Re: rep clause in generics

[dewar@merv.cs.nyu.edu (Robert Dewar)]

Robert Duff said

<<(As a compiler writer, I find it annoying that 'Size is *ever* static --
it messes up one's compiler design.  You'd like to say 'Size is a
machine-dependent thing, and relegate it to the back end, or some such
late, machine-dependent, phase.  Unfortunately, its value is needed
during overload resolution: e.g.,>>

Interestingly, with GNAT, we have found that the rules about whether 'Size
is static or not are essentially perfect from the point of view of the
organization of tasks between our front and backend. The frontend has no
trouble knowing what Integer'Size is, but it would be hard for it to
know what Rec'Size was even for a record with a single integer field.
If such sizes were static, we would need a completely different design
which would entangle the front end and backend much more extensively.

But having sizes of simple scalar stuff be known to the front end is no
problem for us. (of course I can certainly see why it might cause toruble
in some other design, but given that all you need is a relatively small
table -- not a complex target dependent algorithm -- to know what the
sizes are of primitive data, i don't see it as a problem.