found in the same function.
|
| |
|
|
| |
Incoming messages now notify through cj/signel--notify, installed as the fork's signel-notify-function. It suppresses the toast while that chat is in the selected window of a focused frame, collapses and truncates the body to 120 characters, and sends through the notify script (info type, --silent unless cj/signel-notify-sound is set). Without the script on PATH it falls back to notifications-notify and warns at load. The decisions are in the Notification slice addendum of docs/design/signal-client.org.
|
| |
|
|
|
|
|
|
| |
The mode-line box in the preview was hardcoded — it showed a box the generated theme couldn't actually produce, since build-theme.el never emitted :box. Made :box a real face attribute instead: a per-face box object (style line/raised/pressed, width, color) stored on UI and package faces, set from a "box" dropdown in both tables, rendered from the attribute everywhere (the mode-line bars, the package previews via ofs, the UI table preview cells), and exported through build-theme.el's --attrs as a proper :box plist (released/pressed → :style *-button; line → :line-width + optional :color). The hardcoded box is gone; mode-line and mode-line-inactive now default to the released-button box that is the Emacs default, so the preview and the export agree.
This also gives the package faces that genuinely use :box a way to represent it — the face audit found several (magit-branch-current/-remote-head, two flycheck list faces, the telega button family, ~15 slack button/dialog faces).
Tests: build-theme gains box-conversion + ui-box-emit ERT tests (24/24); the app-core deep-equal tests account for the new box slot; all 9 browser gates, 20 python, and 55 node tests stay green.
|
| |
|
|
| |
F8 is bound globally to cj/main-agenda-display, but ghostel's semi-char mode forwarded it to the terminal program, so it did nothing inside a ghostel buffer. I added F8 to ghostel-keymap-exceptions and rebuilt the semi-char map, which lets it fall through to the global map, the path F10 and the F9 family already take.
|
| |
|
|
| |
The tool authors themes from scratch -- palette, faces across every tier, live preview, export to a loadable deftheme. It never selects among existing themes, so "selector" mis-described it. Renamed the directory, the generated HTML and its title, the design spec, and every reference in the code, README, tests, and todo. No behavior change.
|
| |
|
|
| |
build-theme/--attrs takes underline and strike flags and emits :underline t and :strike-through t in canonical order (after slant, before height). The UI and package spec builders read the two new fields off each face object; syntax and default faces pass nil since they never carry them. Two new ERT tests plus updated ordering cases; an end-to-end convert confirms a shr-link face round-trips to :underline t and shr-strike-through to :strike-through t. 22/22 green.
|
| |
|
|
|
|
|
|
|
|
| |
build-theme.el is the last link in the theme-selector pipeline: a theme.json export becomes a single self-contained themes/<name>-theme.el. All four tiers convert: default from assignments.bg/.p, the syntax categories to their font-lock/tree-sitter faces with the bold/italic sets applied, UI passthrough, and package faces with :inherit/:height/weight/slant.
The output is a flat generated deftheme, not the palette/faces/theme trio the hand-authored dupre ships. A theme.json carries resolved per-face hex, not dupre's semantic-mapping layer, so a flat deftheme is the faithful output and never clobbers the curated dupre files.
I omitted the dec (decorator) key: Emacs has no dedicated decorator face and renders decorators with font-lock-type-face, which the type key already owns, so coloring dec independently would clobber types. Decorators follow the type color, as they do in stock Emacs.
20 ERT tests cover the attribute builder, each tier, the dec omission, and an end-to-end convert-and-load with a WCAG-AA assertion on the round-tripped default.
|
| |
|
|
|
|
| |
diff-refine-changed rendered the default foreground (#f0fef0) on the bright-gold yellow-1 (#ffd700) at WCAG contrast 1.35, unreadable wherever the face is a plain background, not inside diff-mode's own foreground overlay. diff-changed was weak too at 2.78.
Both now use the default foreground on the dark amber yellow-2 (#875f00), already in dupre's palette, for a contrast of 5.49. diff-refine-changed keeps its bold weight so it stays stronger than diff-changed. A WCAG-contrast test guards both faces from dropping below 4.5.
|
| |
|
|
|
|
| |
The pearl package moved from a single Linear workspace to multi-account, so the config follows. pearl-accounts now declares two workspaces, one for work and one personal, each resolving its own API key from authinfo and rendering to its own Org file. pearl-default-account picks which one opens, and pearl-switch-account swaps at runtime.
This replaces the old single-account setup (pearl-org-file-path plus one pearl-api-key lookup). The module file moves from linear-config.el to pearl-config.el to match the package name. init.el, the module-headers allowlist, and the module-inventory row follow the rename.
|
| |
|
|
| |
M-F9 close deleted the agent's window, collapsing the layout. Close now swaps that window to the working buffer and kills the agent buffer, so the split stays put. F9 hide still collapses the split. Close no longer does.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Three things in the calibredb search buffer.
A curated transient on ? exposes just my frequent workflows (switch library, filter by format or author, sort by author/title/pubdate/format, open, describe) instead of leaving me to remember calibredb's top-level single keys. calibredb's own full dispatch moves to H. which-key can't help here: these are mode-map single keys, not a prefix. So a curated menu on ? is the discoverable entry point. The transient is defined in :config rather than top-level: its macro expands against the elpa transient, but a batch Emacs loads the older built-in transient and breaks the load.
The book-detail view (d, or v) now docks to the bottom 30% and q dismisses it. calibredb-show-entry-switch goes to pop-to-buffer so a display-buffer-alist rule applies. The default switch-to-buffer-other-window ignores it. It's the same bottom-dock pattern as the signal chat buffer.
Sorting kept dropping the active filter. calibredb's macro-generated sort commands all end in calibredb-search-refresh-and-clear-filter, which nulls every filter flag. I override each to refresh via calibredb-search-refresh-or-resume, which re-applies the filter, so a sort over a filtered list keeps it. I used named advice, so reloading the module doesn't stack it.
Tests cover the describe command and the sort helper. The transient, bindings, dock, and advice wiring need the elpa transient and a live calibredb, so they're verified in the running daemon.
|
| |
|
|
|
|
|
|
|
|
| |
In a nov buffer m is bound to bookmark-set. nov's nov-bookmark-make-record names the record after the buffer name, the raw EPUB filename. So bookmarks read like "Frege_ A Guide for the Perplexed - Edward Kanterian.epub".
I advise nov-bookmark-make-record to rebuild the name from the record's filename: split on the last " - " into title and author per Calibre's "<Title> - <Author>.epub" naming, restore the colon Calibre sanitized to "_ ", and reorder to "Author, Title". That book becomes "Edward Kanterian, Frege: A Guide for the Perplexed".
I pulled the name from the filename rather than the embedded EPUB metadata on purpose. In real books the embedded metadata is the worse copy (truncated titles, authors in "Last, First" sort form, lost punctuation), while the filenames come from Calibre's curated database. A separate task tracks embedding the good metadata back into the files.
cj/--nov-clean-title and cj/--nov-bookmark-name-from-file are pure and carry the logic. The advice is a thin wrapper. 10 ERT tests cover colon restoration, the last-separator split, and the no-separator fallback.
|
| |
|
|
|
|
|
|
| |
C-c c t (Task) and a new C-c c b (Bug) now file into the current projectile project's todo.org under its "... Open Work" heading instead of always landing in the global inbox. Bug stamps the entry [#C]. Task stays a plain TODO.
Both share one function target, cj/--org-capture-project-location. It matches an existing top-level "... Open Work" heading when there is one, so a directory like .emacs.d resolves to the existing "Emacs Open Work" rather than a name derived from the basename. It only creates "<Project> Open Work" when the file has none. Outside a project, or in a project whose root has no todo.org, it falls back to the global inbox under "Inbox". In the no-todo.org case it also warns, naming the project. It never creates a project's todo.org.
I split the logic into pure helpers (project name, target decision, find-or-create heading) so they test directly, with the impure buffer-positioning left thin. 15 ERT tests cover the helpers and the wiring. I confirmed a real capture lands the entry under Open Work at the right level in the running daemon.
|
| |
|
|
|
|
|
|
| |
Entering copy-mode from C-; x c left the cursor at the live column, far right after a prompt, so scrolling up ran the cursor up the right edge instead of the left.
In the tmux branch I append C-a after C-b [, which tmux's emacs copy-mode reads as start-of-line. It has to go after C-b [: before copy-mode is active, C-a would hit the shell's readline instead. In the non-tmux ghostel-copy-mode branch I call beginning-of-line after entering the view, for the same column-0 result. Both branches now run the cursor up the left edge.
The non-tmux test asserts ghostel-copy-mode runs before beginning-of-line, since the move only repositions inside an active copy view. Its tracker variable is dwim-order, not calls, to avoid clashing with the variable the tmux mock macro binds.
|
| |
|
|
|
|
|
|
| |
The C-; M Signal prefix didn't take effect on a fresh Emacs launch. signal-config.el was the only feature module that bound into cj/custom-keymap directly, wrapped in (with-eval-after-load 'keybindings (when (boundp 'cj/custom-keymap) ...)). The boundp guard turned a load-order miss into a silent no-op, so the binding never landed at startup. A later live-reload always papered over it because keybindings was loaded by then.
I switched to the documented cj/register-prefix-map helper and added (require 'keybindings) at the top, matching every other prefix map. The require guarantees keybindings loads before registration, so the guard is gone.
I verified at a full emacs --batch init.el launch, the actual failing scenario, that C-; M resolves to the signel prefix. I added a contract test asserting the registration, since the boundp guard was robust under unit timings and only failed at full launch.
|
| |
|
|
| |
Inside a ghostel terminal or agent buffer, semi-char mode forwarded F10 and C-F10 to the pty, so the music-playlist toggle and the server-shutdown command never ran. Both are global bindings with no ghostel-mode-map entry, so I added them to ghostel-keymap-exceptions and rebuilt the semi-char map. The lookup then falls through to the global map. Same shape as the earlier F9, F12, and window-nav fixes.
|
| |
|
|
|
|
|
|
|
|
| |
Two keystrokes weren't reaching Emacs inside a ghostel terminal, both because of how ghostel routes keys in semi-char mode.
C-SPC was the worse one. ghostel forwards the `C-@' event but not the distinct `C-SPC' event GUI Emacs produces, so C-Space fell through to the global `set-mark-command' and set an Emacs region in the terminal buffer. That region followed point as output streamed (a stuck "selection" Escape and C-g couldn't clear), and it meant tmux copy-mode's begin-selection never started, so M-w copied nothing. I bind C-SPC to cj/term-send-C-SPC, which forwards NUL like a terminal key.
The C-M-arrows (buffer-move, window swap) were being forwarded to the terminal program the same way the F9 family was. I added the windmove S-arrows and buffer-move C-M-arrows to ghostel-keymap-exceptions and rebuilt the semi-char map. The S-arrows already reached Emacs by keymap precedence, but listing them makes the window-nav contract explicit rather than accidental.
Regression tests cover all three: C-SPC bound to the forwarder, and the window-nav keys in the exceptions with the semi-char map no longer forwarding them.
|
| |
|
|
|
|
|
|
| |
F9 did nothing in an agent buffer: ghostel's semi-char mode forwards every key not in ghostel-keymap-exceptions to the pty, and ghostel-semi-char-mode-map outranks the major-mode map, so the F9-family and F12 bindings I'd put in ghostel-mode-map never fired. The keys went to Claude/the shell, which ignored them.
I added the F9 family (in ai-term) and F12 plus C-; (in term-config) to ghostel-keymap-exceptions and rebuilt the semi-char map with ghostel--rebuild-semi-char-keymap. add-to-list updates the list but not the already-built map, so the rebuild is what actually lets the keys through. C-; had the same latent bug for the same reason.
Two regression tests assert the keys are in the exceptions and that the rebuilt semi-char map no longer forwards them. I also corrected the spec note that claimed binding in ghostel-mode-map was enough (true for vterm, wrong for ghostel) and codified the gotcha.
|
| |
|
|
|
|
|
|
| |
I swapped the terminal engine from vterm to ghostel (libghostty-vt) everywhere. term-config replaces vterm-config (the F12 terminal, the C-; x menu, tmux history capture), and ai-term replaces ai-vterm (the F9 Claude-agent launcher). ghostel renders the agent TUI without vterm's flicker under heavy streaming, and one engine now covers every terminal workflow.
Two behavior changes fall out of the swap. F9 launches in a terminal frame now: ghostel renders in TTY frames, so the old GUI-only guard is gone. Terminal windows no longer dim when unfocused: ghostel resolves its palette into the native module per-terminal, so there's no per-window color hook to dim through the way vterm had.
auto-dim drops its vterm color-advice path, the dashboard Terminal button launches ghostel, and the vterm and vterm-toggle packages are removed. The tmux pane-history and copy-mode machinery carried over unchanged. It keys on the pty tty, which ghostel exposes.
|
| |
|
|
|
|
|
|
| |
The config carried a full custom command surface: a hand-built C-; L keymap, which-key labels, a lazy API-key loader wired in as advice, and a pinned DeepSat team id. That's worth keeping only if the out-of-box pearl experience isn't good enough on its own, and the point now is to find out.
Stripped it back to exactly what pearl's README documents for a first install: pearl owns its own keymap (pearl-mode binds the command surface under pearl-keymap-prefix, default C-; L, in any Linear buffer), no global binding, no advice. The API key comes from authinfo.gpg via auth-source-pick-first-password, and the synced file moves to gtd/linear.org under org-directory. The only deviation from the README is loading from the local checkout at ~/code/pearl instead of an archive.
Deleted tests/test-linear-config.el — it covered the custom helpers (the key-loader advice, the keymap wiring) that no longer exist. What's left is declarative use-package config with nothing of its own to test.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
F9 toggle-off used quit-restore-window to dismiss the agent. With several agents alive sharing one slot, switching among them (C-F9) reuses the window via set-window-buffer, which leaves the window's quit-restore parameter pointing at the first agent shown. Once stale, quit-restore-window falls back to switch-to-prev-buffer and surfaces a different agent instead of removing the window, so F9 appeared to "show another agent" rather than hide the split.
Toggle-off now collapses the split with delete-window, which is independent of the slot's buffer history, so the working buffer reclaims the frame. Geometry is captured first so the next toggle-on re-splits at the same width.
Toggle-on reopens the exact agent that was hidden (new cj/--ai-vterm-last-hidden-buffer), falling back to the most-recent agent only when that buffer has been killed. Hide-then-show is now a faithful round trip, not a jump to whichever agent is most-recent in buffer-list.
Sole-window toggle-off returns to the most-recent non-agent buffer instead of other-buffer, which could land on another agent.
I updated the two reuse-edge-window tests that asserted the old restore-displaced-into-a-kept-slot behavior to match the new always-collapse behavior.
|
| |
|
|
| |
The daily-prep workflow now keeps its stable symlink at the project root as daily-prep.org instead of inbox/today-prep.org. I repointed cj/open-project-daily-prep (C-c p d) to match, updating its docstring, not-found message, and test for the new path.
|
| |
|
|
|
|
| |
AI-vterm launches a graphical vterm side window, so F9 / C-F9 / M-F9 now decline with a message in a terminal frame instead of opening a vterm. The guard checks the current frame at command time rather than at load. That matters under the daemon, which serves GUI and terminal frames both with display-graphic-p nil at load, so a load-time gate would have disabled the launcher in its GUI frames too.
Routed the three window-behavior tests through a GUI-frame stub, since a batch run is itself a terminal frame.
|
| |
|
|
|
|
|
|
| |
I added a display-buffer-alist entry matching "*Signel:" chat buffers and routing them through display-buffer-at-bottom with window-height 0.3. The signel fork's signel-chat now uses pop-to-buffer instead of switch-to-buffer, which is what makes the rule apply. Without that switch the buffer replaces the current window and skips display-buffer entirely.
Two new tests in test-signal-config.el lock the entry shape and the regex's buffer-name match set. A new test-signel-cancel-input.el covers the fork's C-c C-k handler. It clears the editable region between signel--input-marker and point-max, then quit-windows so the buffer survives the dismiss.
Closes the "Chat buffer placement + exit keys" task filed during the 2026-05-28 manual-verify walk.
|
| |
|
|
|
|
|
|
|
|
| |
cj/signel--ensure-started in modules/signal-config.el was reading signel--process-name in the first branch of its cond before the use-package autoload of signel had fired. The forward-declared (defvar signel--process-name) at L137 silences the byte-compile warning but doesn't actually bind the variable. Its value comes from signel.el's defconst, which doesn't run until signel is loaded.
The first call to cj/signel-connect (C-; M SPC) after Emacs launch produced "Symbol's value as variable is void: signel--process-name" instead of starting the daemon. Surfaced tonight during the manual verify walk of the initiate-message workflow.
I added (require 'signel) at the top of cj/signel--ensure-started so signel loads before any of its variables get read. The require is idempotent, so callers that hit the function after signel is already loaded pay nothing.
The new ERT test test-signal-config-ensure-started-requires-signel-first asserts ordering: require must be the first call inside the function, not just called somewhere. A future refactor that moves the require below the cond would fail this test instead of passing silently.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I built pieces 2-7 of the initiate-message workflow from docs/design/signal-client.org and added tests covering the fork's clobber-fix (commit 5ec56c0 over there). The picker is the user-facing change: a single key opens a name-based completing-read for any contact, with "Note to Self" pinned first.
The picker stack from the bottom up:
cj/signel--ensure-started is the daemon guard. With a live process it's a no-op. With signel-account set but no process it calls signel-start and pre-warms the contact cache. With signel-account nil it user-errors naming the remedy. Pre-warming on start means the picker feels instant on first use.
cj/signel--fetch-contacts issues a listContacts RPC through the new request-callback contract (signel--send-rpc with a success-callback). The callback runs the result through the verified cj/signal--parse-contacts and stores the (LABEL . RECIPIENT) alist in cj/signel--contact-cache, a cj-owned variable kept separate from signel's receive-time contact-map. An empty result populates the cache as nil, distinct from an RPC failure (which never invokes the callback so the prior cache survives). cj/signel-refresh-contacts is the user-facing command that clears and refetches.
cj/signel-message is the picker. Warm cache opens completing-read immediately. Cold cache kicks off a fetch and accept-process-outputs up to cj/signel-fetch-timeout seconds (3s default), then user-errors if the daemon hasn't responded so a wedged process can't hang Emacs. The candidate list pins "Note to Self" first (resolves to signel-account) with a display-sort metadata function that preserves the given order rather than alphabetizing.
cj/signel-message-self skips the picker and goes straight to signel-account. cj/signel-connect is the friendly verb on the prefix key.
cj/signel-prefix-map binds m / s / d / q / SPC and attaches under C-; M via with-eval-after-load keybindings so the binding survives load-order. l stays unbound for the future link command.
15 new ERT tests cover the ensure-started branches, the fetch + cache contract (issued, populated, empty), refresh-contacts, the picker's four scenarios (warm-cache contact, warm-cache Note to Self, cold-cache resolves in time, cold-cache timeout), message-self, and the keymap bindings. Plus 4 new tests in tests/test-signel-input-preservation.el for the fork's clobber fix: pending-input captures typed text and returns nil when empty; both signel--insert-msg and signel--insert-system-msg redraw the prompt without clobbering "halfwritten".
todo.org closes three tasks as dated event-log entries: the contact picker, the input clobber, and the use-package wiring.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The fork commit (4740d97 in the signel fork) added a request-callback table, extended signel--send-rpc with an optional success-callback, and routed result responses through signel--dispatch. These tests lock that contract from the consuming project so a future fork change can't silently break the picker that will read listContacts through it.
Five tests, Normal / Boundary / Error categories plus a reconnect-invalidation case:
- result-invokes-callback (Normal): a result response with a registered id fires the callback with the value and removes the handler.
- send-rpc-registers-success-callback (Normal): passing a success callback stores it under the returned id.
- unknown-id-is-noop (Boundary): a result with no registered id is silent — no receive or error handler fires, map stays empty.
- error-cleans-up-handler (Error): an error response removes the handler without firing the callback, so a retry starts clean.
- stop-clears-handler-map: signel-stop empties the map, so a restart can't replay stale callbacks waiting on responses that will never arrive.
The dispatch tests synthesize JSON alists directly. No live process is needed. The send-rpc test stubs get-process and process-send-string so it doesn't need a running signal-cli.
Refactor audit on signel.el surfaced one unrelated pre-existing smell I'm not fixing here: signel--handle-error reads from signel--request-buffer-map but never remhashes, so error responses leak request-id → buffer-name entries. Filed as a separate [#C] follow-up under the Signal parent task; the maps clear on stop/start so the impact is bounded to a single live session.
todo.org: the dispatch task flips to DOING (the fork commit is in, the test contract is locked) and gets the leak follow-up appended.
|
| |
|
|
|
|
|
|
|
|
| |
The *AI-Assistant* side window always opened at a fixed 0.4 width, so resizing it by hand was lost the next time it opened. Now the F-key toggle captures the panel's width when it closes and reopens at that width for the rest of the session, the same way the music playlist remembers its height.
The panel has three entry points that all open the same buffer: the toggle, loading a saved conversation, and escalating a quick-ask. I gave them one shared remembered-width var (cj/--ai-assistant-width, owned by ai-config; the other two forward-declare it to avoid a circular require), so the panel comes back at one consistent width whichever door opens it. Capture lives only in the toggle's close path; the other two just replay.
One latent edge: ai-conversations replays with its configurable side, which defaults to right. If that's ever set to top or bottom, the remembered width fraction would land as a height. It can't happen at the default, so I left it as a known edge rather than complicating the call now.
The escalation test needed a top-level defvar for the shared var: a value-less defvar in the module is only file-local to the byte-compiler, so without it the function read the var dynamically and hit void when the test loaded ai-quick-ask without ai-config. ai-quick-ask 13/13 and ai-conversations 47/47 green.
|
| |
|
|
|
|
|
|
| |
F9 brought the agent window down from the top of the frame. The toggle remembers where the window last sat and replays it, and "above" was a position it could capture and replay: move the window to the top with the buffer-move keys, toggle off, and the next toggle reopened it up there. The host default never picks the top, so a remembered "above" was the only way in.
I added an optional allowed-directions list to cj/window-toggle-capture-state, the helper both F9 (ai-vterm) and F12 (vterm-config) share. When the captured direction isn't in the list, it falls back to the default direction and clears the saved size, since that size was measured on the disallowed axis and wouldn't transfer. Both dispatchers now pass (right below left), so neither can remember a top placement. They go through the same helper, so the rule stays in one place.
Three tests cover the new branch: a permitted direction is kept, a disallowed one falls back with the size cleared, and an omitted list preserves the old unconstrained behavior so existing callers are unaffected.
|
| |
|
|
|
|
|
|
|
|
| |
A folded heading that also has a property drawer renders two glyphs after its right-aligned tag: the org-tidy inline dot (" ·") and the fold ellipsis (" ▾"). The tag-align reserve was 5 columns, and that " · ▾" spilled onto a second screen row, leaving a stray indicator line under the heading.
The trailing glyphs measure 4 columns nominally, so the old 5-column reserve looked like it should fit. It didn't: the fallback triangle renders wider than its reported width and the :align-to stretch rounds, so the real overflow exceeds the column count. I sized the new reserve (9) from a rendered measurement, not arithmetic, and the docstring now says so and tells the next reader to verify by screenshot. That mismatch between column math and what actually renders is what let the earlier reserve ship broken.
I verified the fix by rendering the real config off-screen at full width and reading the result: before, the two folded property-drawer headings wrapped "· ▾" to a second row; after, every heading is a single line.
tests/manual/headline-wrap/ holds the fixture and a README so the check can be repeated by opening the file and looking, which is the only honest way to test a redisplay bug like this.
|
| |
|
|
|
|
| |
Plain left-click on an Org file link keeps org's default of opening in the other window. I added S-mouse-1 and mouse-3 as a second gesture that opens the file link in the current window instead, for when I want the file to replace the buffer the link sits in.
The bindings live in org-mouse-map, the keymap org attaches to each link as a text property, rather than org-mode-map. That layer outranks both org-mode-map and the mouse-trap-mode emulation keymap, which otherwise swallows clicks in org buffers. Off a link the gesture does nothing, so a stray right-click is a silent no-op instead of a "No link found" error.
|
| |
|
|
| |
C-c p d (cj/open-project-daily-prep) always forced the prep doc into another window. I switched it to cj/--find-file-respecting-split, the same helper the sibling C-c p t (project todo) uses: it opens the other window when the frame is split and reuses the current window when it isn't. The two project-open commands now behave the same way.
|
| |
|
|
|
|
|
|
|
|
| |
The F10 music playlist opened at a fixed fraction every time, so any manual resize was lost the moment it was toggled closed. Now the toggle captures the window's size on close and reopens at that height, for the rest of the session.
The mechanism is generic, not music-specific. cj/window-size-fraction (geometry-lib) is the pure kernel: a clamped window/frame ratio. cj/side-window-capture-size and cj/side-window-display (toggle-lib) wrap it for any display-buffer-in-side-window consumer — height for top/bottom, width for left/right — storing the remembered fraction in a caller-supplied state var. It mirrors the direction-split toggle pattern the vterm dispatchers already use, but for atomic side windows that can't be split.
music-config wires F10 to it: cj/music-playlist-window-height is the default, cj/--music-playlist-height holds the remembered value (in-memory, resets each session).
12 new tests across the two libs, Normal/Boundary/Error each covered.
|
| |
|
|
|
|
|
|
|
|
| |
I'm building a Signal client in Emacs on signal-cli (linked as a secondary device) with a fork of the signel package as the front end. signel is on MELPA but effectively abandoned, and the behavior I want needs internal edits, so owning a fork beats advising a dead package. Full rationale and the rejected alternatives are in docs/design/signal-client.org.
This lands the signal-cli-independent foundation: contact-list parsing for a completing-read picker, and the predicate that suppresses a notification for the chat being actively viewed. Both are pure and unit-tested without a linked account. cj/signal--parse-contacts was corrected against a live account (signal-cli 0.14 puts givenName/familyName at the top level, not under profile), and verified across all 94 real contacts.
The use-package wiring loads the fork from ~/code/signel, sources the account from a gitignored signal-config.local.el (a phone number is an identifier, not a credential, and this keeps it off the mirror without a GPG prompt), and turns off auto-open so an incoming message can't steal a window. Verified live: signel-start spawns the jsonRpc process, loads the account, and receives over the channel.
The fork edits (notify routing, the upstream input-clobber bug) and the contact-picker command are still to come.
|
| |
|
|
|
|
| |
I added cj/open-project-daily-prep on C-c p d. It opens inbox/today-prep.org under the Projectile project root in another window, project-scoped, so a project without a prep file just reports it rather than erroring. The binding had only ever been eval'd live into the daemon in a past session and vanished on the next restart, so this persists it to the module.
Freeing d meant reworking the deadgrep bindings. deadgrep-in-dir moves to C-c p G (replacing plain deadgrep, which stays M-x-callable), and deadgrep-here keeps C-c p g. Plain project-wide deadgrep dropped off the projectile prefix because it overlapped the context-aware and arbitrary-directory variants. Tests cover the open, missing-file, and not-in-a-project paths.
|
