On Fri, Mar 20, 2026 at 11:53:51AM +0200, Jean-Christian CÎRSTEA wrote: > Le jeu. 19 mars 2026 à 13:25, Nathaniel Shead <[email protected]> a > écrit : > > > Sorry if I wasn't clear in my earlier message, I meant have two tests: > > one using the implicit module mapper (no `-fmodule-mapper`), and the > > other using an explicit one. > > Why are two tests needed? Does the implicit module mapper behave differently > than the explicit one regarding non-ASCII module names? If so, how should the > implicit module mapper test look like?
It's more that I felt it would be good to have tests to ensure that they _don't_ behave differently, and that neither gets accidentally regressed at some point. The implicit module mapper test is the one that you had in v1/v2 of your patch; I asked for an additional test for the explicit module mapper too to make sure that reading the module names and paths from file didn't cause complications in comparison to just generating them directly. That is, I would like to be able to check that both g++ -fmodules -S a.cpp b.cpp and g++ -fmodules -fmodule-mapper=some.map -S a.cpp b.cpp work correctly, even in the presence of non-ASCII module names. Yours, Nathaniel
