fix(tui): handle wide Bengali graphemes - #46112

Independently verified this on Linux with ghostty 1.3.1, which also fills in the "Terminal: Unknown" field from #46107.

Isolated comparison of the two versions, same sample, 24 column bordered box.

0.4.5, requesting unicode-wide, silently resolves to unicode:

│বাংলা ভাষায় কথা বলি বাংলা       │

0.5.9, requesting unicode-wide:

│বাংলা ভাষায় কথা বলি    │
│বাংলা                  │

0.5.9 on the plain unicode profile still overflows, so this fixes the drift only when the wide profile is negotiated, rather than correcting widths unconditionally. That negotiation does happen here: the native lib carries ghosttyWideGraphemeWidths and term_from_xtversion, and ghostty answers the XTVERSION query.

Confirmed end to end as well: running the TUI from this branch in ghostty 1.3.1, the Bengali sample stays inside the panel and the border stays aligned. Word internal spacing from ghostty's two cell advance looks the same as it did on 0.4.5, so that part is not a regression.

Bump checks on a local branch off dev:

• lock diff touches only the opentui packages plus bun-ffi-structs 0.2.4 to 0.3.1
• packages/tui: 193 pass, 1 skip, 0 fail
• bun turbo typecheck from root: 30/30 successful

One suggestion: consider dropping packages/tui/test/bengali-width.test.tsx. It renders a bare box and text with capabilities forced, so no opencode code is under test, and asserting renderer.widthMethod checks the dependency rather than this repo. Because it forces the profile, it would keep passing even if detection broke for every real terminal, which is the part the fix actually hinges on. The frame assertions are captured output with exact trailing space counts, so any future width table change breaks them without indicating whether the new output is correct.

Also, 0.5.10 is published now.

Clarification on my numbers above: those were from a bump-only branch, without this PR's test file, which is why I saw 193 pass where you report 194. I have since run this branch directly: bengali-width.test.tsx passes on its own, and the full packages/tui suite is 194 pass, 1 skip, 0 fail. Your numbers reproduce exactly. My note about the test was about what it covers, not whether it works.

I tested with Hindi also. same problem, same fix.

Mentioning @nexxeln because he is maybe from India.

Cross-linking another issue that may benefit from this OpenTUI upgrade: #36454.

OpenCode on OpenTUI 0.4.5 still emits the same shutdown warning:

Code highlighting failed, falling back to plain text: warn: TreeSitter client destroyed

Upstream tracked that teardown race in anomalyco/opentui#1319 and closed it as completed after renderer teardown ordering changed. The reproduction above is on macOS with OpenCode 1.18.30.

Would be useful to verify #36454 against this PR / current OpenTUI before merge. If the warning disappears, this PR may address both renderer issues; if not, that gives a clear remaining upstream gap.

No claim yet that the bump fixes #36454 — just flagging the likely overlap.

hello @nexxeln, it's urgent

This branch has been upgraded to V2. We're no longer taking PRs for V1. If you think this is still relevant, please port it over and open it for V2.

— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊