Mark all properties as nullable? PHPStan "conflict"

[bbprojectnet]

Summary

Hello,

The IDE Helper marks a property as nullable only when the corresponding database field is nullable. But in reality, when we create a new, empty model, all properties are effectively nullable. The PHPDoc generated by the IDE Helper only reflects the state of a model after it has been retrieved from the database.

From a pure PHP perspective, this isn’t correct, because any property can be null when the class is first instantiated.

I have an “edge case” where many of my attributes are read-only and depend on other attributes. For example:

	protected function foo(): Attribute
	{
		return Attribute::make(
			get: fn (): ?string => $this->bar !== null ? $this->bar->description() : null,
		);
	}

I need to check whether $this->bar !== null before calling a deeper method to avoid errors when accessing this property on a freshly initialized model.

However, I also use PHPStan, which raises an error because, based on the model’s PHPDoc, the check is unnecessary — it assumes bar is not nullable.

So… my question is: is there any way to tell the IDE Helper to mark all properties as nullable?

[mfn]

I'm not aware of such an option, but I can provide some thoughts on that matter: YMMV

But in reality, when we create a new, empty model, all properties are effectively nullable.

In practice, whenever I write a migration which expresses defaults on a column, I also define that default value in <Model>::$attributes.
So, when you new Model(), those attributes are already set. In my experience this works for majority of cases and does not clash with anything.

Also: are you using larastan? I don't know about this particular case, but AFAIK it has additional logic for database things too.

Btw, what phpstan level are you working on and which errors exactly are you getting?

---

For this reason, in all the years and all the big projects I worked on, I never had an issue with this; correctly defining the default attributes worked really well. Given that, I'm also not convinced such an option would be a good approach.

HTH

[bbprojectnet]

Yes, in that case (default attributes values) it's works fine :) but i can't go that way.
But, even that, you still can assign to attribute null value, it's not protected by php type, so this type is still nullable.

Yes, i'm using larastan, but larastan error it's correct, based on generated phpdoc for model, my comparation in attribute logic is not needed. For now, i have so many @phpstan-ignore to pass static analyses, but the main problem is in model phpdoc.

I'm on 8 level. This is my real life code, and phpstan error:

	/**
	 * Date & time frame attribute.
	 *
	 * @return \Illuminate\Database\Eloquent\Casts\Attribute<\App\Support\DateTimeFrame|null, never>
	 */
	protected function datetimeFrame(): Attribute
	{
		return Attribute::make(
			get: fn (): ?DateTimeFrame => $this->datetime && $this->interval ? $this->interval->dateTimeFrame($this->datetime) : null,
		);
	}

  :110   Anonymous function never returns null so it can be removed from the return type.
         🪪  return.unusedType

datetime and interval columns are not nullable in database, but i need that check to avoid (new Model)->datetime_frame error or when i serialize this model. You can ask, why i do that thing? ;) It's hard to explain without rest of application. Beside of that, i'm trying to write strongly typed code, so, when it's possible to raise this kind of error (access this attribute on just initiated class) i want to handle that in correct way.

What i'm trying to achieve, it's to avoid so many ignore annotiation caused by incorrect model phpdoc.
I'm understand that ide helper base on database columns, and php doc of model corresponding with retrieved model - but in my case it's generate many problems. Maybe some switch/option to change this behaviour?

[bbprojectnet]

Another example, let's assume we have model with not nullable column hash (some unique, random value beside of id).
And, in model creating observer event that logic:

if ($model->hash === null) {
   $model->hash = Str::random();
}

When we don't assign any hash, it's generated just before save to database, or not when we earlier assign something.

In this scenario, the null value it's tangible same as string. But ide helper mark this property as just string which also cause so many warnings in phpstan. Just string does not reflect well the real type of this property, only when we retrieve this model next time from database.