Explicit mutation sites¶
solvi.testing.mutants.generate and run mutate numeric literal match patterns
(two threshold nudges), string literal match patterns (a changed literal), and
boolean singleton patterns (True/False flip). Captured/sequence/class/OR patterns
and the None singleton are not pattern mutation sites.
A multi-statement if body ending in a direct raise, without an else, can be
dropped by making that guard false. An assert can separately be made to pass.
These mutations deliberately change the selected guard; they do not infer arbitrary
control flow or guarantee that every exception site can be dropped.
Checks registered through a helper decorator or call form can be declared by source name, without importing the task to discover callbacks:
from solvi.testing import mutants
report = mutants.run("cases.json", check_functions=["known", "Rules.positive"])
check_functions is an explicit registration manifest, not a callable or automatic
Catalog scan. Names must uniquely identify functions in the task source; unknown,
ambiguous and malformed declarations are rejected. functions= still limits the
selected source scope. Direct @cat.check / @cat.constraint decorators need no
manifest. No helper or registration call is executed during generation; the normal
isolated baseline and mutant lifecycles still execute the task when run is used.
The report records the manifest. AST deduplication and the optional exact-builtin-int equivalence grouping retain their existing score units and alias records. This is coverage of these finite source constructs, not mutation completeness for Python.