Questions about `rustc_mir_dataflow` analyses

help
1

I am trying to write a backward rustc_mir_dataflow analysis.

Before iterate_to_fixpoint is called, I intentionally remove some edges from the graph (using MirPatch) to prevent data from flowing along those edges.

But this creates a problem, because the rustc code that initializes an analysis's work queue assumes that all blocks are reachable. Because some edges have been removed, not all blocks are reachable.

My questions are:

  1. Is it a "bug" (for lack of a better term) that the work-queue-initialization code assumes all blocks are reachable?
  2. If not, is there some other way to prevent data from flowing along an edge, without removing the edge?
2

This question about rustc's implementation may be better suited to IRLO. (Leave a note here if you cross-post.)

1 Like
3

Thanks, @quinedot.

After looking into this further, I convinced myself that it is a "bug" in the API. So I submitted a PR to fix it.

Still, I agree that the internals forum would have been a better place to ask this question. I'll ask similar, future questions there.

Thanks again.

4

You might also be interested in the #t-compiler/help stream on zulip.

5

Thanks, @scottmcm. I appreciate the recommendation. :pray:

6

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.