fix(web): prevent long links from creating per-character elements#26843

Approvability

Verdict: Approved at 3f8d57f

Macroscope's review found this PR approvable — Long external-link labels now use CSS wrapping instead of per-character elements, reducing rendering overhead. The change is presentation-only and preserves link text and destinations.

Notes:

• All code in this push has already been reviewed. Approvability was decided on eligibility alone.

^{You can add or adjust custom eligibility rules. Learn more.}

Review in Change Stack →

No actionable comments were generated in the recent review.

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

Verified in a real dev environment

verified as-is at 9bc74123d71a919c384b0b6da8c3c21b43682b79. no source changes or pushes.

• installed and ran github/main (32ecd71726) in a separate worktree, then stopped its captured process group before starting the exact pr head. both used isolated development state.
• used the web app through repo playwright and /usr/bin/google-chrome. created a project for the real t3code git repository and submitted real codex gpt-6.1-sol turns through the composer on both revisions. no mocks, replay provider, fixtures, or demo data.
• preserved the actual main provider response in the pr development state for matching before/after data. the longer plain label has 2,140 characters: its descendants fell from 2,136 to five. the shorter 156-character label also has five descendants on the pr. all five link labels and hrefs are byte-for-byte unchanged, and bold details survives.
• the formatted link made the 728px main table scroll across 16,542px. on the pr, both visible and scroll widths are 728px. a fresh provider turn also rendered plain/formatted links live without reloading. at a 768px dark web viewport, its table stays within 456px with no horizontal overflow.
• nearby behavior: the short #26843 label stays intact; Copy Link returns the complete formatted-link destination (1,948 characters in the matching response, 276 in the fresh response); Open in system browser opens https://example.com/, with title Example Domain.
• inspected the screenshots and trimmed recording. the uploaded video plays to the end and supports seeking in the t3 browser player.

runtime/check transcript:

before, the actual main provider response:

main: long plain link and overflowing formatted link

after, the same response on the pr:

pr: same links wrap with five descendants and preserved bold text

Not verified

• native electron on linux, macos/windows shells, native ios/android, and remote relay/tunnel connections were not exercised. this diff changes the shared web/desktop renderer; mobile has a separate renderer. other providers were not exercised.
• no click-latency or cpu improvement is claimed. the real-app measurements above establish dom growth and wrapping only.
• github ci is not green: the existing run has one failed aggregate Check, 18 canceled jobs, four skipped, and six successful. its Check log reports canceled prerequisite jobs. the independent blacksmith checks above passed; no ci reruns or unrelated changes were made.
• on main, the first provider response persisted but became visible only after reload. the fresh pr turn rendered without reload. this verification does not establish the cause of that separate main behavior.

verification performed through codex in t3 code using the takeover and file-upload skills.

merged github/main into the existing head without conflicts at ba9cebf90a3498f58df5192f1db7e17f8c8c9ac7 and pushed without force. no source edits; the pr remains two files, +64/-12. both existing bot findings are addressed in the contributor commits and their threads are resolved.

blacksmith verification on this head: vp run check:changed passed format, lint, both knip passes, affected typechecks, browser regression tests, and react compiler coverage. the focused MarkdownExternalLink and ChatMarkdown browser suites passed all 13 tests. two independent read-only reviews approved this head with no findings. existing real-app before/after evidence remains above. github checks are pending.