I actually need the opposite here: the parallel part is easy enough*, but the missing test result grouping is a hard blocker for me. With 32 combinations across 3 matrix jobs and a total of around 6000 tests grouping is critical.
Jesse Glick I haven't checked this for a while, there still is no test result grouping of parallel steps, right? Is that planned somewhere?
*: actually the execution order of the matrix combinations is bad too: they get scheduled in alphabetical order, but scheduling based on execution time / diagonal across all axes would be much more useful to get test result feedback earlier.