refactor(ogc): clarify module boundaries#346
Draft
thodson-usgs wants to merge 7 commits into
Draft
Conversation
- Replace pd.json_normalize with pd.DataFrame in _get_resp_data. OGC API properties are flat dicts; json_normalize's recursive flattening is unnecessary overhead (~2.4× faster per page: 3.2 ms → 1.35 ms on 2500-feature pages). - Skip drop_duplicates in _combine_chunk_frames when the plan has no filter axis. List-axis chunks produce non-overlapping partitions, so dedup is pure overhead in the common parallel_chunks case (~2.7× faster combine: 3.4 ms → 1.25 ms on 52k-row results). - Add ChunkPlan.has_filter_axis property to distinguish plans that need dedup (filter-axis overlap possible) from those that don't (list-axis only).
Move response/frame recombination helpers (_combine_chunk_frames, _combine_chunk_responses, _merge_response, _lowest_remaining, _safe_elapsed, _set_response_url, _QUOTA_HEADER) into a dedicated combining.py. planning.py re-exports them for backwards compat. Separates 'what to split' (planning) from 'how to reassemble' (combining), each with its own reason to change.
…g.py These type aliases define ChunkedCall's contract, not anything about interruptions. They lived in interruptions.py only to avoid a circular import. Move them to chunking.py where they belong; interruptions.py keeps TYPE_CHECKING re-exports for backwards compat.
…perties Drop the .copy() snapshot of call.partial_frame and the stored partial_response from ChunkInterrupted.__init__. Instead, partial_frame and partial_response are now @Property delegates to self.call when available, falling back to empty/None when call is None. This avoids copying potentially large DataFrames on every ChunkInterrupted raise — the primary usage pattern is exc.call.resume(), not inspecting the partial data on the exception itself. Pickling still works: __getstate__ snapshots the current partial_frame and partial_response into private backing attributes that __setstate__ restores, so the degraded (call=None) unpickled form retains the data that was available at serialization time.
…helper Extract the per-exception type matching from _classify_chunk_error into a new _classify_transient(exc) helper that classifies a SINGLE exception (no chain walk) into (interrupted_class, retry_after) | None. _classify_chunk_error now calls _classify_transient on each step of its __cause__ walk. _retryable continues to inspect only the top-level exception (preserving the intentional divergence: initial-page raw transients are cheap to retry; wrapped mid-pagination failures escalate to a resumable ChunkInterrupted instead). When a new transient error type needs recognition, it now goes in ONE place (_classify_transient) and both the retry and escalation paths pick it up automatically.
…hims Follow-up cleanup on the module-boundary refactor so the split lands in the import graph, not just the file layout. - Import the recombination helpers straight from ogc.combining at every call site (chunking, engine, wateruse, tests) and delete planning.py's backwards-compat re-export block + __all__. The new module previously had zero direct importers; everything routed through planning. - Drop the dead TYPE_CHECKING re-export of _Fetch/_Finalize/ _passthrough_result from interruptions.py (no importers) and repoint the lone Sphinx xref in shaping.py to ogc.chunking._Finalize, their real home. - Remove ChunkInterrupted.__setstate__: __getstate__ already writes the _pickled_* snapshot keys and the base DataRetrievalError.__setstate__ restores them via __dict__.update, so the override was redundant (the property getattr fallbacks cover the hand-constructed case). - Correct _classify_transient's docstring and _retryable's comment, which claimed a delegation that does not — and deliberately should not — exist (the two use intentionally different type sets). No behavior change; mypy --strict and the affected test suites pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…tate Revert the live-view change (item 3 of DOI-USGS#346): restore exc.partial_frame / exc.partial_response as raise-time snapshots rather than properties that delegate to exc.call. Snapshot is the better contract for an exception — a record of the failure moment, not a live handle: - In a resume loop each interruption stays a faithful record of what it saw. Live-view aliased every exception to the one shared call, so after a second failure exc1.partial_frame and exc2.partial_frame returned the same (later) state. - After a successful resume, live-view's .partial_frame returned the COMPLETE result — a field named "partial" that no longer was. - Exception payloads shouldn't mutate after you catch them (cf. CalledProcessError.output). Removing the delegation also drops the properties and the _pickled_* / __getstate__ snapshot machinery it required: partial_frame/partial_response are plain instance attributes again, pickled by the base __getstate__ (net -21 lines in interruptions.py). No behavior change versus released main — this PR is unmerged, so live-view never shipped. DOI-USGS#346 is now a true no-behavioral-change refactor. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Structural refactoring of the OGC chunking module to clarify responsibilities and reduce coupling. No behavioral changes — all existing tests pass (one test updated to reflect the live-view simplification in item 3).
Includes the perf commit from #345 as the base.
Changes
1. Extract
combining.pyfromplanning.py(item 2)Moves response/frame recombination helpers (
_combine_chunk_frames,_combine_chunk_responses,_merge_response,_lowest_remaining,_safe_elapsed,_QUOTA_HEADER) into a dedicatedcombining.py.planning.pyre-exports them for backwards compatibility.Rationale: planning.py was conflating "what to split" with "how to reassemble" — two concerns with different reasons to change.
2. Move
_Fetch/_Finalize/_passthrough_resultintochunking.py(item 3)These type aliases define
ChunkedCall's contract, not anything about interruptions. They lived ininterruptions.pyonly to break a circular import. Now they're inchunking.pywhere they belong.3. Simplify
ChunkInterruptedby dropping snapshot copies (item 5)ChunkInterrupted.__init__used to.copy()the partial frame at raise time soexc.partial_framewas a frozen snapshot separate fromexc.call.partial_frame(the live view). The distinction was subtle and unlikely to be used by callers (the primary pattern isexc.call.resume()), and cost a full-frame copy proportional to completed data.Now
partial_frameandpartial_responseare properties that delegate toself.call— same as the live view, no memory copy. Pickle still works (snapshots at serialize time).4. Unify error classification into shared
_classify_transienthelper (item 6)_classify_chunk_error(chain-walk) and_retryable(top-level only) shared ~80% of their logic. Created a shared_classify_transient(exc)that classifies a single exception, so adding a new transient type only needs one update.Deferred to follow-up PRs
gather_paginatedprimitive — unifyingChunkedCall._runandwateruse._fan_out(invasive; needs careful testing of the retry/interruption model)engine.py— separating request construction, pagination, and argument normalization (~750-line module with 3 concerns)