feat: Add a feature to disable map support

[DeltaEvo]

The map feature, enabled by default, can be disable to remove map support, for codebases not supporting diffs over map to be expressed.

See redbadger/pathogen#1 for some context on when map support might not be present.

Also in the current codebase, removing a value from a map, trigger an unimplemented!

difficient/src/serde_visit.rs

Line 178 in 7d616b8

KvDiff::Removed => {

[StuartHarris]

Hey @DeltaEvo thanks for this. I fixed some formatting but CI is still failing. The tests fail for no-default-features. I'm happy to fix them, but first, I wasn't quite sure I understand why map needs to be a separate feature. I read the pathogen PR, but still don't quite get this (I'm probably being dumb). Would you explain a little further for me please?

[DeltaEvo]

Thanks for helping with the CI, will take a look at the test failure. You're not being dumb, my fault for being too lazy to do a correct explanation 😅 .

So the issue is that the patch format (Update or Splice) used by pathogen cannot express removing a non numbered key from something (we could cheat by making it a Update to null but then how do you represent a Map<K, Option>).

This is also why the map visitor implementation of difficient is incomplete, currently if it it encounters a removed key, it panics, this is another issue, to fix it we need to extend the Visitor trait to add a removed/deleted method.

Focusing back on the main topic, we are left with two solutions:

• Prevent diffing structs containing maps (What this PR does)
• Extending the diff format to support removal (What I explored in the past and why that pathogen PR as been created), but after some discussion with I think it was Viktor, I was convinced that exposing a Map on a ViewModel is not as easy as it sounds, as you cannot express ordering easily (As each language have different Map implementations) :/ and is why I didn't pursue this road.

[StuartHarris]

@DeltaEvo thanks for the additional context. I understand better now. I agree that there should be a first-class way to signify a delete (as an alternative to splice for lists, and as a sound way to remove keys in a map). But I'm not sure how putting maps behind a feature in this crate helps. What are the situations in which this feature would be enabled/disabled?

[DeltaEvo]

It would be disabled for users that don’t want to add this extra operation just to support maps, basically turning using a map for a Diffable struct an Error.

For example, I want to disable it for Photoroom and prevent using maps in the view model so that we can express all updates with just Update and Splice.

[StuartHarris]

Right. Cool. Thanks. We should still add the changes to pathogen I reckon :-) And I'm not quite sure how ordering matters with maps, but that's a discussion we can have in that repo.

[DeltaEvo]

I can’t be there for today’s crux community meeting as it’s a day off in France, but I can bring that topic for next week.

PS: Sorry for the test, should have ran the full command, will fix them asap

[StuartHarris]

That'd be great. There isn't a call today (it's once a month), so hopefully see you on Tuesday