🚀 go-pugleaf

In article <1993Jan20.173950.20531@murdoch.acc.Virginia.EDU> sdm7g@elvis.med.Virginia.EDU (Steven D. Majewski) writes:

[Lots of unassailable stuff - has anyone not worked this out yet? Then:]

|[ (**) This imbalance has other implications. Processor speeds are 
|increasing faster than disk i/o speeds. At what point is uncompressing a
|compressed instruction stream, or incremental compilation of a minimal
|I-code faster than paging in code from virtual memory?  ] 

You lose big on optimisation this way. When compressing the results of
translation tasks (whether that's parsing natural language or generating
code) *for rapid decompression* you want to save the input **and
selected choicepoint oracles**. As the compression folks will tell you,
data modelling is the important part.... So the "I-code" shouldn't be so
minimal, or rather, it should be augmented with the results of any
computationally expensive decisions that the full compiler might once
have made. (The advantage to two streams is architecture independence,
of course).

Damn, here I go giving all my secrets away.... ;-)
----------------------------------------------------------------------
stephen p spackman    +49 681 302 5288(o) 5282(sec)    stephen@acm.org
      dfki / stuhlsatzenhausweg 3 / d-w-6600 saarbruecken 11 / germany
----------------------------------------------------------------------