akvo / akvo-mis / #1087

90%

main: 90%

Pull #405

coveralls-python

[#402] Add the `complete` flag from the GEO-005 design doc

Section 4 of the design document specifies a completeness flag returned
alongside pagination. This branch shipped `geometry_total` instead,
because a boolean cannot express what the device actually needs to know,
and the reasoning still stands: the cursor advances when the final page
is delivered rather than confirmed, so a device that dies mid-page and
drains the next run cleanly would read a naive "complete" as true over a
gapped index. The flag would be computed from the same pagination that
lost the rows.

So the flag is here, scoped to the one thing it can honestly claim:
this response delivered the last page of this listing. It is not a claim
about the device's stored candidate set, and both the comment and the
schema help_text say so explicitly, pointing at `geometry_total` for
that. Shipping it under a vaguer meaning would be worse than omitting
it, because the failure it invites is exactly the silent pass this task
exists to prevent.

It appears only alongside `geometry` and `geometry_total`, not
unconditionally as the document's example shows. Adding a key to every
response would break the first technical acceptance criterion in the
same document -- that a form without `detectOverlaps` sees a
byte-identical response -- and that criterion is explicit and testable
where the example is illustrative.

Also adds the document's fourth testing-strategy row, which had no
coverage with geometry switched on: that pagination and the
`last_updated` cursor still behave. Geometry is attached by a follow-up
query keyed on the page's ids, so paging is precisely where it could go
wrong -- rows duplicated across pages, dropped between them, or the
cursor quietly ceasing to filter.

7361 of 8411 branches covered (87.52%)

Branch coverage included in aggregate %.

13743 of 15075 relevant lines covered (91.16%)

0.91 hits per line