Fix reduce regression 13304.

[monarchdodra]

https://issues.dlang.org/show_bug.cgi?id=13304

Also cover test case 10709.
https://issues.dlang.org/show_bug.cgi?id=10709

Truth be told, I'm not yet entirely sure what went wrong, and why this fixes the issue. It seems that the compiler was trying to actually CTFE-evaluate the fun call in s = fun(s, e);, whereas the is(typeof(s = fun(s, e)) merely checks the semantics are correct.

I'm still trying to reduce the actual cause, but in the mean time, this (should) fix a regression.

[monarchdodra]

Groan... why is the compiler trying to link in a function that never gets called?

[monarchdodra]

I deactivated the unittest that is causing the failure (the test itself works, but creates a link error). I'll have to try to reduce and submit.

[quickfur]

Should we wait until you've finished reducing it?

[mihails-strasuns]

I think we should

[monarchdodra]

I think we should

Well, it is a regression, so faster is better. What's more, it seems relatively easy to trigger.

Furthermore, the code that's failing is merely there trying to give a good diagnostic eagerly, rather than an obscure one in internal later. If worst comes to worst, we could just strip it out entirely.

I'll have some time Friday to see what I can do.

[mihails-strasuns]

If it has resulted in immediate unittest failure how likely it is that it will just create another regression with some of the user code out there? Quite likely in my opinion.

[monarchdodra]

Failure actually had nothing to do with my code:

auto red()()
{
    return foo();
}

static int foo();

void main(string[] args)
{
    int[] data;
    assert(is(typeof(red()))); //Temporarilly disabled.
}

I just removed the unittest then. Should be fine now.

[monarchdodra]

https://issues.dlang.org/show_bug.cgi?id=13350

[mihails-strasuns]

ah must be yet another issues with speculative template instantiations
ok, lets go ahead with this one

[mihails-strasuns]

Auto-merge toggled on