Feature request: make types marked with exportc always emitted

[cooldome]

Currently compiler eliminates all unused types. Make it possible to force type definition to be emitted.
Pragma exportc seems to be suitable. Another option is used pragma.

[Araq]

exportc is what should be used.

[dom96]

Is this an issue? What's wrong with exportc?

[cooldome]

Currently exportc has no effect on types, types declarations will be eliminated if compiler will not find the evidence of type usage in the generated procs

[dom96]

Can confirm that this is annoying. Just tried this:

{.push exportc, used.}

type
  DisplayKind* {.exportc.} = enum
    Text, Circle, RawPixels

  RGB565* {.exportc.}  = uint16

  Display* {.exportc: "Display", used.}  = object
    case kind*: DisplayKind
    of Text:
      text: cstring
      textColor: RGB565
    of Circle:
      radius: uint8
      circleColor: RGB565
    of RawPixels:
      data: array[16, array[16, RGB565]]
    pos: tuple[x, y: uint8]

  AnimationKind* {.exportc.}  = enum
    Scroll,

  Animation* {.exportc.} = object
    case kind*: AnimationKind
    of Scroll: discard

The C code does not have these types.

[timotheecour]

Exportc affects mangling and shouldnt be conflated with feature request

[HugoP707]

I would be fine with it
what do you suggest instead?

[timotheecour]

Used (extending its current meaning) or export

[disruptek]

Based on @dom96 repro, I don't see how this isn't a bug and not a feature request.

[timotheecour]

dead code elimination (DCE) (whether for procs or types) is useful, and exportc + exportcpp should IMO only affect mangling, not dead code elimination.
Granted, currently there is an inconsistency between types and procs:

• proc fn(){.exportc.}=discard inhibits DCE
• type Foo {.exportc.} = object honors DCE

if we change semantics to make exportc types inhibit DCE, we at least need a new pragma to honor DCE.

Alternatively (IMO, preferably), we can change semantics for exportc procs to honor DCE (with a legacy transition switch), and make {.used.} inhibit DCE. Currently IIUC used is only for modules to prevent import unused.

[Clyybber]

{.exportc.} should definitely prevent DCE, after all its use is to export something to C, regardless of wether the Nim side uses it or not.

[Araq]

Well if you export to C, you want a stable name too so this conflation of concerns is entirely reasonable. Not sure what exporting a type to C brings us though. If you export a proc, it ends up in your binary, if you export a type, it ends up in a single C file, what's the point? We don't generate C header files.

[dom96]

We don't generate C header files.

We do. There is a --header flag that can be passed to get a header file.

[HugoP707]

It was not supposed to be used and was going to be deprecated iirc

[timotheecour]

Why? Any links to where this is discussed?

[Araq]

It's one of these features we cannot maintain and the code generator wasn't build for... It was useful to somebody once and now is a burden. Some lessons to be learned from this here...

[timotheecour]

deprecating --header should be the type of thing that requires an RFC, IMO.