Refget Sequences start/end query parameter and range request corner cases

[jmarshall]

The specification is not explicit as to whether clients must use either both or neither of start=… and end=… query parameters, or if just one of them may also be used. If it is permissible to use start=… by itself or end=… by itself, the specification is not explicit about what that should mean.

The specification does not specify what should happen when a request contains an end query parameter larger than the total sequence length. Reasonable behaviours could be to return a sub-sequence from start to the end of the sequence (i.e., “clipping” the end value) or to fail with Bad Request (as the requested end is out of bounds).

Note that the specification does say that out of bounds Range requests must fail with Bad Request. However this is inconsistent with RFC 7233, which specifies that out of bounds last‑byte‑pos is to be clipped while out of bounds first‑byte‑pos is unsatisfiable so should fail.

[nsheff]

At this point, we could just update the spec to reflect what the reference implementation actually does.

[andrewyatz]

I mean I am okay with it clipping if that is better behaviour and consistent with RFC 7233.

In the other cases though the compliance suite I believe allows for start or end to be omitted, one used or both used

[nsheff]

Proposed clarifications based on testing EBI implementation:

Scenario	Current EBI behavior	Tested in compliance suite	Recommendation
start only (query)	200, returns start→end	✓ Yes	Add to spec: allowed
end only (query)	200, returns 0→end	✓ Yes	Add to spec: allowed
end > length (query)	416	✓ Yes	Add to spec: 416
start > length (query)	416	Partial (circular only)	Change spec from 400 to 416
last-byte-pos > length (Range)	206, clips	✓ Yes	Add to spec: clip per RFC 7233
first-byte-pos > length (Range)	⚠️ 200, full sequence	✗ No	EBI bug — should be 416 per RFC 7233

Summary:

1. Independent params — EBI already supports using start or end alone. Spec should explicitly allow this.
2. Out-of-bounds values — For query params, 416 is fine. For Range headers, RFC 7233 requires clipping last-byte-pos but rejecting first-byte-pos > length with 416.
3. EBI bug — Range header with first-byte-pos > length returns 200 with full sequence instead of 416. Recommend fixing implementation.
4. Compliance suite gap — No test for Range header with first-byte-pos > length.

Proposed spec changes: PR forthcoming.

Test script: test_edge_cases.py | Compliance suite: https://github.com/ga4gh/refget-compliance-suite

[andrewyatz]

Just checking is this against https://www.ebi.ac.uk/ena/cram or https://beta.ensembl.org/data/refget?