dmd 1.068 and 2.053 release

[Don]

Jonathan M Davis wrote:
> On 2011-05-15 03:50, Joel Christensen wrote:
>> Looks like enum's are tighter (eg. 'enum last = media[ $ - 1 ];' doesn't
>> work now). It was working in 52. I had heard it might be relaxed, not
>> tightened. I get the error, 'cannot be read at compile time'.
>>
>> Also immutable imstr = "test"; printf( toStringz( imstr ) ); wasn't
>> working at first, but works now for some reason.
>>
>> Good to have an update though.
> 
> A lot of CTFE stuff was rewritten. What all of the implications of that are, I 
> don't know, but according to Don (who did the rewrite), there are cases which 
> compiled before but didn't generate correct code. I don't know if there were 
> any cases which compiled which were supposed to be illegal.

There are VERY MANY cases which compiled before, which were supposed to 
be illegal. The compiler used to accept a variable where it needed a 
compile-time constant!

 > Regardless,
> because there was a major rewrite for CTFE, the risk of CTFE bugs or 
> behavioral changes is higher than is the case for most releases.

To clarify:
Two massive fixes were made, which are independent of each other:
(1) CONSTANT FOLDING: any case where a compile-time value is required 
now MUST be a compile-time value. If a compile-time value is not 
required, there is no attempt to interpret it. This fixed many 
accepts-invalid bugs.
(2) CTFE: array literals no longer use copy-on-write (which gave totally 
wrong semantics). This fixed many wrong-code bugs.

Fixing (2) also allowed a huge number of CTFE bugs to be fixed.

This particular example is a consequence of (1), and has nothing to do 
with the CTFE changes.