Skip to content

TODO: report progress across outputs during a long LEC - #272

Draft
oharboe wants to merge 1 commit into
keplertech:mainfrom
oharboe:solver-progress
Draft

oharboe wants to merge 1 commit into
keplertech:mainfrom
oharboe:solver-progress

Conversation

@oharboe

@oharboe oharboe commented Oct 6, 2026

Copy link
Copy Markdown
Contributor

A request with a failing test, not a fix. The flag name is only a proposal.

The whole-design LEC builds one miter over every output and makes one SAT call. On a 256-word, 21-port register file (33,888 outputs compared), it built the miter in 4 minutes, then logged nothing for over half an hour, when we stopped it. A long solve can't be told from a hang, or from one that's nearly done.

KeplerCliSubprocessTests.LecReportsProgressAcrossOutputs in MiterTests runs a two-output pair with --po-progress and expects the miter log to say checked 2 of 2 outputs. It fails today with Unrecognized option: --po-progress and runs in milliseconds.

A possible shape: with the flag, outputs are checked in batches (one cone, or n at a time), with a line per batch. The existing per-output analysis after a SAT result already walks the outputs one by one.

🤖 Generated with Claude Code

One miter over every output is one SAT call, and it logs nothing for as
long as it runs. With a --po-progress flag the outputs would be checked a
cone at a time and the miter log would count them.
LecReportsProgressAcrossOutputs asks for that on a two-output pair. It
fails today (Unrecognized option: --po-progress); the flag name is a
proposal.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant