Compile out editor-only logic within _validate_property in export template

[beicause]

Addresses #103357.
Wraps editor-only logic in validate_property with Engine::get_singleton()->is_editor_hint() or

if (!Engine::get_singleton()->is_editor_hint()) {
    return;
}

to make it only run in the editor (mainly PROPERTY_USAGE_NO_EDITOR PROPERTY_USAGE_READ_ONLY and property hint ,except PROPERTY_USAGE_NONE, PROPERTY_USAGE_STORAGE).
This speeds up Object.get_property_list Resource.duplicate and Node.duplicate for certain classes that do many string comparisons in validate property, and should also speed up scene loading.

Benchmark:

func _ready() -> void:
	var m := StandardMaterial3D.new()
	var e := Environment.new()

	var t := Time.get_ticks_msec()
	for i in range(1000):
		m.duplicate()
	var t1 := Time.get_ticks_msec()

	print(t1 - t)

	t = Time.get_ticks_msec()
	for i in range(1000):
		e.duplicate()
	t1 = Time.get_ticks_msec()

	print(t1 - t)

	t = Time.get_ticks_msec()

	for i in range(1000):
		m.get_property_list()
	t1 = Time.get_ticks_msec()

	print(t1 - t)

	t = Time.get_ticks_msec()
	for i in range(1000):
		e.get_property_list()
	t1 = Time.get_ticks_msec()

	print(t1 - t)

Edit:
In editor: 200 105 452 360
In game:
before: 124 107 241 217
after: 57 34 186 132

[dsnopek]

Thanks, this makes sense to me in principle.

I'm assuming the benchmarks in the description are with this PR? Can you also give benchmarks without this PR to compare against?

[Calinou]

• Related to Rework the Configuration Warning system #90049
  (where I proposed surrounding all _get_configuration_warnings() with #ifdef TOOLS_ENABLED).

This could also help with binary size significantly.

[beicause]

Recompiled and tested. Indeed, it is still faster in the game than in the editor without this PR, but this PR still has significant improvement.

without this PR:
In editor: 207 112 498 386
In game: 122 106 247 201

with this PR:
In editor: 211 112 502 382
In game: 66 37 204 155

[arkology]

Interesting, is the TOOLS_ENABLED check actually necessary? I thought compilers were smart enough to optimize out if (false) {} (as is_editor_hint() returns false for non-editor builds - but maybe compilers don't expand Engine::get_singleton()->is_editor_hint() to false during compilation?).

[beicause]

Interesting, is the TOOLS_ENABLED check actually necessary? I thought compilers were smart enough to optimize out if (false) {} (as is_editor_hint() returns false for non-editor builds - but maybe compilers don't expand Engine::get_singleton()->is_editor_hint() to false during compilation?).

After this PR, the size of linux release template is reduced by 31KB. Removing the TOOLS_ENABLED doesn't affect the size, so I believe they are still compiled out without TOOLS_ENABLED.

Not sure if we should remove the TOOLS_ENABLED. I found most of the other Engine::get_singleton()->is_editor_hint() in the codebase are not surrounded by TOOLS_ENABLED, but there are also some exceptions.

[KoBeWi]

I checked all changes and they are correct.

Though I still think that methods completely wrapped in is_editor_hint() should just use early return. Partially disabled methods are fine, early return would be error-prone in their case.

[beicause]

I have changed to use return for completely wrapped methods, but I will check it again later.

[beicause]

I'm surprised that we use the  ^=  operator in  _validate_property, which means the results will be different when it runs repeatedly. Is this a bug?

[KoBeWi]

_validate_property() runs on a copy of PropertyInfo, so each validation is independent.

[TokageItLab]

Now this PR should be rebased.

[beicause]

Rebased and checked all _validate_property again.

[akien-mga]

Thanks!