Error : Unexpected Duplicate. with Warp backend.

[sheganinans]

I keep getting this error when using the jsaddle-warp backend.
However the same exact reflex widget when using jsaddle-webkit2gtk will work perfectly fine.

To replicate this error:

git clone --recurse-submodule https://github.com/sheganinans/reflex-warp
cd reflex-warp
reflex-platform/try-reflex
nix-shell -A shells.ghc
cabal new-repl warp-reflex
main

Connect a web browser to http://localhost:3911, and click the button a few times and check the output in the cabal repl. Soon enough the error will show up and the webapp will freeze.

However if you instead run cabal new-repl webkit-reflex and try pushing the button, the app will work exactly as expected.

[sheganinans]

Any luck on the replication of this bug? I tried it on a different installation of NixOS and got the same behavior.

[nomeata]

Does the browser freeze with high CPU usage? That’s what I am observing here at least…

[nomeata]

According to https://www.reddit.com/r/haskell/comments/88k18x/monthly_hask_anything_april_2018/dwm8qzr/ it is a Firefox-specific problem, and using Chrome avoids it.

[schell]

I see this too, it’s a bummer because FF is my main browser. Any headway?

[schell]

@3noch - you seem to have some insight into this bug. What exactly is the problem on FF?

[vincent-prz]

Upping this issue, as I really care about not having to use Chrome to use JSaddle.

[niteria]

I took a look at this and minimized the example to https://github.com/niteria/jsaddle-repro

As expected it works in Chrome Version 81.0.4044.122 (Official Build) (64-bit) and fails on Firefox 77.0.1 (64-bit).

This is using the latest jsaddle and jsaddle-warp, version 0.9.7.0 for both.

[niteria]

Ok, this is really the same as #51.
It fails in Firefox because Firefox completely disallows synchronous XHR requests. Chrome is less strict.

There's sync-xhr that's supposed to allow this https://featurepolicy.info/policies/sync-xhr, but it's not implemented on Firefox.

[danwdart]

So would the solution be only using async xhr?

[danwdart]

Seems to be the same as #54 too maybe

[danwdart]

Has progress been made towards this? I've been compiling to JS every time since seeing this issue - seeing as the default browser on deployments gets what looks like memory leaks.