Make the streaming-relevance baseline reproducible in-repo#39
Merged
estebanzimanyi merged 1 commit intoJul 10, 2026
Merged
Conversation
The generator codegen_facades.py projects the JMEOS FFI surface against
tools/baseline/streaming-relevance-baseline.json, but that baseline was a
frozen snapshot produced by an out-of-repo classifier and recorded a
dangling absolute source_catalog path, so it could not be regenerated
here — the last non-reproducible input in the generation chain.
Vendor the deterministic classifier as tools/classify_streaming_relevance.py
(the tier of a function is decided purely by its name, object-model role,
and temporal-parameter count) plus a tools/regen_baseline.sh wrapper, and
document the full input chain in GENERATION.md:
MobilityDB @ tools/meos-source-commit.txt
-> MEOS-API/run.py -> meos-idl.json
-> classify_streaming_relevance.py -> streaming-relevance-baseline.json
-> codegen_facades.py -> org.mobilitydb.meos.MeosOps* facades
Regenerate the committed baseline through the vendored classifier so its
provenance is portable. The function classification is byte-identical to
the previous baseline, so the 57 generated facade classes are unchanged;
only the metadata (source_catalog, generated_at) differs.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
The generator
tools/codegen_facades.pyprojects the bundled JMEOS FFI surface againsttools/baseline/streaming-relevance-baseline.json(it emits a facade only for the streaming-relevant tiers). That baseline was the one non-reproducible input left in the generation chain: a frozen snapshot produced by a classifier that lived outside the repo, recording a dangling absolutesource_catalogpath — so nobody (and no CI) could regenerate it.This PR closes that gap, bringing Flink in line with the peer bindings (Spark regenerates its UDFs from the tracked commit via an in-repo generator).
Changes
tools/classify_streaming_relevance.py— vendored deterministic classifier. A function's tier is decided purely by its name, its object-model role, and its temporal-parameter count (zero per-function judgement), so the same MEOS catalog always yields the same baseline.tools/regen_baseline.sh— thin wrapper to regenerate the baseline from a catalog.GENERATION.md— documents the full input chain:tools/baseline/streaming-relevance-baseline.json— regenerated through the vendored classifier so its provenance is portable.Non-breaking — proven
The function classification is byte-identical to the previous baseline (3548 functions, 0 tier changes, identical order); only the metadata (
source_catalog,generated_at) differs, andcodegen_facades.pyreads onlyfunctions[]. Regenerating the 57 facade classes from the old vs new baseline produces byte-identical output (diff -rqempty). The classifier is verified deterministic (two runs byte-identical).