Andy Porter

Am getting test compilation errors because (in the kernel-extraction driver tests) I think we're getting the wrong module definition inserted into the generated code - it has an empty `testkern_code`...

It turned out I only needed to rename one module in one test source file to fix that problem so I've done that. To fix the problem in a real...

Am trying to apply this to Clem's refactored sea-ice branch of NEMO. It's revealing quite a few problems. Currently I'm hitting problems merging symbol tables when doing inlining. I've also...

If a routine accesses a variable brought into module scope by a wildcard import, then attempting to module inline a second routine that does the same thing (for the same...

Can I break-out my changes to `ModuleManager` and `Call` into separate PRs as this is currently rather large?

NEMO integration tests fail with: ``` Inlining of 'ctl_warn' failed: Transformation Error: Routine 'ctl_warn' contains one or more CodeBlocks and therefore cannot be inlined. Inlining of 'dia_hsb_rst' failed: Transformation Error:...

Performance is now back to the level it is on master (no faster unfortunately). However, this has required that I turn off the automatic addition of profiling calipers to non-OpenACC...

> With that you meant the automatically applied psydata-based profiling regions (using `--profile kernels/invokes`)? It is strange that this would cause a huge overhead (maybe verify that they are indeed...

I've added a test on `1651_nested_where`.

An update: I've just brought the branch mentioned above to master and the 'new' test that has been added still fails in the same way.