[sathabbott]

🤖 AI generated content

Fixes #135.

FileContext.type_by_name was annotated -> type[ROOTSerializable], but a lookup can return an annotated builtin. TString, TStringLong and string resolve to Annotated[bytes, ROOTString(...)] (#127), and a file's context resolves TDatime to Annotated[int, Fmt(">I")]. mypy therefore accepted .read() on the result, the TypedTKey bug #127 fixed. tests/test_read.py's basket reader still did it.

As decided on the issue (option 1, object rather than Any):

• FileContext.type_by_name, _BootstrapContext.type_by_name and DynamicFileContext.type_by_name return object. The base class's docstring states the contract: read the result with read_value, never .read(), and narrow with isinstance(t, type) and issubclass(t, ROOTSerializable) where the class itself is needed. mypy now rejects .read() on a lookup.
• read_value takes object, so it accepts a lookup's result unchanged.
• tests/test_read.py's _ReadBasket reads its items with read_value.

Beyond the issue, and why:

• TKey.read_object (without objtype) and TKey.read_from return object. They were annotated ROOTSerializable, but they return whatever the lookup reads: bytes for a TString record, int for a TDatime one. This is the same bug one level up, and TKeyList: key by (name, cycle), with get_by_name for the highest cycle #145 surfaced it once key lookups became typed. With objtype given, read_object still returns that type.
• The Locator protocol's type variable is no longer bound to ROOTSerializable. A TKey is a locator, and its object can be a builtin, so the bound no longer holds. Fetcher.__call__ then types reader.fetch(key) as object. Callers narrow with isinstance, which every caller in the tests already does where it needs a class. Remove the DataFetcher-era methods now that locators and rootfilespec.reader exist #114 reworks the locators; this changes only the bound.

Not changed: TypedTKey.objtype stays type[ObjType]. TypedTKey.read keeps its # type: ignore[arg-type], since for a string record objtype is the alias. type_by_checksum is not implemented anywhere and is left as it is.

Tests

New tests/test_type_by_name.py:

• test_return_type_is_object: the return annotation of FileContext.type_by_name, of both implementations and of TKey.read_from is object. All four fail on main.
• test_lookup_of_tdatime_is_an_annotated_builtin, on root-io-spec's serialization/unframed-records.root:
  • The file's context resolves TDatime to the alias, which is not a class.
  • read_value reads the record's payload as 0x7E6CC000, the value its case.toml pins at 318, with nothing left over. reader.fetch(key) returns the same value.
  • .read() raises AttributeError at run time. The test calls it with # type: ignore[attr-defined], so CI's mypy pins the type statically: on main that ignore is unused and the line before it unreachable, and mypy fails.
• test_lookup_of_strings: TString, TStringLong and string resolve to their aliases in the bootstrap context and in a file's.
• test_lookup_of_a_class_narrows: TNamed narrows to the class.

Since #145 merged, test_root_string.py's two string-record reads compare reader.fetch(...) with bytes directly again. #145 had typed them object to work around TKey.read_from's old annotation.

Full suite: 577 passed, 59 skipped, 49 xfailed (main at 2a65519: 568 passed).

Breaking changes

• FileContext.type_by_name (and its implementations) returns object. Code that calls .read() on the result fails type checking; read with serializable.read_value(t, buffer), or narrow with isinstance(t, type) and issubclass(t, ROOTSerializable) first. Run-time behaviour is unchanged.
• TKey.read_from and TKey.read_object without objtype are typed object, so reader.fetch(key) is too. Narrow with isinstance before using the result as a particular class.
• serializable.Locator's type variable is unbound: a locator may return a builtin.

Assisted-by: claude-code:claude-opus-5-5

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.31%. Comparing base (2a65519) to head (db3e332).
⚠️ Report is 2 commits behind head on main.
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #147   +/-   ##
=======================================
  Coverage   93.31%   93.31%           
=======================================
  Files          37       37           
  Lines        3277     3277           
=======================================
  Hits         3058     3058           
  Misses        219      219           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sathabbott]

@nsmith- this is ready for you, but I don't fully understand it.