From Mike: Yeah, this is a decrease of precision, but the new DPath implementation changes came with some loss of proper categorization. Basically, some areas where the code was clearly compilation time or clearly runtime have been replaced by common code that is used in both cases, and that code issues an SDE. Nothing recaptures this and converts it into a "Runtime" SDE currently.
So I'd say it's a bug that we've lost this precision, because I like having the separation where SDEs that are detected to occur at runtime are explicitly identified. But the DFDL spec doesn't say when anything has to be detected.
So if we view the test as testing the DFDL spec requirement, then making the test less precise is ok. If we view the test as verifying our desired behavor for Daffodil, then the loss of precision is a bug.
I suppose really there are two tests then.... I would prefer not to fix this particular issue right away however.
Changes in diagnostic behavior are probably the biggest observable places where behavior will have changed