can't alias to existing D type

[chriscamacho]

when translating the header from https://github.com/ggambetta/libz80

dstep emitted the following code (in part)

alias ushort = ushort;
alias byte = ubyte;

fortunately its not a large header so I could just comment these alias's out and change any byte to ubyte ...

[ciechowoj]

@chriscamacho The first case is easy to fix - just ignore the alias, but what about the second case? Do you think having dstep to automatically convert byte to ubyte would solve the problem?

[jacob-carlborg]

This looks like a case where DStep doesn't handle D keywords. A quick fix would be to just suffix the alias names with _ as we do in other places.

[chriscamacho]

although in the C world _ is supposed to be reserved I haven't honestly seen it in practice, I'd say an initial scan to prefix any D keywords in the C source with _ or similar would be a good approach and would probably solve other undiscovered corner cases lurking to catch you out !

[jacob-carlborg]

We're currently adding a suffix, not a prefix, i.e byte -> byte_.

[ciechowoj]

Yes, I guess the pragmatic approach with _ is fine. But it doesn't look 'nice', I'm wondering if renaming all bytes to ubytes isn't a better approach...

[chriscamacho]

what if its not actually unsigned ?

[jacob-carlborg]

I'm wondering if renaming all bytes to ubytes isn't a better approach...

That seems quite brittle. What if byte is part of the API and now that symbol doesn't exist anymore and cannot be used by users of the binding.

[jacob-carlborg]

what if its not actually unsigned ?

Or not corresponding to D's ubyte but it's instead a typedef for int or whatever. (We need to account for stupid things).

[ciechowoj]

If it's not actually unsigned then we can just ignore the alias as in the case of ushort above.

[jacob-carlborg]

What about code like this: typedef int byte?

[ciechowoj]

Yes, I would rename it only if it has form like in the current bug. The stupid cases have to be resolved by other means (_ suffix for example).

[ciechowoj]

Or in case of typedef int byte, rename all the bytes to ints

[ciechowoj]

I wouldn't worried about the symbols, if naming of the types influenced the mangling we are screwed with any renaming (e.g. adding _). I don't think the aliases to primitive types are exported at all.

[jacob-carlborg]

I'm not worried about mangling. What I'm referring to is if the C header defines a symbol called byte and then the user tries to code like byte a = 0;, byte won't exist. But I guess we have the same problem with adding _.

[jacob-carlborg]

I guess the best would be to check all the usages of the typedef and then translate to what the typedef resolves to. This could be bundled in the existing --reduce-aliases flag.