[rehearsal] Rebase fork onto v0.0.46-nightly.20261007.2761

[aidenwboudr]

Rehearsal — do not merge. t3-fork-watch rehearsed moving the fork onto upstream v0.0.46-nightly.20261007.2761 (a nightly, not a stable release).

Decision: rebase (not retire)

Upstream	Status	In v0.0.46-nightly.20261007.2761?
pingdotgg#15067 Mermaid diagrams	merged (5e35272)	yes, dropped from the fork patch
pingdotgg#14574 LaTeX math	open	no, carried (current PR head)
pingdotgg#16198 math labels in Mermaid	open	no, carried
pingdotgg#16221 Mermaid renders while streaming	open	no, carried
pingdotgg#16249 copying a diagram copies its source	open	no, carried

Upstream doesn't render math on its own and still has the backtick bug (It costs $5 and $10, and \$HOME`` — the fix is fork-only and pingdotgg#14574's head doesn't have it).

How it was rebuilt

Instead of re-applying the squashed fork commit, the branch starts at the tag and cherry-picks each open PR's current head commits (-x), then three fork-only commits: the backtick fix + tests, math on by default, FORK.md + scripts/fork-build-linux.sh. All the PR heads are the same commits the fork carried on 2026-10-07; none gained new fixes since.

Conflicts resolved:

• ChatMarkdown.test.tsx (fix(web): render completed Mermaid blocks while streaming pingdotgg/t3code#16221): upstream added a bare-anchor describe block where fix(web): render completed Mermaid blocks while streaming pingdotgg/t3code#16221 adds its streaming block; kept both.
• MermaidDiagram.tsx / markdown-clipboard.ts (fix(web): preserve Mermaid source when copying selected text pingdotgg/t3code#16249 vs feat(web): opt-in LaTeX math rendering in chat pingdotgg/t3code#14574/fix(web): render completed Mermaid blocks while streaming pingdotgg/t3code#16221): same resolution the fork used before. The wrapper keeps chat-markdown-mermaid plus data-markdown-mermaid/data-markdown-copy, and one clipboard branch handles [data-markdown-math], [data-markdown-mermaid].

No lockfile changes beyond the PR commits (pnpm install was clean).

Results

• pnpm install: clean, no lockfile changes.
• apps/web pnpm run typecheck: pass.
• apps/web pnpm run test: pass, 461 files / 6166 tests.
• scripts/fork-build-linux.sh 0.0.46-nightly.20261007.2761: pass, wrote release/T3-Code-0.0.46-nightly.20261007.2761-x86_64.AppImage (161 MB).
• render-check.mjs: exit 0. {"ok":true,"katex":4,"katexErrors":0,"diagrams":2,"diagramErrors":[],"homeIsCode":true,"pricesAreText":true,"screenshots":"/home/boudreau/.local/state/t3-fork-watch/runs/rehearse-v0.0.46-nightly.20261007.2761/render"}
• I looked at the dark and light screenshots myself: inline and display math typeset, both Mermaid diagrams render and are themed, $5 and $10 stay prose, and $HOME stays a code span. The screenshots are kept on the homelab in the run folder (~/.local/state/t3-fork-watch/runs/rehearse-v0.0.46-nightly.20261007.2761/render/).

Found in rehearsal: the update feed for a nightly version

Because the version string has a -nightly… prerelease part, electron-builder put the update channel in release/nightly-linux.yml. The release/latest-linux.yml sitting next to it is stale, left over from the 0.0.45 build (version: 0.0.45). If the ship step uploaded release/latest-linux.yml for a nightly tag as written, the laptop updater would read a 0.0.45 manifest. For a stable tag (e.g. 0.0.46), the build writes latest-linux.yml and this doesn't come up. If a nightly is ever shipped, upload the fresh nightly-linux.yml as latest-linux.yml and check the version: line first.

🤖 Generated with Claude Code

[aidenwboudr]

Rehearsal for t3-fork-watch: rebase, 6166 tests, build and render check all passed. Closing; real runs only target stable releases.