Fix Timeoutable timeouts flash hash keys

[janraasch]

See #1777 for an in-depth discussion of the unexpected behaviour.

See #4038 for the fix, but awaiting tests.

The behaviour still works as expected for redirects, see

test 'error message with i18n'
test 'error message with i18n with double redirect'

in https://github.com/plataformatec/devise/blob/master/test/integration/timeoutable_test.rb#L141-L166.

[tegon]

@janraasch Thanks for tackling this.

It seems really odd to me that the tests still pass after the code was deleted. It makes me wonder whether the tests were actually covering the intended scenarios in the first place.
I'll try to simulate the issue this code was trying to solve in the past, but I'm also worried about what @josevalim said about the multi-thread/process environments.

Did you have a chance to manually test this?

[janraasch]

Did you have a chance to manually test this?

Long time no see 😉. If I remember correctly we monkey-patched this on our app, but to be honest I no longer am affiliated with that application so I cannot look into the source code. So \_o_/ I guess 🤓.

It seems really odd to me that the tests still pass after the code was deleted. It makes me wonder whether the tests were actually covering the intended scenarios in the first place.

That does seem strange, I absolutely agree.

[tegon]

Ok, get it. Thanks for replying.

I'm going to close this then, as I don't have the context of why this change was needed or what side effects this pull request might cause. I hope you understand my reasoning.

Thanks again!