Re: Consensus Gathering: is_initialized

On 26.05.21 12:13, Joe Watkins wrote:

Not absolutely convinced that it's a good idea, I asked Nikita to review, and he's unconvinced also and suggested a discussion should be started. You can read both of our thoughts about it on the PR. What I'm looking for now, is reasons that we are wrong - good use cases that aren't born from code that is doing bad things ... Equally if you think this is really bad even if the reason has already been mentioned, please make noise.

I like it because it is very clear and takes into account the current scope. The alternatives are more messy: - When using "isset" on class properties, it is not obvious if you are checking for initialized state or for null, as it does both - When using reflection, scope does not matter (and you have to explicitely set a property to accessible for private properties) While it is not common that I need to check if a property is initialized, I do think there should be a straightforward way to do it. Setting properties through a constructor is not the only way of initializing parts of a class - some use cases were already brought up, like caching or lazy loading. Also, is_initialized cannot be done in userland, as passing an unitialized property to a function already leads to an error, so every way of checking now is a workaround that cannot be made clearer for the reader of the code. I would prefer it if it also worked for untyped properties, just to be consistent. Static analyzers would also have a much easier time understanding the code compared to now. When using isset on a non-nullable property a static analyzer would rightfully complain that the property is not nullable, as it assumes you are checking for null, not for uninitialized - isset in this case is abused to check for uninitialized, hiding the actual intent. Not sure how good static analyzers are at understanding the reflection code, but is_initialized would make things clearer for both humans and analyzers.