fix(terminal): continuation fragments of wrapped references are clickable - #175
fix(terminal): continuation fragments of wrapped references are clickable#175ghackett wants to merge 3 commits into
Conversation
…able Live testing of the stitcher showed clicks on the second half of a hard-wrapped path — the half holding the file name — did nothing, ever. Two causes: - A continuation fragment often contains no slash (`o.py:7)`), so FILE_PATTERN never matches it and the click handler bailed before the stitcher could run. When nothing under the pointer matches, the raw whitespace-delimited token at the clicked cell now stands in as the candidate; the stitcher's geometry gates and existence check keep prose clicks inert. - Joins were matched with an anchored match, which can only shed trailing junk. The emitter wraps whatever surrounds the path along with it, so fragments arrive glued to prefixes — `Read(/a/b/c` is Claude Code's own tool-call format — and any leading junk killed the join. Matching now searches the join, with span guards so a hit must overlap the clicked fragment itself. Verified live (headless, scrollback, pixel scrolling): both fragments of `⏺ Read(collins/fo` + `o.py:7)` resolve with the line suffix, a head token glued to `Update(` resolves, plain references still resolve, and prose tokens at row edges stay inert. 5 new unit tests, 64 pass total. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEJjBkzLLfVtVk6HhzFEn
Live testing in the app showed every wrapped-reference click still dead. Ground truth from running the real claude CLI inside a headless VTE: the CLI's repaint-style renderer leaves VTE's internal ring a full page away from what the adjustment describes (adjustment says rows 0-40, content lives at ring rows 42-80), so every adjustment-derived get_text_range read returned an empty row and the stitcher never saw any text. Feeding text into a terminal keeps ring and adjustment aligned, which is why the earlier probes never caught it. Row texts now come from the visible-screen snapshot (get_text_format), indexed by screen row straight from the click's y — no ring coordinates involved. Soft-wrapped rows come back joined, so they are re-chunked at the column count; pixel scrolling's partial top row is compensated by the adjustment fraction. Trade-off: neighbour rows scrolled out of the viewport are no longer reachable, which only matters for a wrapped reference straddling the viewport edge. Verified against the real CLI: a click map over every cell of a wrapped absolute path (both the plain and the Read(...):42 form, echoed at 64 columns) resolves on all four rows, box borders and prose stay inert, and a tilde path resolves to its directory. The feed-based probe (scrollback, pixel scrolling) still passes all 8 scenarios; 24 unit tests pass. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEJjBkzLLfVtVk6HhzFEn
|
Follow-up from live testing: wrapped clicks were still dead in the app even with this branch. Ground truth from running the real claude CLI inside a headless VTE found the root cause — the CLI's repaint-style renderer leaves VTE's internal ring a full page away from what the adjustment describes (adjustment claimed rows 0–40 while the content sat at ring rows 42–80), so every adjustment-derived Fixed in d5d86bd by reading row texts from the visible-screen snapshot ( |
A wrap-position sweep (every split of a long path, clicking every cell of both rows) found one failing shape: the wrap landing right after a character FILE_PATTERN refuses to end on — a head row ending `…/collins/` or `…/collins/.`. The match candidate arrives without the shed run, so the end-of-row gate never opened, stitching was skipped, and direct resolution then opened the *parent directory* instead of the file — the one hazard stitch-before-direct exists to prevent. The gate now also opens when the row ends with the candidate plus a run of exactly those shed characters, and the run seeds the downward join so `…/collins` ⏎ `linkpatterns.py` stitches back to `…/collins/linkpatterns.py`. Prose punctuation hanging off a reference at a row end opens the gate too, but its joins resolve nowhere and the click falls back to direct resolution as before. 3 new unit tests (27 pass); the 143-case sweep and the 8-scenario feed probe pass clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEJjBkzLLfVtVk6HhzFEn
|
Second live-testing follow-up: continuation fragments containing a slash sometimes still misfired — clicking the head half opened the parent directory. A wrap-position sweep (all 143 viable splits of a long path, plain and Fixed in 99d70f9: the gate also opens when the row ends with the candidate plus a run of exactly the shed characters ( |
|
@claude review |
|
Claude finished @ghackett's task in 4m 44s —— View job Review
I read the full diff against Correctness — no bugs found. I hand-traced
The Minor, non-blocking nits:
Verification note: I wasn't able to execute the test suite directly in this environment (Bash tool calls require interactive approval that isn't available here), but No changes requested — this looks solid and ready as-is. |
1 participant
Stacked on #174. A live testing pass there found that clicks on wrapped file references did nothing — and the failing half was the continuation fragment, the one holding the file name, i.e. exactly where you'd naturally click. Verified against a live headless terminal: with PR 174 as-is, only head-fragment clicks ever worked in Claude Code's real output format.
The two causes
Slashless continuations were never candidates.
FILE_PATTERNdeliberately requires a slash, and a wrap point usually falls inside the basename, so the continuation row showso.py:7)— no VTE match under the pointer, and the click handler returned before the stitcher was consulted. Now, when nothing under the pointer matches, the raw whitespace-delimited token at the clicked cell stands in as the stitch candidate (token_at_column, unit-tested). The stitcher's geometry gates plus the existence check keep Ctrl+clicks on ordinary prose inert — verified live for row-edge prose tokens and empty areas.Leading junk killed the join. The emitter wraps whatever surrounds the path along with it, so fragments arrive glued to prefixes —
⏺ Read(/a/b/cis Claude Code's own tool-call format, and the row-above token it contributes isRead(/a/b/fo, not/a/b/fo. The anchoredmatchcould only shed trailing punctuation, so every up-stitch through that format failed even when the continuation did contain a slash. Joins are now matched with a search over the joined string, with span guards requiring the hit to overlap the clicked fragment itself — one mechanism that sheds junk whether it sits in a contributed token or in a token-derived candidate, and that still refuses to resolve text the click never touched.Verification
Read(-glued head, a slashless token candidate keeping its:7suffix, a glued-prefix token candidate, the span guard (a junk-leading candidate must not resolve the row above's path), andtoken_at_columnedge cases. 64 pass total.⏺ Read(...)/⏺ Update(...)wrapped output): all 8 click scenarios pass — both fragments of a wrapped reference resolve with the line suffix, a head token glued toUpdate(resolves, plain unwrapped references still resolve, and prose tokens at row edges plus empty areas stay inert. Before the fix the same probe failed 4 of 8.🤖 Generated with Claude Code
https://claude.ai/code/session_01AXEJjBkzLLfVtVk6HhzFEn