Subject: publishing
From: Todd D Millstein (todd@cs.washington.edu)
Date: Tue Feb 20 2001 - 09:55:43 PST
I'm publishing my system. Here are the details:
*) Added a slice to compute the class hierarchy. Specifically, it
builds up, for each scope, two things:
- A class hierarchy object. This object contains a table mapping
class names to corresponding class_nodes (distinct from the ClassNode
IR node). class_nodes have a set of parent and children class_nodes.
Right now, their interface is fairly meager. More can be added as
necessary.
- A table mapping names to gf_nodes, each of which contains a set of
methods.
The computation of the class hierarchy makes a few assumptions about
the given wil program, which are currently unchecked:
1) Generic functions and classes should be created only in initial
declarations, and they should always be given a name.
2) Methods are only added to generic functions in the same scope.
3) Classes only inherit from classes in the same scope.
It may be necessary to relax some of these assumptions, particularly if we
want to make non-trivial use of class declarations in nested scopes.
*) Initial decl checks now run on the AST during rep-compute and over the
DFG during rep-check.
*) Variables declared "initially" or "once" are now immutable -- the
syntax does not allow you to say "var" for these declarations. This
change necessitated some updates to the Jil code and the Java wil-stdlib,
so you should update these.
*) The AssnNode hierarchy, which represents assignment nodes in the
IR, has been refined. The gist is that we used to have something called
DelayedInitNode, which was used to represent two things:
1) An initializing variable declaration (InitNode) in the AST gets
translated into a VarDeclNode followed by a DelayedInitNode in the CFG.
2) Any initializing assignment (signified with the ":==" syntax) to a
previously-declared variable gets translated into a DelayedInitNode.
We now have two different classes for these uses of DelayedInitNode:
InternalInitAssnNode for #1 above, SourceInitAssnNode for #2 above. It is
necessary to distinguish these uses for the checks on initial
declarations.
*) To compute the class hierarchy, it was necessary to make InitNodes
multiply inherit from WriteNodes (which they already inherited from)
and VarDeclNodes. That being the case, it probably makes sense to
represent InitNodes directly in the CFG now instead of translating them
as described above.
*) Various cleanups to the initial decl checks.
*) Some small changes to Keunwoo's exception library, to allow it to
typecheck. I originally wanted to use exceptions, but then, with
Sorin's help, I found a nice way to do things without them.
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Tue Feb 20 2001 - 09:56:03 PST