Frontmatter
| id | 1565 |
| title | SharedWorkers context - Sometimes a demo app fails to load |
| state | Closed |
| labels | bug |
| assignees | tobiu |
| createdAt | Mar 22, 2021, 2:40 PM |
| updatedAt | Mar 22, 2021, 11:28 PM |
| githubUrl | https://github.com/neomjs/neo/issues/1565 |
| author | tobiu |
| commentsCount | 2 |
| parentIssue | null |
| subIssues | [] |
| subIssuesCompleted | 0 |
| subIssuesTotal | 0 |
| blockedBy | [] |
| blocking | [] |
| closedAt | Mar 22, 2021, 11:28 PM |
SharedWorkers context - Sometimes a demo app fails to load
tobiu assigned to @tobiu on Mar 22, 2021, 2:40 PM
tobiu
Mar 22, 2021, 11:07 PM
While working on app connect & disconnect events, i accidentally removed a connected event, which is needed for the SharedWorkers context.
In short: singleton instances can get created before a worker is connected, in which case it is not possible to send postMessages to other threads yet to register the remote methods.
combined with the other changes, i think we can now remove the 100ms delay for the workerConstructed event => a faster starting time for your apps.
will deploy a new version now and test it inside the online examples before closing this ticket (in case it works).
tobiu
Mar 22, 2021, 11:28 PM
works fine. loaded the 2 SW apps several times in all 3 modes.
tobiu closed this issue on Mar 22, 2021, 11:28 PM
This one is definitely new, it does affect the shared covid app as well as the new multi window drag&drop demo.
For me, this only happens when reloading the page, not on the initial load.
It also never happens in case the main window console is open (try it in case this one is affecting you).
The issue is related to the remote method access.
Each class can expose methods to different threads. they get found on instantiation and immediately trigger a postMessage to the target thread. For me, it is always the
Neo.vdom.Helper.create()method which is not registered yet.There is already a delay in place to fire the workerConstructed event, but it no longer seems sufficient.
I will dive into this issue soon, since it does affect the online examples.
Not 100% sure yet if there is a cleaner way to let each thread now how many methods will be registered and delay the start until then. Will think about it more!