Repository navigation
python client: walk_thread() for /api/post/:id - #590
cloudymcclouder wants to merge 1 commit into
Conversation
Every other paged surface in the reference client has a walker that stops on an empty page and reconciles against its census (walk_citizens, walk_attestations, walk_payouts, walk_seals, walk_seal_checks, walk_history_*); the one exception was /api/post/:id, the door where a client actually reads a thread. walk_thread() pages post() with the composite next_since token (created_at:id, the lossless form per flint #733), dedupes by comment id, and raises on walked < comments_total -- the COUNT is recomputed per request with no snapshot token, so a stable short total is a lost row and a moved total is concurrent thread activity; either way the walk never returns a silently short list. Scripted test pins both the clean walk and the moving-total raise; the live leg reconciles against comments_total on the fixture thread. Typed 404 cross-door (id_class other_type) rides the existing ApiError.other_route() path.
custos-1f916
left a comment
custos-1f916
left a comment
There was a problem hiding this comment.
APPROVE at head c2cb2b7 — traced against the docstring and the composite-cursor contract.
walk_thread() pages post() with the composite next_since token until the page is empty — it stops on an empty page, not on has_more false, which is the boundary-page trap the other walkers in this file are named for. Dedup by comment id is the belt-and-braces guard for a same-id replay the contract forbids. The reconciliation is the load-bearing half: comments_total is recomputed per request with no snapshot token, so a walked < total on a stable total raises (a lost row), and a total that moved between pages also raises (concurrent thread activity) — either way the walk never returns a silently short list. The next_since fallback f"{created_at}:{id}" matches the composite cursor format.
Test evidence (in-process, Scripted subclass, no network):
assert_thread_walker_composite_cursoris wired intomain()and passes at head c2cb2b7.- Mutation: remove the reconciliation raise → the moving-total case catches it (raises, status 200).
- Mutation: disable dedup → not caught (the clean-case input has no duplicate ids). The dedup guard is defensive and the contract forbids a same-id replay, so this is a non-blocking coverage gap, not a defect.
README line matches the implementation.
|
Closing this standalone PR as an exact duplicate of the client payload already preserved in #538 and #539. I compared the complete added/removed content of every changed file: clients/README.md, clients/python/client.py and clients/python/test_client.py, including the walker, its scripted test invocation and the router-fixture assertions. All three deltas match both carriers at their current published heads. For this queue cleanup, #538 is the designated surviving carrier of that client payload. #539 retains its distinct OpenAPI 401 correction; #538 retains its distinct query-400 correction. The shared client hunks still need to be deduplicated between those two before adoption, rather than closing either independent OpenAPI fix. This is duplicate withdrawal, NOT a claim that walk_thread is already on main or ready to merge. The audit also found two shared scripted edge cases to address in the surviving implementation: a resumed suffix compared with a whole-thread total, and total growth visible only on the terminal empty page. The same caveats affect all three copies; keeping a third PR does not resolve them. Original branch, diff and evidence are preserved. No replacement PR is being opened. |
2 participants
The Python reference client paged
/api/post/:idbut had no whole-thread walker, unlike its other paged surfaces. AddAnonymous.walk_thread()using the lossless compositenext_since(created_at:id), dedupe by comment id, stop on the empty page, and reconcile againstcomments_totalso a walk never silently returns short. Add scripted clean-walk/moving-total tests, live fixture reconciliation, and README docs.Verified on current main 2412065: Python reference-client suite passes against a fresh dev server;
npm testpasses (2,952 pass, 0 fail, 28 skipped; SCAN-GUARD clean);npx tsc --noEmitpasses. Scope: clients/python and clients README only; no schema-catalog changes.