V0.14rc

[daler]

From changelog:

• If a value contained a semicolon there would be unexpected behavior (reported in #212). This is solved by adding a new entry to the dialect, semicolon in quotes, and running the necessary regular expression only when inferring dialect, or, if semicolon in quotes is True, on every feature. In the latter case, this can dramatically increase the parsing time, since in Python regular expressions are relatively slow, but it does correctly parse. Thanks to @DevangThakkar for the fix.
• While working on that, refactored the attributes parsing to make it clearer to follow along, and added more tests. The refactoring fixed some subtle bugs on corner cases:
  • Previously, for features with repeated keys, the order key of dialects would list the repeated keys each time they appeared (i.e., the list had duplicates) which could result in undetermined behavior. The order key is now unique and only the first occurrence of a repeated key will be added to the order.
  • Previously, the ensembl_gtf.txt example file had a leading space in front of the attributes. This looks to be an error in the creation of the example file in the first place, but had previously parsed fine. Now the parser (correctly) mis-handles it. Since I'm unaware of any cases in the wild that have a leading space, I actually consider the new parsing, which complains about the space, to be more correct.
  • Added tests to directly inspect the inferred dialects for the test cases.
• Preserve GFF directives when create_db() imports from a file path, matching the behavior for string-backed iterators and fixing #213. This was due to a different path through the code when using a pathlib.Path object. In addition to this fix, pathlib.Path objects are now converted to str throughout the code base with os.fspath where appropriate.
• CI, testing, and docs infrastructure updates (miniforge instead of mambaforge; GitHub Action version bumps; skip biopython test if it's not installed (#233); reduce build errors for docs)
• Fix #224, which was caused by changes to the argh package used for the command-line tool.
• Address #242 (typo in docstring)
• Migrate to using pyproject.toml for packaging. This changes how versions are calculated and reported, and removes the need for setup.py. Version is only ever recorded in pyproject.toml; version.py gets the installed version or parses the TOML if not installed; setup.py just calls setup() with no arguments since everything has been migrated to pyproject.toml.