[Feature Request] Non-constant static initialized nested delegate literal expression better error message about ldc HOT 13 CLOSED

It doesn't have anything to do with the array syntax, it's just an invalid non-constant initializer for a global as expected. I've added a full example in https://issues.dlang.org/show_bug.cgi?id=24469#c2. And it's actually a DMD regression (since v2.079).

from ldc.

Okay, I'll put there, thanks for the help :)

from ldc.

@kinke Some update on that: Although it compiles with DMD, it starts crashing randomly on a later step function return... Maybe this have an alternative?

from ldc.

Well in that case LDC might be doing the right thing, contrary to DMD. ;) - Your code somehow requires the delegate as a static constant, as part of an initializer for a global or so. And that cannot work if the delegate uses an outer context, e.g., the outer gemName parameter in your snippet, as that context is obviously dynamic.

from ldc.

You can use -vv when compiling with LDC, to see what it's doing right before the error; that could help.

from ldc.

Well in that case LDC might be doing the right thing, contrary to DMD. ;) - Your code somehow requires the delegate as a static constant, as part of an initializer for a global or so. And that cannot work if the delegate uses an outer context, e.g., the outer gemName parameter in your snippet, as that context is obviously dynamic.

I opened an issue on DMD: https://issues.dlang.org/show_bug.cgi?id=24469

What is a static constant delegate? I'm pretty sure I'm not using any kind of fancy keywords with delegates. Maybe is it possible to create a better error message like showing why have it turned to be required to be static constant?

I'll test with -vv and comment again

from ldc.

@kinke I was able to reduce the error after looking in all my code:

struct ExistenceChecker
{
    ///Optional. 
    bool delegate() checkExistenceFn;

}
ExistenceChecker makeExistenceChecker(string gemName)
{
    return ExistenceChecker(()
    {
        return executeShell("gem list | grep "~gemName).status == 0;
    });
}

ExistenceChecker[] chks = [
    makeExistenceChecker("test")
];

from ldc.

Maybe is it possible to put in the error that because of the array syntax there, it can't access the outer variables?

from ldc.

I'll put this as a feature request for showing which part requested for static initialization of a non constant delegate

from ldc.

I suggest you ask for better diagnostics (agreed; what would IMO be very helpful is knowing where the problematic initializer is) upstream, not here.

from ldc.

Wrt. how to work around this - create the array at runtime, not at compile-time, in a module ctor or so.

from ldc.

@kinke Btw I was checking that:

int delegate() makeClosure(int start)
{
    return (){return start+1;};
}

auto b = makeClosure(1);

void main()
{
}

You can see it shows from where it comes from. Does that also requires to be asked on upstream? Because the upstream doesn't even show an error 🤔

from ldc.

With run.dlang.io's current DMD v2.105, it shows the same CTFE error as LDC for your example above, nicely including the initializer line:

onlineapp.d(3): Error: closures are not yet supported in CTFE
onlineapp.d(6):        called from here: `makeClosure(1)`

This doesn't come from LDC directly, but the frontend's CTFE interpreter. The original error with the struct container seems to slip through there, and that's most likely where the fix should land (so in the DMD frontend, that's why I asked for the upstream issue).

from ldc.