Issue 5498 - array of elements of subtypes of a common supertype

[yebblies]

arrayExpressionsToCommonType incorrectly uses condexp.e2->type rather than condexp.type to determine the common type of the array.

http://d.puremagic.com/issues/show_bug.cgi?id=5498

[Aatch]

Any reason this hasn't gone through?

[yebblies]

Well, it fails in the autotester. My guess is that something special needs to be done for template lambdas.

[yebblies]

With the following code:
pragma(msg, typeof(true ? (x => x) : ((int x) => x * 2)));

The issue is that in CondExp::semantic e1 has type void and it is special cased to combine void+anything to void.

If this is avoided by excluding TOKfunction expressions, typeCombine/typeMerge fails.

@9rnsr Are there existing functions that could be used to do the type merging for function literals, or does it need to be done from scratch?

[9rnsr]

In basic, if a type t is implicitly required for a expression e, it should be set by calling e->inferType(t) before the semantic of e.
But, typeMerge should look at the both side of operands at the same time. It is yet not implemented enough. IMHO we need to add a case for FuncExp vs FuncExp in typeMerge

[yebblies]

It seems this change is correct, so I'll leave this pull open pending typeMerge improvements.

[andralex]

@yebblies time to revive this?

[yebblies]

This pull is correct, but it will have to wait until typeMerge is updated to merge function literals.

[andralex]

@yebblies ping?

[mrkline]

@yebblies What's the current status of this? As you saw, this came up again in https://d.puremagic.com/issues/show_bug.cgi?id=12283

[yebblies]

Here's a hack that keeps the old behavior when merging two FuncExps.

[yebblies]

Hmm, actually seems to work.

[andralex]

no guts, no glory

[yebblies]

@andralex Did you merge this based on trust or are you actually confident this is the correct way to solve the problem?

Either way, thanks!

[andralex]

Some of both...

[9rnsr]

With the following code:
pragma(msg, typeof(true ? (x => x) : ((int x) => x * 2)));

For the case, I opened an enhancement issue 12605, and as its first step I opened a refactoring PR #3473.