openvax / pyensembl / 25816358406

85%

master: 83%

Pull #355

github

Address PR #355 review: drop pickle schema bump, surface fasta_version

* Reverted FASTA_PICKLE_SCHEMA_VERSION + the .v2.pickle filename suffix.
  v1 (bare-keyed) pickles load cleanly under the new code path - the
  rebuilt _stripped_index stays empty, but lookups still resolve via
  the version-stripped retry path so Ensembl-style callers see no
  regression. New test test_old_bare_keyed_pickle_still_loads_under_new_code
  pins this.

* Added Transcript.fasta_version - returns the int version that the
  cDNA FASTA header carried for this transcript, or None when the
  genome has no transcript FASTA / the header was bare.

* Added Protein.fasta_version - same shape but for the protein FASTA.
  Protein now carries an optional genome= reference so the view object
  can consult genome.protein_sequences.fasta_version(). Default is
  None for back-compat with callers constructing Protein outside of
  Transcript.

* Opted out of gtfparse 2.7.0's cast_version_columns=True in
  database.py's read_gtf call. When the *_version columns are missing
  on some rows (e.g. start_codon rows don't carry transcript_version),
  pandas' nullable Int64 routes through float on the sqlite write path
  and the value lands as '7.0' text, which broke pyensembl's
  int(result[0]) parse. Keeping strings and doing the int conversion
  on the property side as before.

* New tests for both fasta_version accessors, including a "GTF claims
  version 5, FASTA claims version 3" disagreement case so downstream
  tools can detect mismatched GTF/FASTA pairs.

60 of 61 new or added lines in 5 files covered. (98.36%)

1653 of 1935 relevant lines covered (85.43%)

3.42 hits per line