Stream: t-compiler/wg-incr-comp

Topic: finding ICEs automatically

view this post on Zulip matthiaskrgr (Mar 14 2021 at 13:36):

So I wondered if we could somehow automatically find incremental ices using the spaceheater approach (throwing cpu at the problem).
So far I have been iterating through all commits of a repo and building them one by one, but we can also use clippy to modify the sources!
So I came up with this command chain:

cargo build ; cargo clippy --all-targets --fix -Zunstable-options  --allow-dirty -- --cap-lints warn  -Wclippy::pedantic  ; cargo  build

We first build the crate, then apply clippy lint suggestions and build again to see if the incremental verification catches anything. :)

view this post on Zulip bjorn3 (Mar 14 2021 at 17:36):

I believe there was a program to automatically run incr comp through all commits in a project. It was used when incr comp was still unstable.

view this post on Zulip Wesley Wiser (Mar 15 2021 at 14:36):

I think that was

view this post on Zulip mw (Mar 16 2021 at 10:39):

Yeah, was what we actually used. It required a lot of maintenance because not every commit would build.

view this post on Zulip pnkfelix (Mar 17 2021 at 12:23):

startup idea: a globally-accessible database recording whether given commits are expected or unexpect to build. (potential features include: recording how may of each; what hosts (i.e. os+cpu) worked or didn’t; recording uploaded diffs against commits to “fix” them) ...

view this post on Zulip pnkfelix (Mar 17 2021 at 12:24):

(I imagine many some search tools would benefit from knowing such info ahead of time.)

view this post on Zulip pnkfelix (Mar 17 2021 at 12:25):

(further next steps beyond that: make a proof-of-work currency based on the results, called “buildcoin”)

view this post on Zulip pnkfelix (Mar 17 2021 at 12:27):

((Last part is a joke: I know part of proof-of-work is ability to verify result, which is not tenable here))

Last updated: Oct 21 2021 at 21:46 UTC