[renatex314]

Fixes #8.

Problem

dims: {"cliId": null} was accepted, and the device was filed under value 0 of the dimension. On a dimension declared with a capacity, slot 0 is the first value ever reported, so every filter on that value also matched every null-reporting device. In the instance from the issue, f.cliId=65 returned 1126 devices instead of 1. When the filter is used to restrict a map to one client, that client sees devices that aren't theirs.

Cause

Schema::cells_for treated a value list that was present but empty the same as a dimension that was absent, and put both at index 0. null and [] both deserialise to an empty list.

Change

• schema.rs: a dimension sent with no value (null or []) is now refused with a 400. The message names the dimension, and the route adds the device id:
  device "orphan": "cliId" has no value (null or an empty list); report an explicit value, such as "none", for a device that has none.
  Compact reports, metadata patches and GeoJSON all go through this check.
• geojson.rs:
  • A null under a dimension's name in GeoJSON properties used to be read as an absent property, which also ends up at 0. It is now refused the same way as the compact form.
  • A null inside a list ([1, null], [null]) used to be dropped silently. It is now rejected while the body is parsed: 422 unprocessable_body, the same response a fractional value gets.
• README.md: one sentence describing the rule.

Behaviour change

Clients that send null or [] for a dimension now get an error instead of a silent mis-filing. The workaround from the issue still works: report an explicit value (e.g. 0 or "none"), which gets its own slot.

Not in this PR

A dimension that is absent still takes value 0, as documented. That covers a new device reported without dims, and a report that names some dimensions but not others. On a capacity dimension this has the same exposure. Fixing it means reserving a slot for "unassigned" (which affects snapshots), so it should be its own change.

Testing

• New HTTP regression test a_dimension_with_no_value_is_refused_not_filed_under_the_first_one (tests/regressions.rs). It replays the issue's setup and sends null, [] and [null] in compact form, in GeoJSON and as a metadata patch. It checks that each is rejected with the dimension and the device (or GeoJSON feature) named, that nothing is stored, and that f.cliId=65 still totals 1. Against the old code it fails with {"accepted":1,...}, which is the bug.
• Updated geojson.rs unit tests for the new null handling.
• cargo test --workspace (run in rust:1.98-slim-bookworm): all 218 tests pass.

🤖 Generated with Claude Code