Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!udel!princeton!cs!atomic.Princeton.EDU!nr
From: nr@atomic.Princeton.EDU (Norman Ramsey)
Newsgroups: comp.lang.modula3
Subject: making pickles smaller
Message-ID: <7927@rossignol.Princeton.EDU>
Date: 27 Mar 91 16:43:18 GMT
Sender: news@cs.Princeton.EDU
Organization: Princeton Computer Science Dept
Lines: 28
Originator: nr@atomic.Princeton.EDU


I have a program that is generating pickles that are too large.  I can
imagine two strategies for making them smaller:
  a) increasing sharing
  b) where sharing is unimportant, using WriteBytesProc and ReadBytesProc

As an example of the first stragety, since TEXTs are supposed to be
immutable, I would like to store a single REF for both s and t
whenever
  TYPECODE(s)=TYPECODE(TEXT) AND TYPECODE(s)=TYPECODE(TEXT) AND Text.Equal(s,t)
I suspect that this change would realize a substantial savings in my
particular pickles, but I can't think of any way of using ConvertProcs
to implement this strategy using the current Pkl interface.  By the
time I know I have a TEXT, it's too late to change it.  Trying to find
all the TEXT-containing REF types might be very painful.

I might apply the second strategy to the type REF CHAR, resulting in a
savings of four bytes per value pickled.

Question 1: are there any general techniques that can be used to
            increase sharing, so that distinct references that point
	    to isomorphic structures are merged into a single reference?

Question 2: how can I dissect a pickle to determine which types are
	    responsible for the most space usage?  Are there any tools?
-- 
Norman Ramsey
nr@princeton.edu