regression 1.50: Duplicate symbol linker error

[rylev]

In the latest crater run, a crate that compiles on 1.49, did not compile on 1.50 with a linker error you can see here.

@rustbot modify labels: +regression-from-stable-to-beta

[nagisa]

Seems like a not-genuine-regression, its just a case of cargo building and linking two different versions of the same crate (secp256k1), which appears to export no_mangle symbols if the error message looks right.

[wesleywiser]

Discussed in the release triage meeting. The above theory sounds plausible but we want to make sure there's not an actual issue here.

One possibility @Mark-Simulacrum mentioned is that our rules around no_mangle symbols being exported has changed somehow in this release. @petrochenkov do you happen to know if something like that has changed in this release?

[petrochenkov]

No, I didn't follow the no_mangle changes.

[apiraino]

Removing I-prioritize label as it looks this is on the radar along with the other crater run issues

[Mark-Simulacrum]

I can confirm this is a regression from 1.49. However, I'm having trouble bisecting this - nightlies seem to fail with the linker error for a while back; I ran bisect up to 2020-07-20.

Not sure what to do - maybe worth pinging cleanup group to try to minify at least.

[apiraino]

@rustbot ping cleanup

[rustbot]

Hey Cleanup Crew ICE-breakers! This bug has been identified as a good
"Cleanup ICE-breaking candidate". In case it's useful, here are some
instructions for tackling these sorts of bugs. Maybe take a look?
Thanks! <3

cc @AminArria @camelid @chrissimpkins @contrun @DutchGhost @elshize @ethanboxx @h-michael @HallerPatrick @hdhoang @hellow554 @imtsuki @JamesPatrickGill @kanru @KarlK90 @LeSeulArtichaut @MAdrianMattocks @matheus-consoli @mental32 @nmccarty @Noah-Kennedy @pard68 @PeytonT @pierreN @Redblueflame @RobbieClarken @RobertoSnap @robjtede @SarthakSingh31 @shekohex @sinato @smmalis37 @steffahn @Stupremee @tamuhey @turboladen @woshilapin @yerke

[steffahn]

https://github.com/devrandom/utxo-oracle does compile for me with stable i.e. 1.49. If fails with nightly with beta but also with 1.48 or 1.47.

[steffahn]

So this is not a regression, AFAICT.

With stable 1.49 (e1884a8) as well as current master (613ef74), the linking error when compiling utxo-oracle only appears with a compiler that was built with the channel = "dev" or channel = "beta" option set (in config.toml). Using channel = "stable" does not result in a linker error. (I didn’t actually test "beta" on e1884a8.)

I also tested current beta (1cd0303) and it doesn’t produce the problem with channel = "stable", so it seems likely that 1.50 won’t regress this build.

Finally, as a double check, 1.48 (7eac88a) does produce the linking failure for me even with channel = "stable". I didn’t go further yet in bisecting where the error disappeared between 1.48 and 1.49.

[Mark-Simulacrum]

Interesting. Can we try to figure out what is causing that dependence on the channel?

[pnkfelix]

Visiting for P-high review.

There seems to be a lot of indication that this is not a true regression, but the oddities in terms of how the channel name is affecting the results here is worrisome.

Self-assigning for follow-up investigation.

[pnkfelix]

Also, review determined this no longer warrants P-high priority, not without stronger signal that this represents an actual regression.

@rustbot label: -P-high +P-medium

[jackh726]

No additional info here in about 3 years, closing for no reproduction.