🚀 go-pugleaf

In article <5430@balrog.ctron.com>, smith@ctron.com (Larry Smith) writes:
|> The correct way
|> to implement this semantic is to always pass aggregate types as pointers and
|> to emit code at the begining of each routine to branch to the end, where
|> the compiler will be after noticing what parameters that were modified during
|> the compilation.  These parameters can be copied onto the stack and the
|> argument pointer adjusted, and the routine started with a backwards jump to
|> the real begining of the routine.  This allows one-pass compilation, removes
|> the necessity for aggregate copies where such copies are not needed to enforce
|> standard semantics.
|> 

Please don't do this! If you really object to keeping a representation of the
current procedure in memory so the code generator can generate the proper code,
do some back-patching, put the code in to copy the parameter and have
the compiler write NOPs over if is the parameter is unchanged. I'd rather
the compiler was slightly slower than have my cache/pipeline emptied by
jumping about whenever I call a procedure.

-- 
Ian W. Moor		+71 589 5111 x 5039
iwm@doc.ic.ac.uk   Department of Computing, Imperial College.            
"Experts from BNFL have advised the Russians to rename Chernobyl as Sellograd
 and open a tourist centre"