Testing: Fail CI build if local changes exist

[aduth]

This pull request seeks to add a failure case to our existing ci script to trigger a build failure if local changes exist after one or both of npm run install (e.g. unsynced package-lock.json changes) or npm run docs:build. This has been increasingly problematic as of late due to failures to update generated files upon changes, causing confusion for later developers who are then forced to consider how to reconcile the change.

Testing instructions:

Verify that no error is produced upon running npm run ci (or, more specifically, npm run check-local-changes) after checking out the branch.

Verify that an error code is produced when making a local change (but not committing) and re-running the script.

[mcsf]

Thanks for this.

[tofumatt]

Tested locally and this worked (didn't error when nothing changed; error'd when changes exist).

Totally awesome, thanks a mil. 👍

[tofumatt]

It might be nice to add a newline \n in at the end of this message; it's just a bit busy with NPM's output in there. 🤷‍♂️

[aduth]

Yeah, especially in Travis, it's quite difficult to parse the failing output, made worse with concurrently which doesn't display the failing command in the final result (https://github.com/kimmobrunfeldt/concurrently/issues/134).

For this case specifically, it would be nice too if we could limit the output by npm, such that the message is the only thing or at least more obvious.

[tofumatt]

Oops, @mcsf beat me to it but my GitHub tab didn't update 😅

[mcsf]

One can never have too many thumbs.

[tofumatt]

[ellatrix]

Looks good for package lock.

[ellatrix]

@aduth Merged #10304 which included the package lock update.

[aduth]

Build is failing because platform in which Travis runs will introduce changes to package-lock.json after an install.

See:

• package-lock.json and optional packages npm/npm#17722
• https://npm.community/t/npm-ci-treats-optionaldependency-with-unsupported-platform-differently-than-npm-install/1114

[aduth]

To accommodate the issue described in #10234 (comment), I've put together a small script which processes the diff via a series of disgusting regular expressions to effectively ignore the "optional": true additions to package-lock.json, while still otherwise capturing local changes which could occur by npm install.

Would love any feedback on the revised approach, otherwise I may move forward with the merge as we're continuing to see the issue crop up (#10510 (comment), #10347 (comment)).

[tofumatt]

Reviewed the changes too; they make sense and are well-explained. 🚢

[gziolo]

Thanks for providing working solution despite this nasty npm bug 👍