Issue 5896 - const overload matching is succumb to template parameter one

[9rnsr]

http://d.puremagic.com/issues/show_bug.cgi?id=5896

[WalterBright]

I don't think this solution is correct. The 'this' and the template parameters should participate equally in the overloading process.

[9rnsr]

Improved. Separate and prefer 'ethis' matching from tiargs matching instead add match level.

[WalterBright]

This is much better. But I disagree with the rationale for preferring an ethis match. The ethis matching should act as just another parameter as far as overload matching goes.

[9rnsr]

The 3rd try.
If the results of td->leastAsSpecialized() are same, run semantic, get FuncDeclaration, and compare results of tf->callMatch and fd->leastAsSpecialized().

1st try 9rnsr@9967e84
2nd try 9rnsr@4dd398d

[9rnsr]

Sorry, I found a test-case that does not work properly. 3rd try is invalid.

[9rnsr]

3rd re-try. Deducing function template process has two steps to shrink overload set.

1. Shrink template overloads by specified template arguments.
2. Shrink template function overloads by inferred template arguments from function arguments.
   I think that reason of issue 5896 is by mixing them.

[9rnsr]

Added fix for bug 4953.
As I talked in here, I think #239 is the right way for bug 4953. And this pull doesn't need to increase matching levels, as like #269.

@WalterBright , I'm sure it is time to fix these two bugs.