isumitjha opened a new issue, #44532:
URL: https://github.com/apache/superset/issues/44532
## Bug description
In the Calendar Heatmap (`cal_heatmap`) with **Domain = Month** and
**Subdomain = Week**, the cell for a month's first week is positioned roughly
one cell-width to the *left* of its month block whenever the 1st of that month
falls on a Monday (the week start day).
It lands on top of the previous month's last cell, so the two overlap and
only one dot is visible. The month then appears to begin at its second week,
and a week of data looks missing.
The data itself is correct and present — only the position is wrong.
## How to reproduce
1. Create a Calendar Heatmap on any dataset with a date column and daily
values
2. Set **Domain = Month**, **Subdomain = Week**
3. Use a time range covering a month whose 1st is a Monday — e.g. June 2026,
or February / March / November 2027
4. Look at that month: its first week is not visible, and the month appears
to start at the second week
## What is actually happening
Inspecting the rendered SVG for June 2026 (`cellSize: 12`, `cellPadding:
10`; screen coordinates will vary with viewport width):
```
month block "June" spans screen x 754–854
cell 2026-06-01 x=-22 screen 754 <-- negative position
cell 2026-06-08 x=0 screen 776
cell 2026-06-15 x=22 screen 798
cell 2026-06-22 x=44 screen 820
cell 2026-06-29 x=66 screen 842
month block "May" spans screen x 686–764
cell 2026-05-25 x=66 screen 752
```
The 1 June cell exists, carries its value and colour-level class, and is
painted — at `x = -22`, which puts it 2px from the 25 May cell. The two overlap
almost exactly, so one hides the other. The month blocks themselves also
overlap (May ends at 764, June starts at 754).
## Root cause
Two off-by-one errors in
`superset-frontend/plugins/legacy-plugin-chart-calendar/src/vendor/cal-heatmap.ts`:
**1. Negative x position** — `getMonthWeekNumber`:
```js
return this.getWeekNumber(d) - monthFirstWeekNumber - 1;
```
The `- 1` assumes a month's first week cell always begins in the *previous*
month, which holds only when the 1st is not the week start. When the 1st **is**
the week start, both week numbers are equal and this returns `-1`, so
`positionSubDomainX` yields `-1 * (cellSize + cellPadding)`.
**2. Month block sized one column short** —
`_domainType.week.defaultColumnNumber` for `domain: 'month'`:
```js
return self.options.domainDynamicDimension
? self.getWeekNumber(new Date(d.getFullYear(), d.getMonth() + 1, 0))
- self.getWeekNumber(d)
: 5;
```
`domainDynamicDimension` defaults to `true`, so this returns `endWeekNb -
startWeekNb`, while `computeWeekSubDomainSize` generates `endWeekNb -
startWeekNb + 1` cells. The block is therefore exactly one column narrower than
the number of cells it must hold, which is why adjacent month blocks overlap.
## Already fixed upstream; Superset vendors the older code
This is wa0x6e/cal-heatmap#222 ("Some weeks info is missing", opened 2016),
which describes the same trigger:
> When we put Month as Domain and Week as SubDomain the weeks which are
staring 1st of the month as Monday are missing
A later comment on that issue matches the behaviour above: *"some weeks
appear out of order, some weeks over-write other weeks, some weeks are
duplicated in more than one month"*.
Upstream reported "Weeks positioning issues is fixed in master" (Dec 2022)
and closed the issue as completed in Feb 2023. The library is now on 4.x.
Superset's vendored copy predates that rewrite — it is built on the d3 v3
`d3.time.*` API — and still contains both lines above, including on `master`.
## A second defect worth fixing at the same time
The natural workaround — switching Domain to Year — avoids the `-1`, but
runs into an unguarded lookup in the data-fill loop:
```js
var index = temp[domainUnit].indexOf(
this._domainType[this.options.subDomain].extractUnit(date),
);
subDomainsData[index].v = data[d];
```
`indexOf` returns `-1` when a data point's week has no generated cell, and
`subDomainsData[-1]` is `undefined`, so the chart fails with:
```
TypeError: Cannot read properties of undefined (reading 'v')
```
With `domain: year`, no cell is generated for the final partial week of the
year, so any dataset containing data in that week crashes the chart outright
rather than dropping a point. (Related upstream report of the same
unguarded-index pattern in a different configuration: wa0x6e/cal-heatmap#52.)
A bounds check there would turn a crash into a skipped value.
## Environment
Superset 6.1.0. Both code paths verified unchanged on `master`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
