Enable or disable removing false data for JsonToFormDecoder

[meyerbaptiste]

The JsonToFormDecoder class remove false data like a classic form.
But, in the case where we have a form with just one checkbox or patch just one checkbox it does not work because form is not submitted due to missing of parameters in the request and so it is invalid.
This issue can be fixed by replacing the value by null instead of remove it.

So in this PR I propose that we add a container parameter for this service which can enable or disable removing false data.

I use a container parameter because it is a tricky case which does not require to implement a configuration system of the decoder but which leaves the possibility to the developer to modify it if necessary.

[lsmith77]

if we do add thigs feature, then this should be handled via the DI extension

[lsmith77]

I always have concerns about adding options that have a global effect since it means that people combining different Bundles that require different config will run into trouble. at the same time adding more features that are configured by path etc add overhead.

sigh :-/

[dunglas]

This limitation only occurs when using the PATCH HTTP method with the JsonToFormDecoder:

• If the HTTP method is set to PATCH, the Symfony Form Component only updates entity properties submitted by the user.
• The component also assumes checkboxes behave like in application/x-www-form-urlencoded. To set its value to false, the data for this checkbox must be unset (it's what currently do JsonToFormDecoder). Or, as @meyerbaptiste noticed, set to null.
• Actually, if JsonToFormDecoder is used together with the PATCH method, a checkbox cannot be set to false because the PATCH mode take precedence and the checkbox is ignored (because unseted).

As the role of JsonToFormDecoder is only converting JSON data to a format accepted by the Form Component, this PR can be simplified a lot: just change the current behavior to always set $value to null and remove the unset call. It will be much simpler and will work in all use cases.

[sroze]

👍

[lsmith77]

I never used the decoder myself but I guess this could be argued to be a bug fix rather than a BC break ..

[sroze]

@lsmith77 I haven't the same idea: if some one was using this decoder before this change, and was testing values for instance with array_key_exists to make a process based on the checkbox value, it will break his process as the key now exists, but its value is null.

That's why I think that the idea of @meyerbaptiste was acceptable: it don't introduce BC break but allow to fix the issue, and also allow to change this default behaviour (that is now unset instead of null value) in the next major version.

[adapik]

OK, now how I cannot distinguish empty field from false field? No chance