Qiskit / qiskit / 36025562704

88%

main: 88%

Build Type

push

github

Committed by web-flow

Commit Message

Add the `ShotLoop` node (#16933)

* Add a shot loop, the op that puts circuits into a program

A shot loop carries a list of circuits and a shot count, takes one operand per
circuit for that circuit's parameter values, and returns one tensor of outcomes
per classical register. Its results are the first an op derives from something
other than its operand types, so it writes its own inference rather than sharing
the catalogue's five: the register widths come from the circuits themselves, and
only the batch axes come from the operands.

An operand must be floating point with one value per parameter in its trailing
axis, so a mismatch is caught where it is made rather than when a backend rejects
the job. Axes ahead of that one are a batch prefix, carried through unchanged onto
that circuit's outcome shapes for any rank of prefix, which is how one program
sweeps many parameter sets. The shots axis is fixed, since a circuit's output
format is static even when the operations it runs are not.

Which register a result belongs to is naming, so it lives in a structure:
`output_structure` describes the results as one entry per circuit keyed by
register name, in the leaf order the results come back in, and a caller reads it
with the same vocabulary as any other structured value. A register whose name
cannot name a slot is rejected when the op is constructed, because such a result
could not be addressed at all.

This is the canonical work Qiskit cannot perform in process, so it reports no
built-in evaluation. A program containing one still builds and reports all its
types; evaluating one fails at the top, naming the instruction that needs a
backend, without computing anything first.

* Stop deriving Clone, PartialEq and Eq on ShotLoopError

These traits bind every future variant, and every error a variant nests, to
deriving them too; `InvalidName` no longer derives them, so this one cannot
either. Tests match on errors instead of comparing them or reading their
`Di... (continued)

Coverage Stats

284 of 296 new or added lines in 2 files covered. (95.95%)

15 existing lines in 3 files now uncovered.

121914 of 138404 relevant lines covered (88.09%)

944267.94 hits per line