removes syntax diagnostic err code uniqueness check

[pnkfelix]

This commit removes the syntax diagnostic validation as it's been plaguing the bots for two separate reasons and needs a quick fix for now:

1. It's possible to read the stale metadata of previous builds, causing spurious failures that can only be solved by blowing away tmp.
2. The code does not currently handle concurrent compilations where JSON files in the metadata directory may be half-written and hence contain invalid JSON when read.

This validation should come back, but it should likely be part of a lint pass run at the very end of the build that doesn't run into consistency or concurrency problems.

---

This commit is a revised version of PR #25592; this approach differs in that it continues to output the JSON metadata, so that the error index page continues to work (assuming that people do not accidentally put in duplicate entries in the meantime).

[rust-highfive]

r? @sfackler

(rust_highfive has picked a reviewer for you, use r? to override)

[michaelsproul]

Looks good 👍

[dotdash]

@bors r+

[bors]

📌 Commit 760daf1 has been approved by dotdash

[dotdash]

@bors p=1

unblock those bots

[michaelsproul]

Sorry for breaking the build 😬

[alexcrichton]

@bors: p=100

No worries @michaelsproul!

[bors]

⌛ Testing commit 760daf1 with merge a32db93...

[bors]

💔 Test failed - auto-mac-32-opt

[alexcrichton]

@bors: retry

[bors]

⌛ Testing commit 760daf1 with merge 35b01e2...

[bors]

💔 Test failed - auto-linux-64-x-android-t

[alexcrichton]

Note that the failure there is because there was an error outputting the metadata. I think for now it may just be best to remove it and then re-add it at a later date.

[michaelsproul]

Damn... yeah, disabling it seems like our only option. We'll also have to prevent the error index from trying to build, just in case it fails in the absence of data.

[michaelsproul]

Sorry I can't put a PR together myself, I'm travelling interstate at the moment.

[michaelsproul]

(Commenting out/removing the dependency that the docs target has on the error index should be sufficient)

[pnkfelix]

(I'm currently revising this PR to provide better feedback about the particular error that was encountered during the metadata-output step. Admittedly such feedback won't actually fix the problem immediately, so if someone would prefer to take the more severe step of removing the error index generation and all of the other code, that's fine; I just figured i would try to acquire more knowledge from the build bots in the meantime...)

[michaelsproul]

Yeah, it would also be nice to know why the output is failing... we should probably pass the io::Error along. I'll be back at my build rig on Sunday.

[pnkfelix]

@bors r=dotdash 723d1df p=1

[pnkfelix]

(I'm hoping that this will either (1.) fail in the same way, but provide much better error feedback, or (2.) magically succeed this time, unwedging various builds, and then future failures will likewise provide much better error feedback...)

[bors]

⌛ Testing commit 723d1df with merge 7e1e838...

[bors]

💔 Test failed - auto-linux-64-opt

[pnkfelix]

(I'm hoping that this will either (1.) fail in the same way, but provide much better error feedback, or (2.) magically succeed this time, unwedging various builds, and then future failures will likewise provide much better error feedback...)

Well, apparently we get the unpredicted variant (3.):

rustc: x86_64-unknown-linux-gnu/stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/librustc
cp: x86_64-unknown-linux-gnu/stage2/lib/librustc_back
/home/ubuntu/src/rust-buildbot/slave/auto-linux-64-opt/build/src/librustc/lib.rs:173:1: 173:52 error: metadata output error os error
/home/ubuntu/src/rust-buildbot/slave/auto-linux-64-opt/build/src/librustc/lib.rs:173 __build_diagnostic_array! { librustc, DIAGNOSTICS }
                                                                                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
error: aborting due to previous error

At this point I'm going to go with @alexcrichton 's original plan (augmented with @michaelsproul 's note) to entirely remove the error code checking and error index page generation.

[pnkfelix]

closing in favor of PR #25706