2.7.7 is ready to cut — and it should be the first signed Windows build

[PathGao]

Summary

14 merges (18 pull requests) since v2.7.6 went out on 2026-08-27. All of them
are fixes and features this time — no CI, no refactors — and six of them close
issues people reported.

master is green: 1627 tests pass (1029 npm test, 434 vitest, 164 cargo).

One thing is worth reading before you cut, and it is the same thing as last
time: Windows signing (#562). This release is the one it should land in,
and the section below says why and what it takes.

---

Sign Windows first, then cut

Where #562 stands: you asked on 26 Aug how to sign both the portable and the
installer through SignPath, and the answer is in the thread — one artifact
configuration signs all four exes from one ZIP. Under it is the list of five
things to do in app.signpath.io and the four strings I need back (organization
id, project slug, signing policy slug, artifact configuration slug). The
workflow change on my side is one pull request, written the day the strings
arrive. Nothing else in #562 blocks this.

Why 2.7.7 rather than 2.7.8:

• Do not clear the temporary files #722 is waiting on it. A user on Windows has an empty file re-created
  at the old path 0.2–2 s after every rename, in document folders and next to
  Markpad's own settings. The log put it outside Markpad — something on that
  machine treats an unsigned process specially — and I closed it telling them
  to try again on the first signed build. If 2.7.7 ships unsigned, that is
  one more release they cannot test.
• The updater serves it too. Every Windows install that auto-updates gets
  whatever 2.7.7 is. Unsigned, that is one more SmartScreen dialog (SECURITY: the windows installer of the new version has a trojan virus #334,
  Windows portable release flagged by Microsoft VirusTotal heuristic detection #466) delivered to the 91% of downloads that are Windows. Signed, it is
  the release where that dialog stops.
• The release notes change with it. The SmartScreen [!NOTE] in the
  release template comes out in the same pull request that wires the
  signing, so it does not have to be remembered separately.

If SignPath Foundation's approval takes longer than a week, cut 2.7.7
anyway. Everything below is finished and tested; nothing in it should wait
on a third party's queue. But if the approval is days away, the signed build
is worth those days.

macOS is unchanged: signed with the identity #707 added in 2.7.5, not
notarized, and not part of this ask.

---

Features

• An imported theme colours the preview, not only the editor (An imported theme colours the preview, not only the editor #736) —
  closes Custom CSS #682. A VS Code theme's tokenColors used to stop at the splitter;
  the seven constructs it names now reach the preview through CSS variables.
  A theme that names none of them renders exactly as before.
• Tabs that share a file name say which folder they are from (feat(tabs): a tab that shares a file name says which folder it is from #728) —
  ref Multiple tabs with same filename get folder slug appened #727. Each carries its containing folder, dimmed, and one more folder
  only when that one is still shared — and only while the name is actually
  ambiguous. Settings › Appearance ›
  Show Folder for Duplicate Names, on by default.
• \ce{} chemistry renders (mhchem registers on the KaTeX the preview renders with #747) — closes Feature Request: Support mhchem syntax \ce{} for chemical equations #745. mhchem was already
  loaded, but onto the CommonJS copy of KaTeX rather than the one the
  preview renders with, so $$\ce{...}$$ was a parse error. The built app
  now carries one KaTeX instead of two. The label on a reaction arrow is no
  longer clipped (The label on an mhchem arrow is no longer clipped #750).

Fixes

• Typing with an IME no longer makes the line sink and snap back (Typing with an IME no longer makes the line sink and snap back #746) —
  closes Bug-typing jitter #724. Monaco's hidden textarea held a page of text during
  composition and WebKit scrolled it 3 px short on every keystroke. A
  build-time patch makes the overlay hold one line; proposed upstream as
  Editor textarea: short content unless a screen reader is attached (fixes IME jitter in WebKit) microsoft/vscode#333909 and removed when a Monaco release carries it.
• Copy works on X11 (What X11 copies outlives the command that copied it #731) — closes Copy doesn't work in the editor (paste works) #729, reported on Debian 13 XFCE.
  The clipboard handle was dropped when the command returned, and on X11 the
  selection dies with its owner unless a clipboard manager happens to be
  running. Cut was the expensive case: the text was deleted and nothing was
  saved.
• The outline, the editor and the preview agree in split view (A click on the outline in split view moves the editor too #748,
  The editor lands where the preview does, and the outline follows from the same line #751) — closes Bug: TOC sidebar, source editor and rendered preview cannot sync scroll position mutually #744. A click on the outline moved only the preview when
  split view was entered from reading mode; following the editor, the
  outline was one heading behind.
• A tab switch shows the document it already rendered (fix(tabs): a tab switch shows the document it already rendered, instead of rebuilding it #730). One
  preview host per tab, all mounted, only the active one displayed — the
  same shape as the Monaco models. Re-showing a tab changes nothing, so the
  visible reload, the motion and the reading position that moved after it
  had been restored are gone together. Four follow-ups landed with it
  (fix(tabs): the four tab-switch follow-ups (#732, #733, #734, #735), retargeted to master #737: fix(tabs): the first read of a large file measures its own tab, not the one on screen #732 fix(tabs): the preview's in-page back/forward stacks belong to the tab, not the window #733 A tab dragged into another window opens where its reader left it #734 fix(tabs): a tab dragged out of an editor arrives where the reader left it #735) — a large file's first read measures its own
  tab, the in-page back/forward stacks belong to the tab, and a tab dragged
  between windows arrives where its reader left it, in either direction.
• A background tab in edit mode reopens where the reader left it (A background tab in edit mode reopens where the reader left it #739),
  across a window-state snapshot and a cross-window transfer.
• A cold start restores the reading position against the layout it ends
  up with (fix(preview): a cold start restores the reading position against the layout it ends up with #740). KaTeX, highlight.js and Mermaid land after the first
  document, and the restore ran before them — measured off by up to 812 px on
  the sample documents, now within 0.8 px.
• A theme change re-draws Mermaid diagrams again (fix(preview): a theme change re-draws the diagrams it has never re-drawn (#741), retargeted to master #743, with fix(preview): a theme change re-draws the diagrams it has never re-drawn #741 and
  One writer decides which Monaco theme the editor wears #742). It has not since perf(preview): replace the blocks that changed instead of the article #632 (2.7.4): the diagram was found by pre code,
  and a drawn diagram has no <pre> left. The theme it was drawn with is
  now kept beside it.
• A wrapped line starts where it should, not one space in (A wrapped line starts where it should, not one space in #726). The
  soft-line anchor from fix(preview): make split-view scroll sync line-accurate #541 sat between <br> and its newline, which turned
  the newline into a space.

---

To cut it — @alecdotdev

Once signing is wired, or a week from now, whichever is first:

npm run release 2.7.7

then commit, tag, push, dispatch Build and Release, and publish the draft
once the assets look right. Publishing is what fires Chocolatey and the Snap
Store. With SignPath wired the Windows job pauses until you approve the
signing request; nothing is uploaded before that.