Conditional compilation issue

[azrafe7]

Have not been able to make a general and succint example out of this, but it was reproducing in my code base and this snippet (http://try-haxe.mrcdk.com/#e66Ed) seems to somehow reproduce the issue (although not faithfully):

class Test {
  static function main() {
    trace("Haxe is great!");
    #if js
      trace('x');
    #elif js
      trace('elif');
    #else
      trace('end');
    #end
  }
}

Maybe "elif" was allowed some time ago and has not been supported in latest versions?

[azrafe7]

haxe 4.0.0 preview

As always sorry if I'm misunderstanding something.

[azrafe7]

PS: from google searches it seems that #elif was never supported, but somehow gets through the parser anyhow.

(EDIT: well the code is somewhat succint, but I don't know if it applies to all targets)

[Simn]

It's a valid token as far as the lexer is concerned. All tokens are skipped in inactive branches, so this never comes up as a problem.

Maybe we should indeed error on invalid #-tokens in these cases.

[EricBishton]

You mean: "It's a valid token as any space-delimited set of characters is a [skipped] token in inactive compilation branches," right? It would be nice if inactive CC branches did spit out syntax errors, but then you couldn't #if 0 ... { invalid, half written, code } #endif' any more just to get a test compile. Well.. actually, you could detect '0' as a special circumstance...

@azrafe7 '#elseif' is the token that denotes a second branch for conditional compilation.

@Simn - If you're going to think about creating an error when you see '#unrecognized' in a source file,
I'd love to see #error "message" be a recognized sequence so that you could do something like:

hxcpp_debugger.hx

  #if !hxcpp
  #error "hxcpp debugging can only be used in C++ targets."
  #endif

And that would stop the compiler with an error message:

  hxcpp_debugger.hx: 2: lines 2-2: hxcpp debugging can only be used in C++ targets.

(or whatever the 4.0 error message format will be.)

[Simn]

You mean: "It's a valid token as any space-delimited set of characters is a [skipped] token in inactive compilation branches," right?

That's not true, we still lex normally in inactive branches. The lexer dies on #if false £ #end or any other unrecognized character.

I'd love to see #error "message" be a recognized sequence

#error "message" has been around for a long time.

[EricBishton]

#error "message" has been around for a long time.

My bad. Not sure why I thought it wasn't.

[Simn]

Probably because it's not properly documented.

[azrafe7]

Thanks @EricBishton (figure that out just after posting).

@Simn, yes, why not error out in this cases? (I see you only added the documentation label)