[pypy-dev] papers

[Douglas McNeil]

While I'm waiting to hear back on the bug tracker about whether some odd
binascii behaviour is intentional, I was doing some reading.  Some of the
work by these guys

http://www.cs.washington.edu/research/progsys/wasp/wasp.html

is very interesting, especially their latest

Automated Soundness Proofs for Dataflow Analyses and Transformations via
Local Rules
http://www.cs.washington.edu/research/projects/cecil/pubs/popl05.html

and Automatically Proving the Correctness of Compiler Optimizations
http://www.cs.washington.edu/research/projects/cecil/pubs/pldi03.html

(There's already been mention on the list of another of their alumni,
Grove's thesis on interprocedural optimization.)

BTW, structmodule is complete in the sense it passes test_struct.py on my
box.  Unfortunately (as mentioned earlier) since I don't know of any way
to get access to certain C information it fails when I turn off my
HAVE_NATIVE flag indicating the hardcoded values are correct.. including,
frustratingly, for the default case.  There are also some 754 issues
involving NaNs I just threw up my hands at, though I have a hack for some
infs.

I've also discovered that it's relatively easy to get the behaviours for
well-formed inputs to match when the algorithm is well-specified, harder
to match half-specified behaviour, and brutal to get the errors to match..
like an idiot I tried to implement this one algorithm from the RFC spec I
thought it was respecting.  Was that ever a mistake..


Doug

--
Queen's University Astronomy Research Group
Planetary Dynamics Division			"We create worlds."