Hi Thomas, I am fine with this being in frontend-passes.c - it was just a question :-) resolve.c has become too large anyway.
The testcase looks familiar! Don't forget to commit and push the additional source too. OK for 10-- and all the affected branches. Cheers Paul On Sat, 18 Jul 2020 at 18:57, Thomas Koenig <tkoe...@netcologne.de> wrote: > Hi Paul, > > > The patch looks fine to me but I have two questions: > > > > (i) Why is this not done in resolve.c? > > Of course it doesn't matter where the function resides :-) I put it > into frontend-passes.c because it makes heavy use of gfc_code_walker, > and out of habit. If you prefer, I can of course move the code to > resolve.c. > > If your question is more like "why is that not done during > normal resolution" - while fixing this bug, I began to understand > why the standard has an explicit PURE attribute. If the compiler > is chasing after a procedure which may or may not be implicit_pure, then > everything else needs to have been resolved beforehand. And in normal > resolution, something needs to be the last. > > > (ii) Is Martin's reduced reproducer not the basis for a testcase? > > There, git got the better of me. I thought it was included in the > patch, but, as I reread it and found that this wasn't the case. > > I have attache the test case now. > > OK for trunk and backport once gcc 10 is open again? > > Regards > > Thomas > -- "If you can't explain it simply, you don't understand it well enough" - Albert Einstein
