feat(nanoviews): hand the tracking key to the row

[dangreen]

A row is the row its tracker named. The key is the one thing about it that cannot change — a different key is a different row — and until now the loop kept it to itself:

for_($data, trackById)(
  ($row, $index) => {
    const { $id, $label } = record($row)

    return tr({ class: () => $isSelected($id()) ? 'danger' : '' })(
      td({ class: 'col-md-1' })($id),
      …

$id is a signal over a value that was known before the row existed and will never move. It costs a child signal, an effect and links, on every row, to read something the loop already had. The alternative — closing over the item the row was built from — is wrong the moment the row is reused for another item.

So the key comes third, as a plain value:

for_($data, trackById)(
  ($row, $index, id) => tr({ class: () => $isSelected(id) ? 'danger' : '' })(
    td({ class: 'col-md-1' })(id),
    …

Three details:

• as_ carries it through. as_(record, ($row, $index, key) => …) works the same.
• The static arm hands over the index, which is the key a list without a tracker is reconciled by. That arm used to pass each_ straight to Array#map, so a third parameter received the array — a trap for anyone who ever wrote one.
• trackById and trackBy became generic over what they return. They were typed => unknown, which would have made the key unknown and the whole thing useless.

Without a tracker the key equals the index and the two never disagree: the lookup is keyed by position then, so the row at position i is always the one whose key is i.

Measured

The row's key is only worth something to a component that takes it, so this was measured with the benchmark app rewritten to use it (a separate PR). 30 iterations per arm, two rounds with the arms alternating:

case	script before	script after	95% CI
01 create 1k	24.70	20.00 (−19%)	[−5.05; −3.95]
02 replace all	34.95	30.50 (−13%)	[−6.15; −3.60]
07 create 10k	205.05	186.25 (−9%)	[−24.70; −14.05]
08 append 1k	25.15	21.30 (−15%)	[−4.60; −3.15]

22_run-memory goes from 3.449 MB to 3.083 — 366 KB less for a thousand rows, with both rounds agreeing to the third decimal.

What leaves each row is the whole reactive apparatus that existed only to deliver the id: the proxy trap, a child computed, an effect and three links.

Cost

+11 B gzip (7558 → 7569). No pin moves. If #216 lands first its Average usage pin already covers the pair; whichever goes second wants its pins re-checked after the rebase.

Four tests: the key in the reactive arm and that it stays with a row across a reorder, the index as the key without a tracker, the same for a static array, and the key through as_. Lint, tsc --noEmit and 131 tests are green.

[codecov]

Codecov Report

❌ Patch coverage is 60.00000% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.38%. Comparing base (5942baf) to head (4e1ca5f).

Files with missing lines	Patch %	Lines
packages/nanoviews/src/flow/for.ts	50.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #217   +/-   ##
=======================================
  Coverage   85.38%   85.38%           
=======================================
  Files         140      140           
  Lines        3154     3155    +1     
  Branches      594      594           
=======================================
+ Hits         2693     2694    +1     
  Misses        332      332           
  Partials      129      129           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.