[WIP, reproduced as failing test] Fix HTMLCS issue affecting <input autocomplete="username"> with implicit type

[danyalaytekin]

Adding a failing test to draft PR.

• Will fix Certain pages result in "Cannot read properties of undefined (reading 'replace')" #691
• Renames pa11y 5.x beta error for empty HTML lang attribute #373's test objects, moves its commentary into code, and inverts and broadens its assertion to simply check for success
  • Unfortunately I can't seem to make the git mv hold here 🤔 so the tracking has been lost

Remote manual test case

nvm install 20
npx \
    pa11y/pa11y#fix-691-including-username \
    http://doctors.mystewardhealthcare.org/user/login

Domain from #691

Notes

• HTML_CodeSniffer error is fixed in squizlabs's master branch but 20 commits await release
  • diff since latest version (html_codesniffer@2.5.1)
    • test case confirmed to pass on version built from master
  • (my initial & incorrect conclusion, but this fork's change may still be good to have) HTMLCS error found & fixed in fork by @Zauberbutter:
    • adds space between input type & translation string squizlabs/HTML_CodeSniffer#303
    • test case confirmed to pass on version built from this fork (because it is forked from master)
• I think this variation of the error also appeared in this comment and below:
  • Error: Cannot read property 'replace' of undefined. #627 (comment)

[danyalaytekin]

As described above this issue appears to be caused by a html_codesniffer@2.5.1 bug that was fixed in their default branch sometime between 2020-2022 but hasn't been released. I can think of two options:

1. Use patch-package to patch html_codesniffer@2.5.1 after installation, but we'd be patching the entire built file, which has most of its functionality on one line, so the patch will be deleting that line (several hundred KB) and replacing it with the patched line, which doesn't seem ideal (example with only a few characters patched).
2. (My preference) Soft fork to pa11y/HTML_CodeSniffer and build from there, cherry picking a few fixes, until the project returns to life.

What do you think @aarongoldenthal @hollsk @josebolos @jpw, and can you think of any other options?

[aarongoldenthal]

As described above this issue appears to be caused by a html_codesniffer@2.5.1 bug that was fixed in their default branch sometime between 2020-2022 but hasn't been released. I can think of two options:

1. Use patch-package to patch html_codesniffer@2.5.1 after installation, but we'd be patching the entire built file, which has most of its functionality on one line, so the patch will be deleting that line (several hundred KB) and replacing it with the patched line, which doesn't seem ideal (example with only a few characters patched).
2. (My preference) Soft fork to pa11y/HTML_CodeSniffer and build from there, cherry picking a few fixes, until the project returns to life.

What do you think @aarongoldenthal @hollsk @josebolos @jpw, and can you think of any other options?

The soft fork seems like the lesser of two evils (although I do wonder if the project will return to life 🤔)

[hollsk]

Yes, I was working up a "state of the project" doc yesterday and noted the lack of activity on HTML_CS then too, and I agree that its future seems uncertain.

I'm also leaning towards soft fork, though I do wonder where that leaves Pa11y over the long term. In theory we could probably replace it with a different rules engine (WebAIM's WAVE, maybe), but that'd be a longer-term decision than here. Soft fork will buy us some time to decide in any case.

[Zauberbutter]

I am gratefully using pa11y together with vNu and a link checker library to provide an internal website quality assurance service. Sadly I don't have the time to develop the service to a open source one, but I have some ideas. So I just want to share some libraries with you, maybe they will be useful for further development of pa11y or the deprecation of htmlcs:

• IBM a11y checker engine
• Kayle a11y checker with rewrites of axe & htmlcs
• Siteimprove's alfa a11y checker