Las poblaciones de examen se declaran, y el servidor verifica lo que hay dentro

alembic/versions/b7c3f91d2a04_query_set_carries_an_exam_population.py

@@ -0,0 +1,84 @@
+"""query_set carries an exam population, which has no evaluation window at all
+
+THE SIXTH KIND, AND THE SECOND ONE THAT IS NOT A WINDOW'S DELTA. ADR-D57 puts
+the vocabulary exams inside the platform, and its decision 2 says each exam
+population enters as a declared set. There are two of them, frozen once by the
+laboratory: 3.335 proteins for the site exam and 4.227 for the domain exam, over
+Swiss-Prot 2022_03.
+
+WHY NOT ``corpus_population`` AND WHY NOT A WINDOW. An exam population is not
+the corpus: it is a frozen sample with a partition into a selection and a
+confirmation half, and the confirmation half is spent once, at the end, by the
+platform. And it has no evaluation window: these exams sit on Swiss-Prot
+2022_03, which has nothing to do with the cut of a transition window. That is
+also why they do not fit ADR-D55's exploration unit, whose
+``evaluation_set_id`` is NOT NULL.
+
+WHAT MAKES IT VERIFIABLE, which is the half this migration can enforce. The
+artifact a row of this kind points at is a MANIFEST, not the population file: an
+exam population is several files (the pairs of accession and md5, the split, the
+concepts, the instances or the sites, and the laboratory's own manifest with the
+scripts that built them), and a ``query_set`` row has one ``artifact_uri`` and
+one ``content_sha256``. So the row names the manifest, the manifest names every
+file with its own digest, and the declaring operation re-reads and verifies each
+one before any row exists. A digest proves nobody altered the bytes; it never
+proves they are the right bytes, so the operation also checks each file's row
+count against what the manifest claims about it.
+
+Revision ID: b7c3f91d2a04
+Revises: a4e91c0d5b83
+"""
+
+from __future__ import annotations
+
+from alembic import op
+
+revision = "b7c3f91d2a04"
+down_revision = "a4e91c0d5b83"
+branch_labels = None
+depends_on = None
+
+SOURCES = (
+    "evaluation_targets",
+    "fasta_upload",
+    "quick_annotation",
+    "corpus_population",
+    "exam_population",
+    "legacy_undeclared",
+)
+
+PREVIOUS = tuple(v for v in SOURCES if v != "exam_population")
+
+
+def upgrade() -> None:
+    op.drop_constraint("ck_query_set_source_kind", "query_set", type_="check")
+    listed = ", ".join(f"'{v}'" for v in SOURCES)
+    op.create_check_constraint(
+        "ck_query_set_source_kind", "query_set", f"source_kind IN ({listed})"
+    )
+    # The same four criteria the corpus kind carries, for the same reason: the
+    # set guard as a constraint and not as a habit. Described, verifiable (a uri
+    # with its sha), produced by a job, and NOT derived from a window. The last
+    # clause is the one that carries work here too, and more of it: an exam
+    # population could LOOK like it belongs to a window, because the exams will
+    # be compared against windows' results, and it does not.
+    op.create_check_constraint(
+        "ck_query_set_exam_is_declared",
+        "query_set",
+        "source_kind <> 'exam_population' OR ("
+        "artifact_uri IS NOT NULL AND content_sha256 IS NOT NULL "
+        "AND description IS NOT NULL AND job_id IS NOT NULL "
+        "AND evaluation_set_id IS NULL)",
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint("ck_query_set_exam_is_declared", "query_set", type_="check")
+    op.drop_constraint("ck_query_set_source_kind", "query_set", type_="check")
+    # Back to the five it had. A row already written as 'exam_population' makes
+    # this downgrade fail, which is correct: dropping the value while rows use
+    # it leaves a kind the CHECK forbids and the data asserts.
+    listed = ", ".join(f"'{v}'" for v in PREVIOUS)
+    op.create_check_constraint(
+        "ck_query_set_source_kind", "query_set", f"source_kind IN ({listed})"
+    )

protea/core/operation_catalog.py

@@ -43,6 +43,9 @@
 from protea.core.operations.declare_corpus_population import (
     DeclareCorpusPopulationOperation,
 )
+from protea.core.operations.declare_exam_population import (
+    DeclareExamPopulationOperation,
+)
 from protea.core.operations.declare_query_set_provenance import (
     DeclareQuerySetProvenanceOperation,
 )
@@ -140,6 +143,7 @@ def _register_window_maintenance(registry: OperationRegistry) -> None:
     precisely because they change what a row says about itself.
     """
     registry.register(DeclareCorpusPopulationOperation())
+    registry.register(DeclareExamPopulationOperation())
     registry.register(DeclareQuerySetProvenanceOperation())
     registry.register(ExportTransitionsDatasetOperation())
     registry.register(RetireEvaluationSetsOperation())