fix: autosize never measured autoSize.colValueArray (#934)

[6pac-ai]

Fixes #934, reported by @doornik in Nov 2023 with the correct diagnosis and a working fix.

The bug

getColContentSize() handed the bare colValueArray to getColWidth() cast as any:

maxColWidth = this.getColWidth(columnDef, gridCanvas, autoSize.colValueArray as any);

getColWidth() measures by walking rowInfo.startIndex..rowInfo.endIndex. On a bare array both bounds are undefined, undefined <= undefined is false, so the measuring loop never ran and the column was sized to an empty cell. It didn't throw only because the loop never executed — rowInfo.getRowVal is undefined too and would have thrown on the first iteration.

It is wider than the original report

The issue was raised as affecting the reporter's own use of colValueArray. But under the default ColAutosizeMode.ContentIntelligent, the grid sets colValueArray itself for three data types:

if (colDataTypeOf === 'boolean') { autoSize.colValueArray = [true, false]; }
if (colDataTypeOf === 'date')    { autoSize.colValueArray = [new Date(...)]; }
if (colDataTypeOf === 'moment')  { autoSize.colValueArray = [moment(...)]; }

So autosize silently failed to measure content for every boolean, date and moment column, falling back to header width.

Measured effect

Instrumenting getColWidth, with ContentIntelligent + IgnoreViewport:

column	rowInfo received	before	after
Date	bare Array, bounds undefined	8px	427px
boolean	bare Array, bounds undefined	8px	34px
string (control)	proper RowInfo	50px	50px

8px is the width of an empty cell.

The change

The values are wrapped in a proper RowInfo. Two details beyond the reported fix:

• colIndex is included — getColWidth() passes it to the formatter as the cell index, so without it formatters receive undefined.
• as any becomes as RowInfo — that cast was the root cause. It silenced exactly the shape mismatch that would otherwise have been a compile error, which is how this survived two years.

Test

cypress/e2e/quirk-autosize-colvaluearray.cy.ts, self-hosting via cy.intercept (nothing added to examples/), following the existing quirk-* specs.

It asserts on autoSize.contentSizePx — the measurement itself — rather than the final column width, because the width distribution that runs afterwards does not discriminate between the two behaviours (I checked: the date column ends up at the same final width either way). Header names are deliberately one character, since getColContentSize() returns max(headerWidthPx, measured) and a wide header would mask a failed measurement.

The precise signal: the bug collapses that max() to exactly headerWidthPx, so "content beats its own header width" is what separates the two. The spec also checks its own preconditions (both columns really do take the colValueArray path) and keeps a plain string column as a control so the main measuring path is not broken.

Verification

• Fails on the unfixed build (fix stashed and rebuilt), both retry attempts: expected 21 to be above 21 — the date column collapsed to its 21px header. Passes with the fix.
• tsc --noEmit, eslint and the prod build all clean, exit 0.
• Full Cypress suite: 652 tests, 650 passing, 2 pending, 0 failing.

Note for whoever picks this up

The issue is currently labelled enhancement + PR welcome; it is a bug. Also worth a follow-up: this survived two years because an any cast hid a shape mismatch at a call boundary, and the autosize code is the least test-covered area of slick.grid.ts — a targeted sweep for similar casts there would be worthwhile.

🤖 Generated with Claude Code

[6pac]

@ghiscoding this is a small fix in response to a bug report in issue 934. I'm merging it because it's simple and you don't use this code.

[6pac]

BTW we are done for the moment, I suggest you do the minor release. There is still a backlog to get through, but I think that will go on forever... I need to focus on work for this week so I'll get back to it next week.

[ghiscoding]

@6pac I can do it tomorrow after my working hours, I'll wait to see if we should merge the other RTL PR as-is or he prefers to leave it open to finish it by then. Anyway, I'll release Monday evening, that would be your Tuesday morning 😆