Repository navigation
Conversation
Since #451 (2.16.0) the sidebar's scrollbar thumb is tinted with the sidebar's own text colour so a long list visibly scrolls. The rule asked for that colour at `opacity: 0.3` (0.5 on hover), but scrollbar pseudo-elements ignore opacity, so the thumb has been painted at full text colour: a bright bar beside every sidebar that overflows, unchanged on hover. The strength now lives in the colour itself: `color-mix(in srgb, currentColor 30%, transparent)` at rest and 50% under the pointer, with the plain currentColor line kept first for a browser without color-mix. The thumb stays discoverable, which is what #451 wanted, without competing with the notes list. Measured in the built app over CDP (isolated stores, an 80-note vault so the sidebar scrolls), sampling the thumb's pixels. Gruvbox dark: text 212,190,152 over 22,23,24, the thumb went from 209,191,157 to 78,73,64 and 115,106,90 on hover (a 30% and 50% blend). Gruvbox light: text 80,73,69 over 229,226,201, from 79,73,70 to 184,180,161 and 154,150,135 on hover. scrollbar-style.test.ts fails on the old rule: no scrollbar part may set opacity, and the sidebar thumb uses the 30% and 50% mixes.
…rsor Cmd+L (Ctrl+L on Windows and Linux) toggles the checkbox on the cursor line, and in Vim mode it works from normal, insert, and visual mode without changing modes. The manual only listed it as "Toggle checkbox" in the general shortcut table and in a concepts card titled "Any line becomes a checkbox", so someone looking in the Vim & Ex section for a way to mark a task done found nothing. Vim & Ex gains a "Check off the task under the cursor" row, and its key label follows an editor.toggleCheckbox rebind the way gd and gD do. The card is renamed "Check off a task with ⌘L" (the shortcut row that points at it follows) and says how it behaves in each Vim mode. Behavior verified in the built app over CDP: normal mode stays normal, insert stays insert, Vjj toggles exactly the selected lines, and one u undoes them. The website mirror landed in ZenNotes/website 2da9d8d.
…ATH probe Julie (Discord, 2026-10-09) runs ZenNotes from a terminal on Linux, and clicking CLI in Settings made it "crash and stop responding". To answer "is zn on your PATH?", and to find node, npm, rg and fzf, the main process asks the user's shell for its PATH, interactive login shell first (`$SHELL -lic`, #634). An interactive shell does job control on its controlling terminal, and a ZenNotes started from a terminal shares that terminal with every shell it starts. Reproduced on Linux with the 2.64.0 tarball, two ways: - zsh, with an rc file that prompts on /dev/tty: each probe took the terminal's foreground and waited there for an answer nobody knew was being asked for. Interactive shells ignore SIGTERM, so the five-second timeout ended none of them, Settings > CLI sat on "Checking install status..." for good, and every visit started more of them. - bash, with ZenNotes started in the background (`zennotes &`): the startup probe, finding its process group outside the terminal's foreground, sent SIGTTIN to that group, which is ZenNotes' own. The main process and the renderer stopped and no window ever came up. Finder, Dock and app-menu launches have no terminal, which is why this only shows when ZenNotes is started from one. The probe now runs in a session of its own (`detached`) with stdin empty. Without a controlling terminal there is no job control to do: nothing outside the probe is signalled, and /dev/tty fails at once instead of blocking. A timeout SIGKILLs the probe's whole process group, so a stuck rc file costs five seconds per shell instead of the page, and callers that ask at the same time share one probe instead of starting three. What it deliberately does not do: change the interactive-first order of #634 or the timeout. rc files still run, only without a terminal to talk to. Verified with a Linux tarball of this change on the same setup (Ubuntu 26.04, app started under a pty, isolated stores): with the prompting rc the page loaded in about a second, from one probe with no terminal and its own process group, and the terminal stayed with ZenNotes; with an rc that hangs, each stuck zsh was killed with its child at five seconds, bash answered and the page loaded; `zennotes &` under bash opened its window and the CLI page. login-shell-path.test.ts fails against the old probe: a shell that ignores SIGTERM (it and its child must be gone), the probe's own session, and three callers sharing one shell.
…the table Tornado300 (Discord, 2026-10-09) selected every row of a rendered table in Vim mode and could copy only the content, never the table. Reproduced in the built app (Vim on, isolated stores), two ways: - V j j y in the table, then p on the line below it, pasted the cells as tab-separated text into the middle of that line: the cell cursor yanked every selection as TSV, charwise. - Cmd+C on the same selection reached the cell cursor as a plain `c`, Vim's change operator. The clipboard stayed as it was, the selected cells were emptied into insert mode, and the Escape that followed wrote the blank table back to the note and to disk. Cmd and Ctrl chords in a cell now go to the app instead of running as Vim letters, except Ctrl+R (redo), Ctrl+Alt (AltGr) characters, and whatever the pane moves (Ctrl+h/j/k/l) are bound to. In a visual selection Cmd+C (Ctrl+C on Windows and Linux) copies it without changing it, and Cmd+X cuts it. A Shift+V selection yanks as Markdown table rows, linewise (one that takes in the header brings its divider row too), so p in the note pastes a working table. A character or block selection still yanks TSV, which is what a spreadsheet wants. The cell menu (right-click / m) gains "Copy table as Markdown" and "Duplicate table", and the manual's "Render tables in live preview" entry says how to copy one. What it deliberately does not do: change how a yank reaches the system clipboard (still the "Sync clipboard with Vim registers" setting), or change table editing with Vim off. Verified over CDP on the built app: V j j y then p pastes the table below it as four Markdown lines; Cmd+C puts those rows on the clipboard and the Escape after it leaves every cell, and the file, intact; the menu copies the table and duplicates it below (two table widgets). cm-table.test.ts gains seven tests: the five yank and copy tests fail against the old code, and the two menu tests drive items that did not exist.
…e it Tornado300 (Discord, 2026-10-09) found the Vim motions and keybinds dead in the Tags overview. The view answers its keys only while the Tags panel holds focus (#412). The editor pane's capture handlers claim the editor panel for any click or focus inside the pane, so the first click on a heading or a tag chip handed the panel to the editor and j/k stopped reaching the view. Tasks and Atlas already win the panel back on their root element; the Tags view now does the same, and since inner capture handlers run after the pane's, its claim is the one that sticks. Pane navigation had the same gap: with the Tags view as the active tab, the panel list still named the editor in that slot, so Ctrl+W h/l and Alt+h/l had no way to land on the view. It names the Tags view now, as it does Tasks and Atlas. Verified over CDP on the built app: before, after one click on a heading, j left the cursor where it was; after, j and k move after heading and chip clicks. vim-nav.test.ts gains the panel-order case, which fails against the old code.
…ags step Tornado300 (Discord, 2026-10-09) wanted every note without tags moved into a folder to check. Saying "carries none of these tags" took one not-tagged step per tag, and "carries no tag at all" could not be said. `tag`, `tagged` and `not-tagged` now take a comma-separated list and match any tag in it, so `tagged #book, #article` is an allow list and `not-tagged #someday, #reference` a block list. `no-tags` keeps the notes that carry no tag at all, frontmatter or inline (a `#` inside inline code is not a tag, as everywhere else in the app). The canvas labels, the step reference, completion (the entry under the caret, `#a,#b` included) and the inspector field (`book, article`) all speak the list, and the file always saves it as `#a, #b`, a repeated tag dropped whatever its case. The manual's Workflows card shows the list form and counts thirteen filter steps. What it deliberately does not do: read tags separated only by spaces as a list. Before lists, `not-tagged #a #b` ran as `not-tagged #a` with a warning, so reading the spaces as a list now would quietly change what an existing `... | trash` touches. It is an error that names the comma form, and an errored statement plans no writes. A build without lists rejects the comma form the same way, so a file shared across versions never runs wider than written. The step is not called `untagged`: a step kind wins over a wire of the same name, and `untagged` is a natural wire name. Verified over CDP on the built app (Vim on, isolated stores): `all | no-tags | move review` moved exactly the three untagged notes (one with `#notatag` only in inline code) and left a note whose only tag is in its frontmatter; `tagged` and `not-tagged #book, #article` wrote the right sets; typing ", someday" into the inspector autosaved `#book, #article, #someday`, Backspace took it back to `#book, #article` without eating into the previous tag, and switching steps re-seeds the field; the ? reference lists no-tags. 22 new tests: 15 fail against the old code, the rest guard what must not change (a single tag binds as before, no completion after a bare space).
New Task from the palette, `:newtask`, and now a phone's Action Button (zennotesandroid#101) wrote the task file and showed nothing. The task appears only in the Tasks view, so unless that was on screen, the only way to know it worked was to go and look. A task created anywhere but the Tasks view now raises "Task added" with an Open action that opens the new task note. On the Tasks view the new row is the confirmation, so it stays quiet there. The toast host gains a `data-toast-host` hook: the phone shells pin their floating button to the corner toasts use, and lift toasts above it by that hook. Verified over CDP on the built app: `:newtask` from a note shows the toast, and Open opens the task note and dismisses it; the Tasks view's "+ New task" shows none. store-new-task.test.ts covers both and a cancelled prompt.
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
ZenNotes 2.65.0.
Features
tag,taggedandnot-taggedtake a comma-separated tag list (any of them), and a newno-tagsstep keeps notes with no tag at all. Space-only separation is an error naming the comma form, since old builds rannot-tagged #a #basnot-tagged #a. (Tornado300, Discord)Fixes
c, which emptied the cells and saved the blank table. (Tornado300)ToastHostgainsdata-toast-hostfor the phone shells. (zennotesandroid#101)opacitynever applied).Cycle verification: every fix reproduced in the built app over CDP on an isolated profile, then checked again after; the terminal freeze on Linux (Ubuntu 26.04) with the 2.64.0 tarball. New tests fail against the old code where they pin a behavior change (5 of 7 table, the Tags panel order, 15 of 22 workflow, 3 probe tests).
Local gates on
21ebc827: typecheck 8/8 and test:run 6/6 with no turbo cache (shared-domain 1,921, app-core 3,140, quicklook 15, desktop 1,055),npm audit --omit=dev --audit-level=highclean (15 low or moderate remain in dompurify, fast-uri, ip-address, katex, smol-toml and sprintf-js).npm run packsigned the app on the first try, deep strict codesign OK, the pack log shows electron-builder's "skipped macOS notarization" next to the hook, bundledzn v0.6.3inResources/zn-cli(drwxr-xr-x); packaged launch check: page target in 1.4 s reporting 2.65.0. Smoke suites on the built app: vim-editor, sidebar-vim and editor-improvements passed on their first run.