wasm needs targettype to be sourceLibrary

[skoppe]

Otherwise lld won't link it correctly with the dependant. For templates it is not an issue, but for regular functions it is.

[9il]

LGTM

[skoppe]

Hmm. It appears what I wrote above is not true. It does fix the problem I was having (unresolved references), but it should be considered a workaround at best.

You see, I recently started using dip1000 in spasm, and suddenly I got an unresolved reference for the BitVector.rep function defined in bitmapped_block.d.

Because BitmappedBlock is a template it was instantiated with the compiler flags of spasm (which includes dip1000), and it inferred an extra return attribute on the BitVector.rep function, causing the mangling to be different vs. the BitVector.rep function in the static compiled library. (I am using dub).

The proper solution (in my case) would be to compile stdx-allocator with dip1000, but that would probably cause issues for dependent projects that don't do dip1000...

[9il]

I think the allocators should be a source library. To solve similar issues in Mir I reworked structs and their methods as templated structs/methods.

[skoppe]

I don't think it scales so well to have every library that uses different compiler flags to be a sourceLibrary. Plus, I need the dip1000 for unittests as well.

It also seems an issue more and more people will run into when using more dip switches.

I don't have a good solution for this. It could be argued to be a dub issue that it doesn't propagate the compiler switches when compiling dependencies. But I reckon that propagating switches to come with drawbacks as well.