You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Changed
server extra: corrected the remaining floors to the lowest versions that actually
resolve together and pass the suite - fastapi>=0.110.1 (was >=0.115) and mcp>=2.0.0,<3 (was >=2.1.1,<3). No change to what the package does.
server extra: pydantic>=2.12,<3. The server has always imported pydantic at module
level and never declared it. It only arrived transitively through fastapi and mcp. The
floor is the one mcp already requires, so nothing resolves differently: the import is
declared instead of inherited.
Fixed
The TSV/CSV export kept only the first synonym of each type for an entry and dropped the
rest: 1579 of the ontology's 7288 synonyms, across 590 entries. MOD:00014 alone has 12 RESID-alternate synonyms, of which 11 were lost. Each synonym_<type> column now holds
every synonym of that type, in file order, separated by " | ". That separator is not the "; " the other multi-value columns use, because 35 synonym values contain "; "
themselves and such a column could not be split back. The columns and their order are
unchanged.
server extra: uvicorn>=0.30 could not be installed. mcp requires uvicorn>=0.31.1
on every platform but emscripten, so no resolution of the extra contained a 0.30.x, and --resolution lowest-direct silently stepped over the declared floor - the floor was
unreachable rather than merely untested. Now uvicorn>=0.31.1. On emscripten, where mcp
does not require uvicorn, this does raise the floor.
search_mass: adding position to a query could add hits instead of removing them. The
terminus-slot rule read position=None as "neither terminus is open", which made the
least-constrained query the narrowest one. Both termini are now open for position=None
and closed for position="anywhere", so position only ever removes matches. For
PSI-MOD data this changes no result. A slot's sites come from origin residues only,
never from N-term or C-term, so the rule could not fire here - a sweep of 1140
result sets is identical before and after. tests/test_position_narrows.py is the guard
that keeps it that way, because the same mass-search module ships in unimodpy and
uniprotptmpy, where the rule does fire.