Subject: Re: tranformations and node visiting
From: Keunwoo Lee (klee@cs.washington.edu)
Date: Thu Mar 01 2001 - 19:40:05 PST
BTW, Sorin and I talked about this and concluded it was probably a bug.
I've made a workaround where we accumulate the graph's initial node
collection in a separate list and iterate over that list, thereby avoiding
the mutation-while-iterating problem.
My code, which was formerly breaking, now appears to be running smoothly.
There's probably a better way to do this that doesn't involve allocating a
list that's the length of the original graph, but right now we just need a
quick fix.
I think I will commit just transform-implementation.cecil (should have no
dependencies on other changes I've made) tonight, if towers builds without
dying.
~k.lee
On Thu, 1 Mar 2001, Keunwoo Lee wrote:
> I put forth the following question about node transformation code:
>
> Given:
>
> a. apply_transformations (in transform-implementation.cecil:15) uses
> graph.nodes_do( ... ) to visit each node and determine if it needs a
> transformation.
>
> b. nodes_do, as defined for most slices, ultimately uses
> visit_graph_nodes to iterate over all nodes.
>
> c. visit_graph_nodes uses each node's succ_nodes to determine where to
> traverse next.
>
> What is to prevent apply_transformations from
>
> 1. Visiting a node which it has just patched into the graph (because that
> node is now a succ_node of some other node in the current graph).
>
> 2. Using this node's ID to index into the list of transformations.
>
> 3. Picking a bogus transformation to perform on that node.
>
> ?
>
> ~k.lee
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Thu Mar 01 2001 - 19:41:04 PST