[dmd-beta] dmd 2.051 beta

[Walter Bright]

Ok, this is a *very* hard problem to track down, as it is subtly 
dependent on the order of instantiations. The problem is that 
std.array.Appender has 3 put() member templates. We are only concerned 
with the first and last one. If we pass an argument of type:

    char[]

to put(), it winds up tripping the recursive constraint detection in 
template.c, causing the constraint to fail, and the put() template is 
not instantiated.

The recursive part is not actually in the template constraint, it is in 
the body of put(). So it only happens when semantic3() is called on the 
template instantiation while inside the constraint of another 
instantiation. The reason for the erratic nature of the bug happening is 
that if semantic3() has already been called for the instantiation, it 
isn't called again. Also, sometimes the call to semantic3() is deferred 
to another module, or is deferred for other reasons, and so the error 
doesn't happen.

I thought about fixing the template so it wasn't recursive, but then 
thought this mysterious error would crop up elsewhere and confound the 
next person. So that was out. I still thought it best to eagerly attempt 
to do semantic3(). But bugzilla 4072 is the reason for the recursive 
protection code in the first place.

Finally, I decided upon a rather hackish solution - let the recursion 
detection code run a couple times before it trips. It doesn't seem 
right, but I can't think of anything better at the moment.