nexedi / erp5

added 1 commit

• ed357ba1 - [erp5_web_renderjs_ui] Fix rjs_gadget_erp5_serviceworker.js. Use cache.add becau…

Compare with previous version

assigned to @romain

added 1 commit

• dac9a653 - [erp5_web_renderjs_ui] Fix rjs_gadget_erp5_serviceworker.js to fix Firefox Cache Storage problem.

Compare with previous version

added 1 commit

• 61c625c7 - [erp5_web_renderjs_ui] Stop calling skipWaiting() and clients.claim() for the new service worker.

Compare with previous version

added 1 commit

• f36c1e65 - [erp5_web_renderjs_ui] New client can use the latest cache without waiting for t…

Compare with previous version

added 1 commit

• 7771d2fd - [erp5_web_renderjs_ui] Fix translation script. Get service worker filename from …

Compare with previous version

added 1 commit

• 5f84fccd - [erp5_web_renderjs_ui] Fix service worker. Don't hardcode the special cache name.

Compare with previous version

added 1 commit

• b90d25e3 - [erp5_web_renderjs_ui] Add more comments because service worker is unstable and hard to use safely.

Compare with previous version

added 1 commit

• afbd3923 - [erp5_web_renderjs_ui] If service worker failed to install cache, unregister thi…

Compare with previous version

added 1 commit

• 84ea41ac - [erp5_web_renderjs_ui] Clean up service worker code. Remove many if.

Compare with previous version

added 1 commit

• bcd04ff4 - [erp5_web_renderjs_ui] Update service worker code. Use lower case for non-global variables.

Compare with previous version

added 1 commit

• 1b40f812 - [erp5_web_renderjs_ui] Update service worker code. Client_id is null when it is …

Compare with previous version

added 1 commit

• 5947bc2b - [erp5_web_renderjs_ui] Use absolute wallpaper image URL to support firefox.

Compare with previous version

mentioned in commit 08637332

added 1 commit

• 08637332 - Revert "[erp5_web_renderjs_ui] Use absolute wallpaper image URL to support firefox."

Compare with previous version

added 1 commit

• 8dcc7d23 - [erp5_web_renderjs_ui] Use absolute wallpaper image URL to support firefox.

Compare with previous version

added 1 commit

• 7ee9acc0 - [erp5_web_renderjs_ui] Firefox wallpaper URL changer must be implemented as a declareService.

Compare with previous version

Merged as 264ded5c

closed

reopened

Gadget inerface

I rebase my development branch with your changes and got a new failing test: testERP5Interface.

This test checks that gadget implementation matches the declared interface.

To achieve this, it loads all gadgets defined in the "now removed" appcache.

I guess the interface checker, the validator test and the erp5 interface test have to be changed to read the service worker, as you did for the testTranslation.

Wallpaper

On my development instance, I also saw that the wallpaper is not handled by the service worker (even if I add it to the cache). This may comes from a Chrome bug.

Thanks. I will check them next week.

This MR has conflicts, but the conflicts seems to be because files deleted in the branch have been modified in master. I tried to rebase to fix conflicts, the result is 0c813702

---

I'm not working on this MR, I'm just working on the alarm to update application cache manifest. I planned to base it on Base_getListFileFromAppcache but I see it got renamed to Base_getTranslationSourceFileList in 36f5e732 .

The logic here is also very different, so I think I will start writing this alarm using Base_getListFileFromAppcache and we'll adjust later.

---

Also, I was wondering, instead of parsing JS to extract REQUIRED_FILES, wouldn't it be easier to use a JSON file as definition of the "URLs to cache for this website" ?

The service worker code could be written as:

/*global self, caches, fetch, Promise, URL, location, REQUIRED_FILES*/
(function (self, caches, fetch, Promise, URL, location, REQUIRED_FILES) {
 // ... in this scope, REQUIRED_FILES is defined
}(self, caches, fetch, Promise, URL, location, REQUIRED_FILES));

We define as a web site property the reference of the json.

At the end of WebPage_viewAsWeb (where we already have some logic), we add something like:

# When we are serving the service worker javascript, dynamically prepend the json definition of cached pages.
cache_url_list_json_url = web_section.getLayoutProperty("configuration_cache_url_list_json_url")
if cache_url_list_json_url:
  service_worker_url = web_section.getLayoutProperty("configuration_service_worker_url")  
  if service_worker_url and context.getReference() == service_worker_url:
    return "var REQUIRED_FILES = %s\n%s" % (
      web_section.getDocumentValue(reference=cache_url_list_json_url).getTextContent(),
      web_content)

In fact, I don't know, I don't understand yet the plan of how this would work, so this idea might be non relevant.

What's the intended scenario for cases where we make a renderjs website similar to web_site_module/renderjs_runner but with a slightly different list of URLs to cache ? Do we write a custom service worker js with different REQUIRED_FILES ?

[ edit to fix indentation in the python code and one more time the code itself ( REQUIRED_FILES = was missing )]

Yes. You only need to clone rjs_gadget_erp5_serviceworker.js give it the higher version, then update REQUIRED_FILES. Aand then if you need to update existing cache, you need to update CACHE_NAME, just increment '_0001' to '_0002'.

Thanks. I don't want to make things more complicated, but cloning rjs_gadget_erp5_serviceworker.js and modifing is not so convenient because this file contain both the definition of what's in the cache and the logic to manage the cache. That would be better if we could customize only the "what's in the cache" part.

One way could also be to split the logic and the cache definition in two scripts and use importScripts, maybe:

• one service worker "entry point" script with the current logic
• another script that the entry point uses importScripts to load. This scripts fetches a json of cache definition (required files and maybe cache name )

This way to customize the list of cached resources, we just need to customize the json. This json is also straightforward to use for tools like "update translation", gadget interfaces tests or what will be equivalent of !1009 (merged)

It seems that for the equivalent of !1009 (merged) we can set in the json a hash of each file content and only update cache for files that have changed. https://www.oreilly.com/library/view/building-progressive-web/9781491961643/ch04.html#idm140659830074624 is an example of an "install" event that copies in the new cache version some responses from the previous cache version, we could do something similar where we copy from previous cache if hash is same and we could achieve something that when we have 200 responses and only 1 of them is modified, we refresh only 1. This could be useful to speed up development. ( Maybe I'm missing something but when I develop some renderjs I have to refresh all pages, which takes about 30 seconds from a server in France )

But all this would be hard to do correctly. I'm certainly not saying we should do something complex like this now, it's just ideas. Maybe first finish a first version of service worker where the update scenario is increment CACHE_NAME in rjs_gadget_erp5_serviceworker.js.

Something was missing except the tests ? ( result is not bad but it was not running all the tests )

For now, I prefer not to split the service worker file, and I prefer to keep the list of files inside it. Ensuring everything is correctly working is already complex enough.

Developpers will have to copy the service worker code for now.

Concerning the hash generation, the "best" solution is probably to generate an immutable URL based on the content result, like "couscous.js?etag=hash". This way, we can set an infinite and immutable HTTP cache value. CDN will be able to keep the files in cache. And browser will not trigger any query in such case.

And we will not have to implement any clever and buggy code on the service worker code. Except that we must handle all hardcoded gadget dependencies which do not use such "etag" value. But as you said, this could be done in a second step.

when I develop some renderjs I have to refresh all pages, which takes about 30 seconds from a server in France

Comment the files you're working on, and only those 2 files will be reloaded.

For now, I prefer not to split the service worker file, and I prefer to keep the list of files inside it. Ensuring everything is correctly working is already complex enough.

Developpers will have to copy the service worker code for now.

Sure, I totally agree, let's make a first version like this.

This MR was replaced by !1069 (merged)

closed