| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| | |
|
| |
|
|
|
|
|
|
| |
Migrate `cj/set-wallpaper' to the new helper. The old code messaged a bare `"%s not found"' when feh or swww was missing -- the helper produces a better message ("feh not found; wallpaper setter unavailable") that lands in *Warnings* instead of flashing once in the echo area.
Add `(require \='system-lib)' to dirvish-config.el per the Phase 2 exit criterion.
Audit notes for the other `executable-find' call sites in prog-shell, prog-c, prog-go, prog-python, browser-config: all silent `:if' / LSP-availability checks (spec says keep silent), or interactive commands with platform-specific install hints in their existing messages (`shellcheck`, `mypy`, `delve`). Migrating those would lose the install hints; leaving them.
|
| |
|
|
|
|
|
|
|
|
| |
Phase 2 of utility-consolidation, first commit per the spec's recommended order. `cj/mail--executable-or-warn' was the right pattern -- check executable-find, return the path, otherwise emit a clear `display-warning' naming the feature -- but it was trapped in mail-config and only mail callers benefited. Lift it into `cj/executable-find-or-warn' in system-lib.el with one new argument: an optional GROUP symbol that flows through to `display-warning' (defaulting to `cj/system-lib') so per-feature warning filters keep working. Mail callers pass `mail-config' explicitly.
Migrate `cj/mail--mbsync-command' and `cj/mail-configure-smtpmail' to the new helper. Drop the local definition. Add `(require \='system-lib)' to mail-config.el per the spec's Phase 2 exit criterion ("consumer modules explicitly require system-lib").
Five Normal/Boundary tests cover the four return-shape cases (program found / program missing / warning content / default vs explicit group).
Other consumers (prog-*.el, dirvish-config.el, browser-config.el) still call `executable-find' directly. Migrating them is a follow-up commit, audited per call site -- the spec flags some `:if' silent checks as intentional and they should NOT switch to the warning helper.
|
| |
|
|
|
|
|
|
|
|
| |
Phase 1 of utility-consolidation per docs/design/utility-consolidation.org. The inventory walks the spec's 30-entry Candidate Extraction Table and, for each helper, records: visibility, dependencies, side effects, callers in modules and tests, test file location, extraction priority, and a Migrate / Leave / Defer decision with rationale.
Decisions: 11 Migrate, 3 Leave, 13 Defer. The Migrate items are grouped by phase in the spec's recommended order: Phase 2 (foundation helpers -- executable lookup, shell quoting, process runner, file-from-context), Phase 3 (Org-safe text sanitizers), Phase 4 (external-open consolidation). The Defer items mostly need a second production caller before promotion is justified.
Discoveries worth recording: `cj/log-silently' already has 10 production callers (more than the spec's table suggested), and `cj/--file-manager-program-for' shipped today in dirvish-config.el is the new form of OS-dispatch consolidation -- Phase 4's `cj/external-open-command' should fold it in rather than re-deriving.
No code behavior changes -- this is the spec's stated Phase 1 exit criterion.
|
| |
|
|
|
|
| |
`cj/dired-mark-all-visible-files' classified the current line as a directory via `(looking-at "^. d")' inline. Lift the classification into `cj/--dired-line-is-directory-p', a string predicate that takes a line and returns non-nil when it's a directory listing. The wrapper still walks the dired buffer line by line and calls `dired-mark' -- that iteration is dired-coupled and stays untested -- but the format-aware predicate is now isolated and verified.
Six Normal/Boundary tests cover unmarked directories, marked directories (`*' prefix), regular files (`-' instead of `d'), symlinks (`l'), empty lines, and dired header lines (` /path:' and ` total N').
|
| |
|
|
|
|
| |
`cj/dirvish-open-html-in-eww' inlined a `string-match-p' against `\.html?\=' to decide whether to hand a file to eww. The check was case-sensitive, so `.HTML' fell through to the "Not an HTML file" message even though every browser treats it as HTML. Lift the predicate into `cj/--html-file-p' and bind `case-fold-search' to t so uppercase and mixed-case extensions match. The trailing-`\=' anchor stays so files like `html-thing.org' still don't match.
Seven Normal/Boundary/Error tests cover lowercase `.html', `.htm', uppercase `.HTML', mixed-case `.Html', embedded `html' (no match), non-html extensions (no match), and no-extension files (no match).
|
| |
|
|
|
|
|
|
| |
`cj/dired-ediff-files' had its pair-determination logic inline: count check, prompt fallback when only one file was marked, and the older-first ordering for `ediff-files'. Lift it into `cj/--ediff-pair-from-files' -- pure given the file list, an injected prompt thunk, and a newer-than-p comparator -- so tests stay independent of mtimes and the dired prompt.
While extracting, surface a latent bug: with zero marked files the original code fell through to `(file-newer-than-file-p nil nil)' and crashed with a wrong-type-argument error. Replace the crash with a clear `user-error' ("No files marked"), and add a regression test. The 3+ files case keeps its existing user-error message.
Five Normal/Boundary/Error tests cover both ordering directions, the one-file prompt path, and both error counts.
|
| |
|
|
|
|
|
|
|
| |
`cj/dired-create-playlist-from-marked' had its audio-file filtering and trailing-`.m3u' stripping inline among the dired marking, prompting, overwrite-loop, and file-write logic. Lift each into its own pure helper:
- `cj/--playlist-filter-audio (files extensions)' returns only files whose extension matches one of EXTENSIONS (lowercase, no dot). Case-insensitive on the file side.
- `cj/--playlist-sanitize-name (name)' strips a trailing `.m3u' suffix; embedded `.m3u' that isn't at the end is preserved.
Ten Normal/Boundary tests cover keep-only-audio, case-insensitivity, files-without-extension excluded, empty inputs, and the sanitize edge cases (bare name, embedded `.m3u', empty string, just `.m3u').
|
| |
|
|
| |
`cj/get-project-root' was already in the right shape (pure dispatch over `projectile-project-root' and `project-current'), but had no tests. Add four covering the four meaningful states: projectile finds a root, projectile returns nil and project.el finds one, projectile errors and the wrapper falls through to project.el, and neither finds anything. The tests stub all three external symbols (`projectile-project-root', `project-current', `project-root') so they exercise the function without a project on disk.
|
| |
|
|
|
|
| |
`cj/dirvish-open-file-manager-here' had its platform-dispatch -- xdg-open / open / explorer / shell fallback -- inline as a four-arm cond branching on `executable-find' and `system-type'. Lift the choice into `cj/--file-manager-program-for', a pure function from (HAS-XDG-OPEN-P SYSTEM-TYPE) to a program-name string or nil. The wrapper resolves the live values, asks the helper, and either calls the program or falls back to a shell-command.
Six Normal/Boundary tests cover the four return shapes (xdg-open across system types, darwin/windows-nt without xdg-open, the nil fallback for plain Linux without xdg-open and for unknown system-types like Haiku).
|
| |
|
|
| |
`cj/set-wallpaper' had two parallel cond arms hardcoding the X11/Wayland dispatch and the success/failure messages inline. Lift the program-and-args choice into `cj/--wallpaper-program-for' -- a pcase from a display-server symbol to a (PROGRAM ARG...) list, or nil for unknown environments. The wrapper now: detect env, ask helper for the command, surface the right message (unknown server / executable missing / success). Adding a third backend (e.g. xdg-desktop-portal) becomes one pcase clause + one test.
|
| |
|
|
|
|
| |
`cj/dired-copy-path-as-kill' was a 57-line procedural body with the path-resolution branching mixed into the kill-ring write and the user-visible message. Lift the four-way decision (force-absolute / project-relative / home-relative / bare absolute) into `cj/--dired-resolve-display-path', a pure function from (FILE PROJECT-ROOT HOME-DIR FORCE-ABSOLUTE) to (PATH . PATH-TYPE). The wrapper now reads as: get the dired file, ask the helper how to display it, format-or-pass-through, kill-new, message.
Seven Normal/Boundary tests cover each branch: project-relative, home-relative, the home-itself "~" glyph, absolute fallback, force-absolute beating both project and home, and project taking precedence over home when both apply.
|
| |
|
|
|
|
| |
The name-mangling logic in `cj/dirvish-duplicate-file' was inline -- inseparable from the dired side effects (existence check, copy, revert). Extract to `cj/--duplicate-file-name', a pure function from FILE to FILE-WITH-COPY-SUFFIX. Seven Normal/Boundary tests cover the cases I care about: typical extension, elisp file, no extension, multi-dot extensions (only the last dot counts), leading-dot dotfiles, relative paths, spaces in the base name.
The wrapper retains the dired-mode interactive shape and now reads as a thin shell over the pure helper.
|
| |
|
|
|
|
| |
Pressing `g' in dirvish (`dirvish-quick-access') failed with `Wrong type argument: command, (keymap (107 . transient:dirvish-quick-access::N))'. Transient auto-groups multi-character keys into nested prefix keymaps; when one entry's KEY is also the prefix of a longer entry's KEY, the shorter slot resolves to the auto-generated sub-prefix keymap rather than to the intended leaf command. The "wallpaper" entry on `pw' and the "project work" entry on `pwk' (added in 4ece1eb) collided. Rename `pw' to `wp'; `pwk' is now the only entry under the `pw'-prefix, so transient builds a clean leaf.
Add `tests/test-dirvish-config-quick-access.el' with five tests: four cover a small `test-dirvish-config--key-collisions' helper against synthetic inputs (collision detected, clean list, empty, single-entry), and one regression test asserts the live `dirvish-quick-access-entries' contains no shorter-key-is-prefix-of-longer-key pair.
|
| |
|
|
|
|
|
|
|
|
| |
The personal vterm map was on `C-; V'. The capital V costs a Shift on every keystroke into the menu, which adds up for the daily `C-; V c' / `C-; V C' bindings. Move the prefix to lowercase `C-; x' -- free, no Shift, faintly mnemonic (xterm/execute). The lowercase `C-; v' stays the version-control menu.
Wire `vterm-next-prompt' and `vterm-previous-prompt' into the menu so they're reachable everywhere, not only inside vterm-copy-mode-map. Lowercase `n' and `p' match Emacs's idiom for next/previous; bump "new vterm" up to capital `N' for the rare new-buffer case.
Drop the `<pause>' binding for `vterm-copy-mode' from `vterm-mode-map'. Modern keyboards rarely have a Pause key and `C-; x c' is the canonical entry now.
Update which-key labels and tests; `test-vterm-keymap-includes-history-and-copy-bindings' now asserts the new prefix, and two new tests cover prompt-nav bindings and the dropped `<pause>' binding.
|
| |
|
|
|
|
| |
The 3-pixel bar was visible but a block matches the rest of my Emacs cursor and lets the standard cursor color and `blink-cursor-mode' behavior carry through unchanged. Same enter/exit semantics: forced visible on entry, buffer-local override killed on exit so the live terminal goes back to the TUI's chosen state.
Update the test expectations and rename the "prior-was-box" boundary test to "prior-was-hbar" so it still proves the override does something (the prior and the override would otherwise both be `box').
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
The unit tests for the cursor-restoration hook only exercised the helper in isolation. The real integration -- toggling the minor mode and watching the hook fire as part of the chain -- wasn't covered. If `vterm-copy-mode-done' or `cj/vterm-copy-mode-cancel' broke their exit semantics (or our hook stopped firing on `vterm-copy-mode-hook'), the unit tests would still pass but the cursor would stay stuck on the bar in real use.
Add five integration tests that toggle the actual minor mode through stubbed `vterm--enter-copy-mode' / `vterm--exit-copy-mode' (so we don't need a live vterm process) and assert the cursor moves through the full lifecycle: nil -> bar on enter, bar -> killed-local on exit. Cover all three exit paths Craig hits in normal use:
- vterm-copy-mode -1 directly (the toggle)
- vterm-copy-mode-done with an active region (M-w / RET)
- vterm-copy-mode-done with no region (line-selection branch)
- cj/vterm-copy-mode-cancel (C-g / <escape>)
Plus a multi-cycle test so a regression in `kill-local-variable' handling shows up.
|
| |
|
|
|
|
|
|
| |
vterm's C module sets `cursor-type' to nil whenever the underlying TUI sends DECTCEM (`\e[?25l') to hide the terminal cursor. Most full-screen TUIs do this on startup — Claude Code in an ai-vterm being a daily example. Once the cursor is hidden at the buffer level, vterm-copy-mode inherits that nil and the user can't see where point is when navigating to select text. Selection still works, but you're flying blind.
Add a `vterm-copy-mode-hook' that forces `cursor-type' to a 3-pixel bar on entry and kills the buffer-local override on exit. The bar shape is drawn between characters rather than by inverting one, so heavy TUI face properties don't hide it either. On exit the live terminal goes back to whatever vterm's tracking says, so the TUI's chosen cursor state resumes.
4 ERT tests cover the hook's enter/exit behavior and confirm registration on `vterm-copy-mode-hook'.
|
| |
|
|
|
|
| |
The combined module had grown to 573 lines covering two unrelated subsystems with no shared state — the eshell shell-mode commands and the vterm/F12 toggle. The header even rendered this with two `;; ----` dividers. Split into two focused modules. eshell-config.el keeps the eshell user commands and package wiring (~170 lines). vterm-config.el keeps the vterm package, the tmux history capture command, the F12 toggle, and the C-; V keymap (~400 lines). Update init.el to require both, point the four vterm test files at vterm-config, and refresh the cross-module commentary in cj-window-geometry.el and cj-window-toggle.el.
No behavior change. Full test suite green; validate-modules clean.
|
| |
|
|
|
|
| |
The F12 commit (554b32d) flagged this as a follow-up: ~120 lines of capture-state and display-saved logic were duplicated between modules/ai-vterm.el and modules/eshell-vterm-config.el. The only differences were the default direction (right for F9, below for F12) and the customization name for the fallback size. Extract the shared logic into modules/cj-window-toggle.el so both consumers reduce to thin delegates that pass their state-var symbols and defaults. The state vars stay where they were, so existing tests against each consumer's helpers keep working.
10 new tests cover the parameterized helpers in isolation. All consumer tests still pass.
|
| |
|
|
| |
Quote the symbol name `below' in three F12 toggle docstrings so the byte-compiler stops flagging the unescaped single quotes. Same change in each: `'below` becomes `\`below'` so Emacs renders it as a code reference. Also drop the unused `cj/vterm-history` alias — no callers in the tree.
|
| |
|
|
| |
Move calendar feed conversion into an isolated batch Emacs worker so large parse/write cycles do not freeze interactive editing. Cover the worker command, isolated logging, quoted settings, and sync success/failure paths with focused ERTs.
|
| |
|
|
| |
Set load-prefer-newer for Makefile batch/test invocations. This matches normal startup and keeps tests from accidentally loading stale .elc files instead of the edited source.
|
| |
|
|
| |
Add an Emacs-first copy workflow for vterm and tmux. C-; V c enters vterm copy mode, C-; V C captures tmux pane scrollback into a temporary Emacs buffer, M-w copies and returns, and C-g/Escape cancel without copying. This also adds clickable URLs, removes the bad vtermf binding, unbinds C-c C-t, and tests the vterm/tmux keymap behavior.
|
| |
|
|
| |
Stop automatically stashing, pulling, and popping dirty repos during reconciliation. Clean repos still pull, dirty repos open Magit for review, and results now include structured statuses, skip reasons, pruning, and a summary.
|
| |
|
|
| |
Remove the duplicate Org cache keymap and keep C-; O owned by the shared Org map. The cache clear command now clears all Org buffers by default, with a prefix argument for the current buffer.
|
| |
|
|
| |
Move the local GPTel tool wiring out of init.el and into ai-config. The tools directory and feature list are now configurable, missing optional tools are non-fatal, and focused tests cover the loading behavior.
|
| |
|
|
| |
Cover the startup archive paths and refresh behavior. This locks down the local-first archive setup, online archive opt-in, cache freshness checks, and priority ordering so package startup changes are easier to reason about later.
|
| |
|
|
| |
Switches from `:vc` to `:load-path "~/code/org-drill"` so I can iterate against the local clone. The use-package block keeps the `:vc` form commented immediately above, ready to flip back when the dev work lands.
|
| |
|
|
| |
Upstream chime renamed `chime-day-wide-time` to `chime-day-wide-alert-times` (now a list of times rather than a single string), and consolidated `chime-time-left-format-short` / `-long` / `-at-event` into a single `chime-time-left-formats` alist. I updated my config to match. There's no behavior change beyond keeping chime alerts firing on this Emacs.
|
| |
|
|
| |
The PROJECT_ROOT guard skipped any .el file outside the project on the theory that out-of-tree files are owned by their own project's hooks. In practice, when I edit an .el file in another repo from a Claude session in this one, the other project often doesn't have its own hook running, so the file was getting no validation at all. I dropped the guard so any .el file edited from a Claude session here gets validated, regardless of which repo it lives in.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Eight test files across the ai-vterm and vterm-toggle suites each shipped a small variant of the same cleanup loop: walk `buffer-list`, kill any buffer whose name starts with a given prefix. Each file also re-implemented the `(string-prefix-p ...)` check inline. One file additionally had its own fake-vterm-mode-buffer constructor for tests that needed `cj/--vterm-toggle-buffer-p` to fire.
I pulled the shared logic into `tests/testutil-vterm-buffers.el`:
- `cj/test--kill-buffers-matching-prefix` is the primitive.
- `cj/test--kill-claude-buffers` and `cj/test--kill-test-vterm-buffers` are thin wrappers for the two prefixes that actually appear.
- `cj/test--make-fake-vterm-buffer` constructs a buffer with `major-mode` set to `vterm-mode` without launching a real vterm process.
Each affected test file now `(require 'testutil-vterm-buffers)` and calls the shared helpers directly. `test-vterm-toggle--buffer-filter.el` keeps a 3-line wrapper that calls both kill helpers in sequence (the only place that needs both prefixes). Net diff: -116 / +72 across 8 test files, plus ~30 lines in the new testutil. Roughly -45 lines after the abstraction is paid for.
No behavior change. 80 ai-vterm tests, 15 vterm-toggle tests, 15 cj-window-geometry tests all pass. Full make test green.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
`ai-vterm.el` (F9) and `eshell-vterm-config.el` (F12) both grew the same geometry-preservation pattern: classify a window's position, capture its body size, map cardinal direction to its frame-edge variant. The shared helpers were sitting as near-duplicates in both modules. With two real consumers established, the abstraction has the right shape. I pulled them into `cj-window-geometry.el`.
The new module exposes three pure helpers:
- `cj/window-direction` returns right/below/left/above based on edges relative to `frame-root-window`. Takes an optional DEFAULT for the single-window-frame fallback so each consumer picks its own (ai-vterm wants 'right, vterm-toggle wants 'below).
- `cj/window-body-size` returns body-cols (right/left) or body-lines (below/above). Same body-vs-total reasoning as before: divider-independent, matches what the user sees.
- `cj/cardinal-to-edge-direction` maps right/left/below/above to rightmost/leftmost/bottom/top, used by each consumer's `display-saved` action.
`ai-vterm.el` and `eshell-vterm-config.el` now `(require 'cj-window-geometry)` and call the shared helpers directly. The consumer-specific `capture-state` and `display-saved` bodies stay in each module because they bind to consumer-specific state vars. Extracting those would either need parameter-passing-via-symbol or a macro, both heavier than the duplication they would remove.
Tests: 15 in `test-cj-window-geometry.el` covering all four directions, body-size on both axes, cardinal-to-edge mapping, default-arg fallback, and the unknown-direction nil case. Deleted `test-ai-vterm--window-geometry.el` (now redundant) and trimmed four duplicate window-direction/size tests from `test-vterm-toggle--display.el`. Net LOC: each consumer ~40-50 lines lighter, with the new module + tests paying roughly half that back. Full make test green. make validate-modules green.
|
| |
|
|
|
|
|
|
| |
The C-F9 project picker now flags projects whose claude buffer is alive with a " [running]" suffix on the abbreviated path. I added `cj/--ai-vterm-format-candidate` to compute the display name and routed the picker through it. Before the change, the picker showed every candidate identically, so you couldn't tell at a glance whether picking a project would attach to an existing session or start a fresh one.
F9 with two or more alive claude buffers used to open the project picker. That meant after toggling claude-A off, opening claude-B via C-F9, then toggling B off, the next F9 dropped into a picker rather than redisplaying B (the one you just toggled off). I renamed `redisplay-single` to `redisplay-recent` in `cj/--ai-vterm-dispatch` and broadened the trigger from "exactly one alive" to "one or more alive". F9 now redisplays the MRU claude buffer, so it consistently means "toggle THE claude I was last using". The project picker stays explicit on C-F9 for "start a different project", and M-F9 still picks among existing claudes.
2 new tests for the indicator (`format-candidate` flagged + unflagged), 2 dispatch tests renamed to match the new contract. 80 ai-vterm tests pass. Full make test green.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
vterm-toggle picked the most-recently-selected vterm buffer as F12's toggle target. After using F9 on a claude vterm, the most-recent vterm IS claude, so F12 ended up toggling claude, which has its own F9 / C-F9 / M-F9 surface in ai-vterm.el and shouldn't be affected. The display rule also had a hard-coded `(window-height . 0.7)` that overrode mouse-resize and orientation flips on every toggle.
I replaced the F12 binding with `cj/vterm-toggle` in `eshell-vterm-config.el`, mirroring the pattern shipped in ai-vterm.el:
- `cj/--vterm-toggle-buffer-p` excludes claude-prefixed buffers from F12's candidate set.
- `cj/--vterm-toggle-capture-state` records direction + body size at toggle-off.
- `cj/--vterm-toggle-display-saved` replays via `(body-columns . N)` / `(body-lines . N)` cons forms with the cardinal direction mapped to its frame-edge variant (`right` -> `rightmost`, `below` -> `bottom`, etc.) so vterm always lands at the captured edge regardless of selected window.
- `cj/vterm-toggle` uses `delete-window` (with `one-window-p` guard) on toggle-off so buffer-move scenarios don't leak ghost windows.
Default direction is `'below` to match F12's traditional bottom split. The vterm-toggle package stays installed so `M-x vterm-toggle` still works. Only the F12 binding changes.
19 new tests across three files: buffer-filter, dispatch, display. Full make test green.
Tradeoff: ~150 lines of geometry helpers and capture/display action logic are duplicated from ai-vterm.el. Worth extracting into a shared module now that two consumers exist. I filed it as a follow-up rather than blocking this ship.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Live-testing surfaced four edge-case failures in the F9 toggle geometry preservation. Each gets a dedicated regression test.
- Multi-window squeeze: a captured fraction-of-frame replayed at the wrong size in 3+ window layouts because `display-buffer-in-direction` interprets float widths as fractions of the new window's parent, not the frame. In a flat 3-window layout the parent is the root, but in nested splits it's a sub-tree, and the captured fraction blew the layout up. I switched to absolute integer body-cols and body-lines as the captured unit. The unit is layout-independent.
- One-col peek: a claude window captured rightmost (no right divider, body=total) replayed into a middle position (with divider, body=total-1) showed 1 col of the sibling buffer peeking through where claude should have ended. I wrap the integer size in a `(body-columns . N)` / `(body-lines . N)` cons so `display-buffer-in-direction` sets the body explicitly, divider-independent.
- Position swap and compounding gap: `direction=right` in `display-buffer-in-direction` splits the selected window, not the frame edge. In multi-window layouts the new claude landed mid-frame instead of where it came from. Each toggle compounded a 1-col loss because the new position picked up a divider the original lacked. I map the cardinal direction to its frame-edge variant (`right` -> `rightmost`, `below` -> `bottom`, etc.) so claude always returns to the captured edge.
- Extra window after buffer-move: buffer-move (C-M-arrows) doesn't update the claude window's `quit-restore` parameter, so `quit-window` falls through to bury rather than delete. The window stays alive showing some other buffer. Toggle-on doesn't recognize it and creates a fresh side window, landing at N+1 windows. I switched to `delete-window` with a `one-window-p` guard for the single-window-frame case.
One tradeoff: in a layout where claude was deliberately in a middle position (e.g. agenda | claude | todo), the next toggle pulls it to the frame edge rather than the middle. The side-panel pattern is the design intent and the common case.
7 new regression tests covering each scenario. 80 ai-vterm tests pass. Full make test green.
|
| |\ |
|
| | | |
|
| |/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
F9 was a single command that always opened the project picker. Three small frustrations stacked up. With one claude buffer open and not visible, F9 was a redundant prompt to pick a project that already had a session. With claude visible, there was no way to bury it without M-x quit-window. With two projects' buffers alive, swapping between them was a buffer-switch chore.
F9 is now a dispatch:
- Claude visible in this frame: quit the window (toggle off) and capture the geometry first.
- Exactly one claude buffer alive but hidden: re-display it (DWIM single-buffer case).
- Zero or two-plus alive: fall through to the project picker.
C-F9 is the always-pick-project entry point for explicit project switches. M-F9 is a buffer picker over the alive claude buffers. If a claude window is currently shown, the picked buffer replaces it in that window so the split orientation and size carry over. The shown buffer sorts last in the picker with a [shown] marker so RET picks "the other one."
Split geometry persists across toggles. Two module-level vars (cj/--ai-vterm-last-direction, cj/--ai-vterm-last-size) capture at toggle-off and feed a custom display action. After M-S-t flips claude from right to bottom, F9 toggle-off-then-on returns it at the bottom. After a mouse resize, the next toggle restores that fraction. State is per-session. Restarts reset to default right/0.5.
Two display-buffer fixes came out of testing:
- save-window-excursion around (vterm name) keeps the dashboard from being buried on a fresh F9 at startup. vterm calls pop-to-buffer-same-window internally, which would otherwise replace the selected window's buffer before the alist could route the new one.
- The action chain swaps display-buffer-use-some-window for a more specific cj/--ai-vterm-reuse-existing-claude. The generic version stole non-claude windows on C-F9 when the user was focused inside claude (claude on bottom, code on top -> new project landed in the code window). The specific version only reuses windows that already show a claude buffer.
I reclaimed C-F9 from the gptel toggle in ai-config.el. C-; a t still binds gptel.
I added eight new test files (claude-buffers, displayed-claude-window, dispatch, pick-buffer-candidates, window-geometry, capture-state, display-saved, reuse-existing-claude) plus a regression test on cj/--ai-vterm-show-or-create for the dashboard-preservation fix. All 73 ai-vterm tests pass and the full make test suite is green.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Two post-ship issues blocked practical use of the new launcher.
The display rule used `display-buffer-in-side-window` with `(dedicated . t)`. Side-window dedication caused `set-window-buffer` to error during `buffer-move` (C-M-arrows), which left a half-finished swap with both sides showing the claude buffer. Then `switch-to-buffer` on a non-claude buffer in that dedicated window split instead of replacing.
I rewrote the rule as `display-buffer-reuse-window -> display-buffer-use-some-window -> display-buffer-in-direction (right)`. The resulting window is ordinary, not dedicated, so swap and replace work normally. I also narrowed `vterm-toggle`'s broad lambda (which matches any vterm-mode buffer) to exclude `claude [` buffers. Otherwise vterm-toggle's `:defer` made it install last and capture our buffers first with its own bottom-split + dedicated treatment.
The tmux side: vterm's auto-launch hook ran a bare `tmux\n`, so each session got an auto-named one. After an Emacs crash the tmux session would survive but I couldn't find it. A second F9 just spawned another. The launcher now sends `tmux new-session -A -s <basename> -c <dir> '<claude>; exec bash'`. The `-A` reattaches to a same-named session if it already exists. The `exec bash` keeps the tmux window alive if claude itself exits. A `cj/--ai-vterm-suppress-tmux` flag tells the existing vterm hook to skip its bare tmux step so the named launch runs instead.
11 new tests across 2 files cover the session-name and launch-command helpers. I updated tests for show-or-create and the display rule. All 34 ai-vterm tests are green.
|
| |
|
|
|
|
|
|
| |
`make validate-modules` had 19 module-load failures, all the same shape: a module references a symbol or feature owned by another module without saying so. Production was fine because init.el orders requires correctly. The batch target loads each module in isolation, though, and surfaces the gap.
I added explicit `(require 'keybindings)` or `(require 'user-constants)` to each affected module. The requires are idempotent at runtime, so production load order is unchanged. For three optional packages (elpa-mirror, mu4e, org-contacts), I switched to `(require 'X nil t)` so the modules load cleanly when those packages aren't installed. The activation calls become no-ops in that case.
`make validate-modules` now reports 0 failures.
|
| |
|
|
| |
`window-swap-states` on M-S-s overlapped with `buffer-move` (C-M-arrows), which I actually use. I cleared the binding plus the M-S -> M-S-s translation in keyboard-compat.el so the keyboard table stays in sync.
|
| |
|
|
|
|
|
|
| |
The new module picks a Claude-template project from a filtered completing-read list. It scans the same roots the `ai` shell launcher uses, then opens or reuses a vterm buffer named `claude [<repo>]` on the right. F9 launches it. The prior `cj/toggle-gptel` binding moves from F9 to C-F9 so both AI tools share the same physical key.
The display rule chains reuse-window -> use-some-window -> in-direction (right). The resulting window isn't dedicated. That matters because side-window dedication was breaking `buffer-move` (C-M-arrows) and `switch-to-buffer` replacement on the claude buffer. I also narrowed `vterm-toggle`'s display rule to skip `claude [` buffers. Otherwise it claimed them first with its bottom-split + dedicated treatment.
I added 23 tests across 5 files: the buffer-name transform, candidate walker, show-or-create dispatch, picker, and display rule. Design lives at docs/design/ai-vterm.org.
|
| |
|
|
|
|
| |
The .claude/rules/ files (commits.md, testing.md, verification.md) were local copies that drifted from the canonical at ~/code/rulesets. I replaced them with absolute symlinks pointing at the canonical files. This repo now picks up rule updates automatically.
The symlinks are absolute. Cloning this repo on a new machine needs ~/code/rulesets at the same path. Otherwise the symlinks dangle.
|
| |
|
|
|
|
|
|
|
|
| |
I replaced the load-time icon-stub block in keyboard-compat with per-call :around advice that checks display-graphic-p against the rendering frame. The old block ran at module-load. Under daemon startup no frame exists yet, so display-graphic-p returned nil and the empty-string stubs installed permanently. Every GUI client connecting to that daemon then saw blanks. The new shape lets one daemon serve real icons to GUI clients and blanks to terminal clients.
I also pulled the nerd-icons-completion and nerd-icons-ibuffer integrations, the package install, and a new tint helper into modules/nerd-icons-config.el. Per-feature use stays in the consuming module (dashboard, dirvish, keyboard-compat). The malformed cons-cell on the marginalia hook in selection-framework.el got fixed in the move.
Added a default darkgoldenrod tint, a :filter-return advice on nerd-icons-icon-for-dir so dir icons pick up a color face, and a buffer-local face-remap in dired-mode-hook so plain files in dired render in shadow grey.
13 tests across 3 new files cover the per-call gate, the dir-color helper (idempotent under nerd-icons' memoized return strings), and the bulk-tint helper.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
I added two sibling design specs in `docs/design/`:
`init-load-graph.org` covers untangling `init.el` from its current "everything eager in a fixed order" shape. It defines a layered architecture (early-init / foundation / core UX / domain workflow / optional), a module category table for every required file, a per-file commentary header standard with seven required lines, a six-phase migration plan with exit criteria, and a testing strategy split into automated batch checks, manual smoke checks, and startup performance baselines via `benchmark-init`.
`utility-consolidation.org` is the sibling project. It covers extracting reusable helpers from feature modules into `system-lib.el` and a small set of topic libraries (`cj-process.el`, `cj-org-text.el`, `cj-cache.el`). It includes a candidate decision criteria section, a library file header standard with worked example, a candidate extraction table with priorities and proposed names, nine helper groups with API plus behavior contracts, naming rules, migration phases, test relocation policy, and a recommended first-three-commits sequence.
Both specs are draft. No code change in this commit. The two projects are intentionally separated because the load-graph project asks "when does this load?" and the consolidation project asks "who owns this helper?". Those are different questions with different rollback shapes.
Implementation tracking lives in `todo.org`.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
`calendar-sync--event-to-org` already cleaned the description body via `calendar-sync--sanitize-org-body`, but the event summary went into the heading line and the location, organizer, status, and URL went into the property drawer without sanitization. Any of those fields containing newlines could create extra Org headings, close the property drawer early with a stray `:END:`, or inject property-looking lines that the agenda would then parse as real properties.
I added two helpers. `calendar-sync--sanitize-org-property-value` trims the input and collapses any run of whitespace or newlines into a single space. `calendar-sync--sanitize-org-heading` composes that over the existing body sanitizer so `*` sequences also become `-`. The event-to-org function now routes the summary through the heading sanitizer and each property value through the property sanitizer.
I added regression tests across two files. `test-calendar-sync--sanitize-org-body.el` gets 4 new tests for the two helpers, covering newline flattening, leading-star replacement, structural-character flattening, and whitespace collapse. `test-calendar-sync--event-to-org.el` gets 2 new integration tests. A summary containing `\n** Hidden task` produces a single `* ` heading with the body inlined. A location containing `\n:END:\n* Not a real heading` collapses to a single property line with no extra `:END:` or heading injected.
515 calendar-sync tests pass together.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
`test-runner.el` stored `cj/test-focused-files` and `cj/test-mode` in single global variables. ERT tests loaded by `cj/test-load-all` accumulated in the same global registry across projects. Switching projects inherited the previous project's focused files and mode. `cj/test-run-all` then ran every loaded ERT test from every project visited this session.
I introduced a per-project state hash, `cj/test-project-states`, keyed by Projectile project root (or `default-directory` when not in a project). New helpers `cj/test--state-get` and `cj/test--state-put` route each read and write through that hash, so the focused-files list and the all/focused mode now live per project. The legacy public variables `cj/test-focused-files` and `cj/test-mode` are kept. They mirror the active project's state via `cj/test--sync-legacy-state` so existing modeline indicators and external code keep working.
I also tracked which project roots had loaded tests (`cj/test-loaded-project-roots`) and added two ERT-isolation helpers. `cj/test--current-project-test-names` filters ERT's full registry to tests whose source file lives under the current project root. `cj/ert-clear-tests` deletes ERT tests loaded from other known project roots, so a fresh project starts with only its own tests. `cj/test-run-all` now uses the filtered name list, and a `projectile-after-switch-project-hook` clears foreign tests automatically when you switch projects.
I added four regression tests to `tests/test-test-runner.el`: focus state isolated per project, mode isolated per project, `cj/ert-clear-tests` keeps the current project's tests and removes others, and `cj/test--current-project-test-names` returns only the current project's tests. Each test creates throwaway projects under the test temp dir and stubs `projectile-project-root` to switch contexts.
33 test-runner tests pass together.
|
