Allow disabling all Callback decorators via environment properties

[Sanne]

The CallbackDecorator provided by Contex Propagation has a strong impact on performance, and in some contexts unnecessary.

It's rather tricky to remove it from classpath, so it would be very handy to have a different option to disable it.

cc/ @jponge @FroMage

[FroMage]

I feel that it could be useful to have a more general approach to disabling ServiceLoader stuff, perhaps with a setting that overrides the list of detected services? This way we could use this approach in more loaded service lists, and perhaps exclude specific services on top of being able to disable them all?

Something like services.fqdn.CallbackDecorator= to disable it, and services.fqdn.CallbackDecorator=foo.BarDecorator,gee.OtherDecorator?

I've needed something like this for Quarkus sometimes…

[Sanne]

I feel that it could be useful to have a more general approach [...]

I agree, butl all this state is currently a global static. That would need some more extensive refactoring which might not be suitable do do now, while I would need a solution in short term.

[codecov]

Codecov Report

Merging #670 (78315b6) into main (e06ce7d) will decrease coverage by 0.06%.
The diff coverage is 85.71%.

@@             Coverage Diff              @@
##               main     #670      +/-   ##
============================================
- Coverage     90.25%   90.19%   -0.07%     
- Complexity     2989     2992       +3     
============================================
  Files           374      374              
  Lines         11762    11764       +2     
  Branches       1476     1478       +2     
============================================
- Hits          10616    10610       -6     
- Misses          584      591       +7     
- Partials        562      563       +1     

Impacted Files	Coverage Δ
...smallrye/mutiny/infrastructure/Infrastructure.java	73.03% <85.71%> (-0.27%)	⬇️
...utiny/operators/multi/MultiWindowOnDurationOp.java	65.71% <0.00%> (-5.72%)	⬇️
...java/io/smallrye/mutiny/helpers/Subscriptions.java	79.00% <0.00%> (-2.77%)	⬇️
...perators/multi/processors/SerializedProcessor.java	90.09% <0.00%> (-1.99%)	⬇️
...mallrye/mutiny/operators/multi/MultiFlatMapOp.java	85.98% <0.00%> (-0.63%)	⬇️
...mallrye/mutiny/helpers/queues/MpscLinkedQueue.java	98.38% <0.00%> (ø)
...erators/multi/builders/SerializedMultiEmitter.java	83.54% <0.00%> (+1.26%)	⬆️
...tiny/operators/multi/MultiBufferWithTimeoutOp.java	74.78% <0.00%> (+1.68%)	⬆️
...y/operators/multi/processors/UnicastProcessor.java	97.34% <0.00%> (+1.76%)	⬆️
.../io/smallrye/mutiny/helpers/queues/DrainUtils.java	82.22% <0.00%> (+6.66%)	⬆️
... and 1 more

[jponge]

Ok we can have this.

How about having DISABLE_CALLBACK_DECORATORS_PROP_NAME as a public constant?

[Sanne]

Ok we can have this.

How about having DISABLE_CALLBACK_DECORATORS_PROP_NAME as a public constant?

If you prefer I can do that. But maybe it then becomes harder to remove later, assuming some better alternative arises from @FroMage 's ideas?

Personally I have no opinion, let me know if you want me to change it :)

[jponge]

@FroMage any better idea?

[FroMage]

Errrr, no better idea. I guess this depends on which other system properties Mutiny uses, and whether you want to advertise this one. I was more thinking about MP Config myself, but it's more complex. I'll start a discussion on quarkus-dev about this.

[jponge]

On second thought I think we can keep the property private, and not advertise this too much. Thoughts @cescoffier?

[cescoffier]

We don't have to advertise it and keep it private.

[FroMage]

There you go, problem solved ;)

[jponge]

Here we go!

[Sanne]

thanks all! Looking forward to use this :)