[cfe-commits] [cfe-dev] [review request] Removing redundant implicit casts in the AST, take 2

[John McCall]

On Feb 23, 2012, at 6:42 PM, David Blaikie wrote:
> Sorry, I'll provide some more context - AnalyzeImplicitConversions
> itself does "IgnoreParensImpCasts" as its first thing - but not before
> recording the QualType of the outer Expr (lines 4202, 4203). Many of
> the diagnostics work simply by comparing the type of the outer
> expression with the type of the inner one - thus they fired /all/ if
> you didn't have that IgnoreParensImpCasts before the
> AnalyzeImplicitConversions call - because they looked just like
> implicit conversions (they /were/ in the AST). Now that explicit
> conversions aren't implemented as implicit ones this problem doesn't
> exist.
> 
> In fact when I made this change I was hoping there were cases where an
> explicit conversion could contain an implicit one as you showed - but
> further, that such cases might have real warnings. Turned out removing
> that outer IgnoreParensImpCasts didn't cause any clang tests to change
> at all... nice, even if it didn't allow us to catch new cases.
> 
> In any case I suspect it's the right thing - there's no reason for us
> to be ignoring those implicit conversions except that we were doing it
> to support our not-so-great AST representation of explicit
> conversions.

There are a few situations where we implement a cast with multiple
implicit conversions.  One example would be something like
  (_Complex float) 0
which is an IntegralToFloating conversion followed by a
FloatingRealToComplex.

John.