The number branch of tomllib's value parser calls match_to_number without wrapping ValueError, so an integer with more digits than the interpreter's integer string conversion limit escapes tomllib.loads as a bare ValueError instead of tomllib.TOMLDecodeError. The sibling date and time branch already wraps match_to_datetime's ValueError into TOMLDecodeError; this mirrors that for the number branch, so a malformed document raises only TOMLDecodeError, which is what the documentation promises.

Positive and negative decimal integers are both affected. Hexadecimal, octal, and binary literals are not, because they bypass the integer string conversion limit.

The wrap is scoped to the integer conversion only. A parse_float callback that raises ValueError still propagates unchanged, since that error belongs to the caller and is not a decode error; the existing test_invalid_parse_float contract is preserved.

The regression test in test_error.py forces a low sys.set_int_max_str_digits through test.support.adjust_int_max_str_digits, so it is deterministic and fast; it fails before the change and passes after.

tomllib is vendored from tomli; an upstream sync of the same wrapping is the maintainer's call.