Stream: t-compiler/help

Topic: cargo-bisect-rustc


Paul Faria (Aug 28 2019 at 04:57, on Zulip):

I'm trying to use cargo-bisect-rustc to figure out which nightly (from a year ago) triggers a certain error in rustdoc. I'm passing a script to the command, with the contents:

#!/bin/sh
cargo rustdoc

And the part that's getting me stuck is that the command always returns that the --start param is regressing. This is occurring even though I can manually install that nightly with rustup, run cargo rustdoc and get a successful output. Any idea why it would report the error incorrectly?
I'm running it like so:
cargo-bisect-rustc --test-dir=docs_crash --script=./bisect-script.sh --start=2018-07-10 --end=2018-09-01 --preserve

Paul Faria (Aug 28 2019 at 04:57, on Zulip):

It gives the output:

checking nightly-2018-07-10
std for x86_64-unknown-linux-gnu: 51.82 MB / 51.82 MB [=======...======] 100.00 % 3.56 MB/s
the --start nightly has the regression
Paul Faria (Aug 28 2019 at 05:00, on Zulip):

And this is the rustup version:

$ rustup override set nightly-2018-07-10
...
$ cargo --version
cargo 1.29.0-nightly (af9e40c26 2018-07-05)
$ cargo rustdoc
    Checking docs_fail v0.1.0 (file:///home/pfaria/projects/docs_crash/lib/docs_fail)
 Documenting docs_crash v0.1.0 (file:///home/pfaria/projects/docs_crash)
    Finished dev [unoptimized + debuginfo] target(s) in 0.33s
nikomatsakis (Aug 29 2019 at 15:41, on Zulip):

cc @simulacrum

nikomatsakis (Aug 29 2019 at 15:41, on Zulip):

(I have no idea :)

simulacrum (Aug 29 2019 at 15:42, on Zulip):

Hm, yeah, that does seem strange. I think there should be a way to show output (--show-output perhaps?) maybe that would point at the problem

simulacrum (Aug 29 2019 at 15:42, on Zulip):

I seem to recall the script needing to use $CARGO and such though

simulacrum (Aug 29 2019 at 15:43, on Zulip):

maybe I'm remembering wrong though

simulacrum (Aug 29 2019 at 15:43, on Zulip):

@Paul Faria ^

Paul Faria (Aug 29 2019 at 16:08, on Zulip):

Thanks, I'll try that out when I get home tomorrow

Paul Faria (Aug 30 2019 at 14:19, on Zulip):

So using --prompt helped me realize that cargo was using an unstable option (color) from way back when. I had to add -Z unstable-options to my script. This narrowed it down to 2018-08-21(I think?) It was between commits 20dc0c5c7 and 3bc2ca7e4. I'm trying to bisect to the exact commit manually (since the builds aren't available anymore, I'm trying to manually build the minimal rustdoc to trigger the issue). However, and maybe no one can help here since it's so old, building ./x.py --stage=1 src/tools/rustdoc src/libstd takes well over an hour (seems to get stuck on building core). Any idea why that might be the case, or what I can check (config.toml or otherwise) to speed that up? I'll have this running while I work so it's not a big deal if it takes a long time.

Last update: Nov 11 2019 at 23:15UTC