Re: Linker ... still useful ?

dmason@uwaterloo.ca (Dave Mason)
Sat, 15 Oct 1994 19:06:03 GMT

          From comp.compilers

Related articles
[10 earlier articles]
Re: Linker ... still useful ? gnb@bby.com.au (1994-09-29)
Re: Linker ... still useful ? andrew@cee.hw.ac.uk (1994-09-30)
Re: Linker ... still useful ? marks@orb.mincom.oz.au (1994-10-05)
Re: Linker ... still useful ? ok@cs.rmit.oz.au (1994-10-06)
Re: Linker ... still useful ? monnier@di.epfl.ch (1994-10-07)
Re: Linker ... still useful ? baynes@ukpsshp1.serigate.philips.nl (1994-10-10)
Re: Linker ... still useful ? dmason@uwaterloo.ca (1994-10-15)
| List of all articles for this month |

Newsgroups: comp.compilers
From: dmason@uwaterloo.ca (Dave Mason)
Keywords: linker
Organization: School of Computer Science, Ryerson Polytechnic University
References: 94-09-122 94-09-155
Date: Sat, 15 Oct 1994 19:06:03 GMT
X-Url: <http://plg.uwaterloo.ca/~dmason>

Sorry to follow up to this so late, I've been busy and gotten behind
on news...


Almost all this discussion about linking has missed out on an important
current research area: early linking/late optimization/late code
generation. Particularly in ``new'' programming languages (Smalltalk,
SML, Self, other dynamically typed OO languages, other functional
languages) the traditional compilation to machine code followed by linkage
produces inferior run-time performance (because of the huge number of
small functions).


The research literature (David Wall at DEC, Dave Unger at Sun, Dave Keppel
at Washington, Craig Chambers at ?Washington) includes on-the-fly code
generation, conventional linkage with post-link optimization. Newer work
by myself and others is looking at linkage of intermediate code, followed
by optimization, and finally code generation. This has many advantages,
but one relevant to this discussion is that (with a sufficiently general
intermediate language) it can handle the multi-language impedance match
problem quite easily.


The research challenges include finding appropriate intermediate
representations and showing that the resulting software development cycle
isn't *too* slow to be useful. I personally hope to have convincing
results in a year or so.


../Dave
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.