Stalling out on CI with Xcode 10

[u2606]

After updating our app to Xcode 10 and Swift 4.2, danger-swiftlint stalls out on CircleCI. After linting some of our files, it eventually stops printing to STDOUT, and after 10 minutes of inactivity, CircleCI terminates it.

This behavior did not occur with Xcode 9 and Swift 4.1.

This is the relevant portion of our Dangerfile.swift:

import Danger
import DangerSwiftLint // package: https://github.com/ashfurrow/danger-swiftlint.git

SwiftLint.lint(inline: true, directory: "Clutter/Application”)

This is the (abridged) log from CircleCI:

#!/bin/bash --login -o pipefail
danger process danger-swift
Ran with: Dangerfile.swift /var/folders/gk/lkr9pm5x039fx6d3j9r52rv80000gn/T/danger-dsl.json /var/folders/gk/lkr9pm5x039fx6d3j9r52rv80000gn/T/danger-response.json
Executing pwd 
Working directory: /Users/distiller/project

Executing swiftlint lint --quiet --path "Clutter/Application/API/ClutterAPI.swift" --reporter json
<189 lines omitted>
Executing swiftlint lint --quiet --path "Clutter/Application/Protocols/TableViewController.swift" --reporter json
Executing swiftlint lint --quiet --path "Clutter/Application/StyleKits/ClutterStyleKit.swift" 
Too long with no output (exceeded 10m0s)

I notice that the final swiftlint lint line doesn’t end with -- reporter json. Additionally, ClutterStyleKit.swift is a 3005-line, PaintCode-generated file. Not sure if either of those is relevant.

[u2606]

On a subsequent run, the CircleCI log ended one line earlier:

Executing swiftlint lint --quiet --path "Clutter/Application/Protocols/TableViewController.swift" --reporter jso

Note that this time, the final line ended mid-word.

[ashfurrow]

@erikstrottmann thanks for opening this issue – no solution comes to mind, so let's dig into this. Is there a way in Circle to get a raw log file? I've seen CI providers cut off text like this before but it does get written. If you could try running the commands locally to verify they work, that'd be great too 👍

[u2606]

@ashfurrow, there’s no native support for raw log files on CircleCI. I tried using https://github.com/artemv/circleci-step-outputter to download the output of the Danger step, which consisted of the same log messages I pasted above, split into two .txt files.

I had the same result when running against my Swift 4.2 branch locally with DANGER_GITHUB_API_TOKEN="<...>" DANGER_FAKE_CI="YEP" DANGER_TEST_REPO="<...>" DANGER_TEST_PR="<...>" /usr/local/bin/danger process /usr/local/bin/danger-swift — it logs that it’s linting some files, then stalls indefinitely midway through a print statement.

However, I can successfully run Danger SwiftLint against a different PR that’s still using Swift 4.1, even when my local machine is using the Xcode 10 Command Line Tools to run Danger Swift.

That makes it look like the failure isn’t being caused by Xcode 10 itself, but by something I’ve changed in my Swift 4.2 branch. Is there any way to run Danger against a specific commit? That would help me narrow down which change breaks Danger SwiftLint.

[ashfurrow]

@erikstrottmann you should be able to run bundle exec danger pr with --head= and --base= options to specify which commit to use. Thanks for looking into this!

[u2606]

@ashfurrow, does Danger JS’s danger process subcommand support --head= and --base=? Since I’m using Danger Swift, I’m unable to use Danger Ruby or Danger JS’s danger pr subcommand.

When I pass --head="<...>" and --base="<...>" to Danger JS’s danger process, it appears to ignore them. When I include the DANGER_TEST_REPO environment variable, it gets stuck midway through printing linting logs, even when I specify a head that has Danger SwiftLint disabled. When I omit DANGER_TEST_REPO, it appears to default to a different PR (the most recently created one?) and succeeds.

[ashfurrow]

Oh yikes, I'm not sure. Hmm, if I were diving into this, I'd focus on reproducing the issue locally. I'd probably use the Circle CI CLI tool to run the CI image locally and get it to stall so I could figure out what was going on.

I'd prefer to dive into this with you but I'm short on time – let me know what questions I can answer, and good luck.

[orta]

Heyo folks, I'm monitoring this issue here: danger/swift#108

[orta]

If someone is still seeing this, can you try add ~/.danger-swift to your build caching? I've got a pretty drastic solution: danger/swift#139 but it might take some time to figure out

[u2606]

We’re no longer seeing this issue. We’ve updated to Danger Swift 1.0.0, changed the command to danger-swift ci, and removed the separate Danger SwiftLint import.

[ashfurrow]

Thanks for following up @erikstrottmann! This issue is safe to close then?

[u2606]

Yep, should be safe to close!