🚀 go-pugleaf

Mark Hamburg (mhamburg@mv.us.adobe.com) wrote:
: Okay, Simon Johnston just pointed me to Intermetrics efforts on an Ada 95
: to Java bytecode compiler, so I probably should elaborate on my previous
: posting:

: First, note that I said "readily be done".

: One could actually implement Oberon's semantics on the Java virtual
: machine, but the performance characteristics would be horrendous.

: For example, since the Java virtual machine doesn't allow stack allocation
: of records or arrays, you would just have the compiler emit code to heap
: allocate them. This will behave essentially the same but will result in far
: more frequent garbage collection than in Oberon.

Somewhat more, but there's the potential to be clever. 

: Similarly, since Java doesn't support embedding structured types inside
: other structured types, you could simply silently allocate a separate heap
: block for each embedded construct. An array of N records results in N+1
: heap allocations, one for the array and N for the individual records.

Make new aggregate structured types. 

: VAR parameters make things particularly nasty. Since the Java virtual
: machine has no support for passing values by reference, you have to pass an
: object pointer. This means that even primitive values like integers have to
: be stored as heap allocated objects if you want to pass them as VAR
: parameters. Since you won't know whether a public record field will ever be
: passed as a VAR parameter,  all public record fields will need to be stored
: as separate heap objects.

These could be wrapped at time of call. 

: Mark