Issue 13010: use VRP for static asserts

[lionello]

https://issues.dlang.org/show_bug.cgi?id=13010

[yebblies]

static assert( (assert(0), true) );

[lionello]

@yebblies Good point :)

Instead, I've followed @9rnsr 's recommendation and am using VRP during constant folding of comparisons. This is actually much nicer: it can be used instead of __traits(valueRange...) to check the range of a variable. It's also less prone to breakage since the range of a variable will only get narrower as the compiler gets smarter.

[lionello]

Yeah, this is not right. Fixing.

[yebblies]

Doesn't this have the same problem? You still need to evaluate the expression, because evaluating might fail.

[lionello]

Constant folding ("optimize") and VRP don't rely on CTFE and don't fully interpret the expression.

[lionello]

(Ugh, still fails. Shouldn't do this at 3am.)

[yebblies]

Constant folding ("optimize") and VRP don't rely on CTFE and don't fully interpret the expression.

Yes, but they can't fold away the failure even if they know the range of the value it returns.

[lionello]

@yebblies I'm not too worried about that case since it can be figured out. I'm more worried about supposedly "dead code" that's being found in Phobos because dchar.max is 0x10FFFF. In real life, dchar can be higher than that.

[andralex]

ping

[lionello]

Needs more tests.

[lionello]

Superseded by #5229