Use GitHub Actions to build jlreq?

[xfq]

jlreq is a very long document, so it takes a long time to open it. In #218 we optimized the images, and the performance has improved. I think the load time can be further optimized by building the document using GitHub actions, like the ARIA folks did here:

https://github.com/w3c/aria-practices/blob/0e7accec081f995fc969f570ec048094791b0bc0/.github/workflows/deploy.yml

[himorin]

rather than building (processing respec and placing processed file) in this repository, I suppose we are encouraged to update as WD on /TR/ more frequently... (also I'm not sure whether these processing eats large volume within full load time.)

[kidayasuo]

How the page load time is optimized by using the GitHub actions? (a novice question)

[xfq]

rather than building (processing respec and placing processed file) in this repository, I suppose we are encouraged to update as WD on /TR/ more frequently... (also I'm not sure whether these processing eats large volume within full load time.)

This is fine, but ED is still slow (unless we synchronize the two commit by commit).

[xfq]

How the page load time is optimized by using the GitHub actions? (a novice question)

I just tried it, and the speed of the two seemed very different. I opened these two pages on the same computer and network (I cleared the HTTP cache before each run):

• https://w3c.github.io/jlreq/
• https://xfq.github.io/jlreq-spec-prod/

The first one took ~9 seconds and the second one took ~3 seconds. I tried running it many times, and the results were about the same. I haven't investigated what caused such a big gap, though.

[kidayasuo]

The first one took ~9 seconds and the second one took ~3 seconds.

Thank you Fuqiao! It seems the difference is significant.

[himorin]

(note, one issue which blocks this, spec generator has shorter limit than takes to process current JLreq - which is also a reason why pr-preview is not working.)

[r12a]

The first one took ~9 seconds and the second one took ~3 seconds.

Fwiw, for me the downloads took ~4 and ~3 seconds, respectively. So not a lot of difference.

[xfq]

(note, one issue which blocks this, spec generator has shorter limit than takes to process current JLreq - which is also a reason why pr-preview is not working.)

This is tracked in w3c/spec-generator#396

We can also try to run ReSpec locally (in the GitHub action environment) instead of relying on spec-generator.

[himorin]

We can also try to run ReSpec locally (in the GitHub action environment) instead of relying on spec-generator.

haven't checked internal processing of w3c/spec-prod, but is this relies on external or local processing?? (yeah, I should secure time for check,,, I know..)

[xfq]

It seems that spec-prod does local processing, and ARIA depends on spec-generator. See:

• https://github.com/w3c/spec-prod/blob/494a9e0108f105413f512c5bafd2faed39e86adc/src/build.ts#L101-L103
• https://github.com/w3c/aria-practices/blob/0e7accec081f995fc969f570ec048094791b0bc0/.github/workflows/deploy.yml#L48