makeColumnsResizable measures once and credits colspan rows to column 0 (packets Details column too narrow)

Found while fixing #244 (PR #252); the reviewer confirmed it there. The default column widths of resizable tables come from a one-shot measurement that can go wrong in two ways. On the packets table, the Details column then becomes so narrow that some advert names are entirely on the hidden, clamped second line, so only the icon shows.

Cause

makeColumnsResizable() in public/app.js (master 310c501a, around line 2254):

It measures only once. It sets table.dataset.resizable = '1' and sizes the columns from the first ≤ 30 body rows present at that moment. If the first render is empty or nearly empty, the widths come from the header alone and stay that way. On packets, that happens when the default 15-minute window has few rows.
Colspan rows are credited to column 0. The measurement reads row.children[i] per column index. A row with one full-width colspan cell, such as a spacer, empty-state or group-detail row, therefore puts its whole content width into the first column. The extra width then stays with that column (expand on packets).

Observed in #252 at 1200 px (reported by the author and the reviewer, [T] there):

	expand	Details
fresh fixture (first render has 29 rows)	15 %	15 % (178 px)
aged fixture (empty first window)	43.5 %	5.4 % (64 px)

When colspan rows are skipped, the review measured Details at 301 px with the fresh fixture. CSS can't fix it: width: 32px !important on th.col-expand had no effect, because Chrome's auto layout still gives the leftover width to that column.

Saved widths in localStorage (meshcore-pkt-col-widths etc.) bypass the measurement. Users who resized a column once keep their own widths.

Expected

Skip rows with a colspan cell, or any row where row.children.length !== ths.length, when measuring.
Do not lock widths from an empty or near-empty body. Measure again once real rows exist, but at most once and only while the user has no saved widths, so there is no per-render cost.
Keep the behaviour for the other callers: nodes.js, observers.js, analytics.js, gps-sanity.js, new-nodes.js, node-changes.js and position-gaps.js.
Update the pinned.truncated guard in test-issue-1122-details-row-clamp-e2e.js in the same change. The test(packets): make the Details advert-link E2E independent of fixture age (#244) #252 review notes that it assumes today's narrow Details column and will fail once the widths are fixed.

Tests

A unit test (vm context) with a colspan row and with an empty first body.
An E2E on packets at 1200 and 900 px with an aged fixture, which checks the Details width and that long advert names show text, not only the icon.
A mutant check: put the colspan rows back into the measurement, and the test must fail.

Perf: the measurement runs once per table per page load (at most twice with the re-measure). There is no per-row or per-render work.

Plan — CS-Macmini #258

Branch codex/issue-258-column-widths from origin/master 3bb2cb89. Tests first (red), then the fix. One draft PR, "Relates to #258".

1. Skip colspan rows (`public/app.js`)

A new helper, columnMeasureRows(tbody, colCount, limit), returns at most 30 body rows that have exactly one cell per header cell and no cell with colSpan > 1.
It skips the vscroll spacers, "No packets found", the group-detail rows and any row with a different cell count.
makeColumnsResizable() measures only those rows plus the header.

2. Do not lock widths from an (almost) empty body

Threshold: COL_MEASURE_MIN_ROWS = 5 usable rows.
Below the threshold, the table gets provisional widths as today: the header plus the usable rows, with handles added. One MutationObserver on that tbody (childList only) then waits for real rows.
The callback is a cheap guard:
- if the table is gone or the storage key now holds saved widths, it disconnects;
- if tbody.rows.length is below the threshold, it returns.
Once enough usable rows exist, it disconnects first and then re-measures once. Before measuring it clears the th widths, so the old percentages do not feed back.
At or above the threshold on the first call, no observer is created. Today's path is unchanged, apart from the colspan skip.
Cost: at most two measurements per table per page load. The guard runs only while the table is still provisional; after that there is no per-render or per-row work.

3. Saved widths

A valid meshcore-*-col-widths entry is applied exactly as today, with no measurement and no observer.
If the user drags a handle while the table is provisional, onUp saves the widths as today, and the observer then disconnects without re-measuring.
The handles themselves are unchanged.

4. The 8 callers

packets: the first render with the default 15-min window and an aged fixture has only colspan rows, so it gets the deferred measure.
nodes, observers and gps-sanity, new-nodes, node-changes, position-gaps: they call after rendering rows. They change only when they render colspan rows or fewer than 5 rows first; then they also re-measure once later.
analytics: small static tables never mutate, so the observer never fires and costs nothing. They keep today's widths.
The measurement keeps today's table-layout and fitting logic. I'll extract the fit step into a pure helper without changing what it does.
Browser check: column widths of nodes, observers and one analytics table, master vs branch, with a fresh fixture.

5. `test-issue-1122-details-row-clamp-e2e.js`

With the fix, Details at 1200 px is about 300 px, so the pinned 21-char name fits.
Truncation is required only where it is real, at 375 px.
At 900 and 1200 px the pinned long name must instead show its whole width. This is a stronger check, and it cannot pass vacuously, because the pinned row must still be on screen.
I will base the exact rule on measured widths.

6. Tests and mutants

Unit, new test-issue-258-column-widths.js (vm context on the real app.js, with a minimal fake table DOM; registered in test-all.sh):
- a colspan spacer or empty-state row and a short group-detail row do not change any width;
- an empty first body is deferred and then re-measured once when rows arrive, after which the observer is disconnected;
- fewer than 5 rows are deferred;
- saved widths: no measure and no observer;
- saved widths that appear while the table is provisional: no re-measure;
- a full first body: no observer;
- idempotent re-call.
E2E, new test-issue-258-column-widths-e2e.js (registered in deploy.yml after the bug(packets): filter UX disaster — help panel overlaps table, toolbar chaotic, path chips spill rows Kpa-clawbot/CoreScope#1122 Details clamp line):
- Setup: packets at 1200 and 900 px. The browser clock is moved forward with Playwright page.clock, so that the fixture's effective age is 20 and 120 min. The offset is computed from the newest packet, so the CI age does not matter.
- Empty first render: the default 15-min window renders with no data rows; this is asserted, so the test cannot pass vacuously.
- After switching to a wide window (24 h at 1200, 3 h at 900):
  - Details has a minimum width;
  - expand stays narrow;
  - every on-screen advert link shows at least 12 px, and the long-name row shows its text, not only the icon;
  - a further window change does not move the widths (only one re-measure);
  - preset saved widths are applied unchanged.
Mutants (at least 3):
- colspan rows back in the measurement;
- no deferral, i.e. today's one-shot behaviour;
- re-measure on every mutation (no disconnect);
- re-measure despite saved widths;
- the 1122 guard reverted.
Suites:
- sh test-all.sh and node test-frontend-helpers.js;
- the bug(packets): filter UX disaster — help panel overlaps table, toolbar chaotic, path chips spill rows Kpa-clawbot/CoreScope#1122 Details clamp, bug(packets): layout chaos — path chips spill rows, +N popover overlaps, filter chips collide with toolbar, Saved dropdown bleeds through Kpa-clawbot/CoreScope#1128 layout and multi-viewport, test: 13 red orphan unit tests + 4 orphan E2E files left after #187; collapsed packet groups show an up-caret #189 and Follow-up to #248: Affinity Debug toggle broken + reversed carets; group-row aria-expanded on mobile #254 E2Es, plus the new one;
- all against a local Go server on e2e-fixture.db prepared as in CI, fresh and aged (DB shifted 20 and 120 min), and with the instrumented frontend if possible.

Gates

No hardcoded colours.
scripts/check-xss-sinks.sh --diff origin/master clean.
Fork guards stay 9 + 1.
One new E2E line in deploy.yml.

Customizer

COL_MEASURE_MIN_ROWS (5) is an internal layout heuristic, not a user setting, so it gets no customizer entry.

Merge pull request #268 from dborup/codex/issue-258-column-widths

Fixed by #268 (merged as 406f8481).

Merge branch 'master' into codex/issue-259-followups-255

Navigation Menu

Repository navigation

makeColumnsResizable measures once and credits colspan rows to column 0 (packets Details column too narrow) #258

Cause

Expected

Tests

Plan — CS-Macmini #258

1. Skip colspan rows (`public/app.js`)

2. Do not lock widths from an (almost) empty body

3. Saved widths

4. The 8 callers

5. `test-issue-1122-details-row-clamp-e2e.js`

6. Tests and mutants

Gates

Customizer

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

makeColumnsResizable measures once and credits colspan rows to column 0 (packets Details column too narrow) #258

Description

Cause

Expected

Tests

Activity

adminopenclaw8-sketch commented on Oct 5, 2026

Plan — CS-Macmini #258

1. Skip colspan rows (public/app.js)

2. Do not lock widths from an (almost) empty body

3. Saved widths

4. The 8 callers

5. test-issue-1122-details-row-clamp-e2e.js

6. Tests and mutants

Gates

Customizer

dborup commented on Oct 6, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

1. Skip colspan rows (`public/app.js`)

5. `test-issue-1122-details-row-clamp-e2e.js`