Disable Jackson's FAIL_ON_UNKNOWN_PROPERTIES by default

[gsmet]

This is what was discussed on the mailing list.

Commits should be reviewed separately as the first one is a cleanup commit, renaming a file.

/cc @maxandersen @loicmathieu @geoand

[loicmathieu]

We should document that the default Jackson ObjectMapper is customized by default.
Is it the only customization that we wants by default (I usually always disabled date as timestamp as we better off with ISO dates for interoperability).

I think we must provide more configuration options for the more customized features of Jackson. Or maybe we can have "generic configuration" where a developer can use :
quarkus.jakson.feature.<feature-name>=value?

[gsmet]

So to be honest, I'm not for having all the features in the config.

People who want to customize Jackson can do it with a customizer, that's practical enough.

So for me it's really about:

• providing more sane defaults if we think they make sense
• providing an easy way to go back to Jackson's default behavior

I will write the docs once we agree on the idea.

[loicmathieu]

@gsmet OK for me then.

providing more sane defaults if we think they make sense

I think serializing dates as ISO (as JSON-B) is a better default than timestamp.
So I'll disabled the features WRITE_DATES_AS_TIMESTAMP by default and add a config item for it.

Keep in mind that it will be a breaking change for the frameworks that currently uses the default ObjectMapper (AWS Lambda, Funqy, GCP Functions, Azure Functions at least).

[geoand]

+1 for this.

But we do need to communicate that this is a breaking change

[gsmet]

I'm not exactly sure about the timestamp change format. @geoand WDYT?

[geoand]

I like the idea, +1

[gsmet]

@geoand so I tried to make the config BUILD_TIME_AND_RUNTIME_FIXED and thought the bean would then be initialized at a proper time.

It somehow doesn't work: no injection errors but the injected bean does not have the configured value.

I ended up using the usual Support pattern but I think it should be doable to inject BUILD_TIME_AND_RUNTIME_FIXED and that this bean should be initialized earlier to avoid the issue we have with a purely runtime config.

As for the timestamp, I'll let others do the work. I have no idea what it does and doesn't really have time to experiment.

Now that the infrastructure is here, it should be easy enough.

If nobody takes it, I'll take care of it sometime next week.

Now let's see if CI prefers this version.

[maxandersen]

Fail on properties is good. But timestamp seems like it risk breaking many or ?

[geoand]

I think it's a reasonable default but indeed it's another breaking change.

We can certainly leave it out for now

[geoand]

Good idea to have this first, as it allows users to override the behavior with their own customizers if need be

[geoand]

You might want to add a note that this only works for the ObjectMapper configured by Quarkus and which they obtain via CDI (or use implicitly via RESTEasy etc).

I know it should be evident, but I am sure it might come as a surprise to some users.

[geoand]

👍

[gsmet]

You're right, it's not obvious, I added some precisions.