Re: type hinting

php.internals

Tomi Kaistila

18 years ago
> I believe the cleanest solution that we could implement would be using > the type casting with "Type" objects.
I experimented with this for a couple of months, a couple of weeks ago. In opinion, it does not work well. I am guessing developers at java also figured the same since they still also have primitive types. Aside from the mentioned performance cost, it results in ugly and clumsy code. Not to mention you need to cast the objects into primitive types if you wish to use the variables with PHP functions. function random(Integer $max, Integer $min) { $n = rand($min->getInteger(), $max->getInteger()); return $n; } Implementing the __toString() method in the classes helps a little but not much. And really, if you think about it, just adding support to scalar type hinting solves the hold issue, thus making this sort of solution pointless and a pit hole wasted energy and time.
> Is there any reason we can't at least add "scalar" and > "resource" (and possibly "object" for generic objects) to typehints?
This sounds like, at least, a partial victory to adding proper type hinting to PHP but in my mind it is not enough. I have counted that about a quarter of all type hints that I would make is about matching the argument's type to scalar. The rest are about separating specific scalars from each other. In most cases integer and boolean. I would probably end up not using type hinting at all, just to save the users of my API from the confusion and just do the type checks myself inside my methods.
> Thinking about this, simply because we all know PHP __CAN__ type > juggle, there is no need for any errors/warnings/notices on scalar > types. It will be up to the API developer to make sure appropriate > values are supplied (if necessary), so this reduces the type checking.
In other words, it is not "in the spirit of PHP" to enable strict type hinting. I can understand this argument, but I still disagree heavily with this approach. 1) It is not consistent with the way that Array and Object type hinting works. While you might make the argument that it is not an issue because you cannot implicitly cast those types, it does create an annoying (yet another) magic behavior thing in PHP that people need to learn and remember. If anything, it will confuse new users, which I believe a lot of people were concerned about. Essentially, if it is necessary (and most of the time it is) to validate the value beyond just the type, you might as well do the casting at the same time as you validate. 2) The reason people are asking for type hinting for scalars is specifically to move some of the responsibility from the API developer to the user of the API and it is not an unreasonable request. After all, if person A has a program that asks for user input, the best place for validation is not in the class that person B wrote. The values should be validated in the program that asked for the input. Since the validation is necessary, person A can easily cast the type of the input at the end of the process, before he passes it to the class. E.g. function validate($input) { // validate the input // return false if it was not valid // otherwise return a properly cast value } $input = $_GET['var1']; $input = validate($input) if(! empty($input)) { doSomething($input); } This seems like a perfectly logical division of labor. The validator validates and casts and the library class acts on the proper input. Tomi Kaistila PHP Developer

Sam Barrow

18 years ago
On Tue, 2008-01-08 at 16:03 +0200, Tomi Kaistila wrote:
> > I believe the cleanest solution that we could implement would be using > > the type casting with "Type" objects. > I experimented with this for a couple of months, a couple of weeks ago. In > opinion, it does not work well. I am guessing developers at java also figured > the same since they still also have primitive types. Aside from the mentioned > performance cost, it results in ugly and clumsy code. Not to mention you need > to cast the objects into primitive types if you wish to use the variables > with PHP functions. > > function random(Integer $max, Integer $min) > { > $n = rand($min->getInteger(), $max->getInteger()); > return $n; > } > > Implementing the __toString() method in the classes helps a little but not > much.
But the syntax is longer ("$a = 5" vs "$a = new Integer(5)"), and if you have a large application with hundreds of integers it starts to add up. Performance is also much worse when using objects for every variable.
> And really, if you think about it, just adding support to scalar type hinting > solves the hold issue, thus making this sort of solution pointless and a pit > hole wasted energy and time. > > > Is there any reason we can't at least add "scalar" and > > "resource" (and possibly "object" for generic objects) to typehints? > This sounds like, at least, a partial victory to adding proper type hinting to > PHP but in my mind it is not enough. I have counted that about a quarter of > all type hints that I would make is about matching the argument's type to > scalar. The rest are about separating specific scalars from each other. In > most cases integer and boolean.
Better than nothing.

Tomi Kaistila

18 years ago
> But the syntax is longer ("$a = 5" vs "$a = new Integer(5)"), and if you > have a large application with hundreds of integers it starts to add up. > Performance is also much worse when using objects for every variable.
I agree. That is what I meant by "ugly" and "clumsy" in my last post.
> > This sounds like, at least, a partial victory to adding proper type > > hinting to PHP but in my mind it is not enough. I have counted that about > > a quarter of all type hints that I would make is about matching the > > argument's type to scalar. The rest are about separating specific scalars > > from each other. In most cases integer and boolean. > > Better than nothing.
No, it is not. It is actually better to not fix it at all, if you cannot fix it properly. Especially in this case when the feature has already once been left "unfinished". The benefit is zero for me, if when I need to hint a boolean and can only hint a scalar. In that case, I will still have to do the validating myself. I might as well leave the type hinting out all together, instead of waste resources for having first PHP check that it is a scalar and then my own code to check that it is a boolean. Actually, if I can only hint "scalar" in general, I am essentially telling the users of my API that they should be ready for two types of errors. That which occurs when the argument is not a scalar and that which occurs when it is not a boolean. One will be a PHP error and the second will be an exception. Would be easier on them if I do not use type hinting at all in that case. I say, do it properly or do not do it at all. Tomi Kaistila PHP Developer

Sam Barrow

18 years ago
On Tue, 2008-01-08 at 22:47 +0200, Tomi Kaistila wrote:
> > But the syntax is longer ("$a = 5" vs "$a = new Integer(5)"), and if you > > have a large application with hundreds of integers it starts to add up. > > Performance is also much worse when using objects for every variable. > I agree. That is what I meant by "ugly" and "clumsy" in my last post. > > > > This sounds like, at least, a partial victory to adding proper type > > > hinting to PHP but in my mind it is not enough. I have counted that about > > > a quarter of all type hints that I would make is about matching the > > > argument's type to scalar. The rest are about separating specific scalars > > > from each other. In most cases integer and boolean. > > > > Better than nothing. > No, it is not. It is actually better to not fix it at all, if you cannot fix > it properly. Especially in this case when the feature has already once been > left "unfinished". > > The benefit is zero for me, if when I need to hint a boolean and can only hint > a scalar. In that case, I will still have to do the validating myself. I > might as well leave the type hinting out all together, instead of waste > resources for having first PHP check that it is a scalar and then my own code > to check that it is a boolean. > > Actually, if I can only hint "scalar" in general, I am essentially telling the > users of my API that they should be ready for two types of errors. That which > occurs when the argument is not a scalar and that which occurs when it is not > a boolean. One will be a PHP error and the second will be an exception. Would > be easier on them if I do not use type hinting at all in that case.
Ok, but a scalar becomes useful for a couple of important things. Scalars are all displayable values, that can be stored in a db, outputted, etc. The scalar type hint prevents errors related to objects, resources, and arrays attempting to be stored in the db or echoed.

Tomi Kaistila

18 years ago
> Ok, but a scalar becomes useful for a couple of important things. > Scalars are all displayable values, that can be stored in a db, > outputted, etc. The scalar type hint prevents errors related to objects, > resources, and arrays attempting to be stored in the db or echoed.
True, but honestly I do not see a whole lot of use for this. In a word, it is very limited. _Too_ limited to be truly useful in my opinion. It is simple statistics; there is more need for specific types than scalars in general, when using type hinting. I do agree that it is a good idea to allow for type hinting scalars in general, but it should not be considered a solution. It is more like a comfortable add-on. And to be frank, this really feels like a secondary price (bronze at most). If pressed I would go bold say that this really feels like a screw. A way of saying that we went out of our way to fix this feature that was limited by adding a seemingly useful attribute to it, but in the end still leaving the feature limited and not really fixing anything. Tomi Kaistila PHP Developer

Sam Barrow

18 years ago
On Wed, 2008-01-09 at 00:58 +0200, Tomi Kaistila wrote:
> > Ok, but a scalar becomes useful for a couple of important things. > > Scalars are all displayable values, that can be stored in a db, > > outputted, etc. The scalar type hint prevents errors related to objects, > > resources, and arrays attempting to be stored in the db or echoed. > True, but honestly I do not see a whole lot of use for this. In a word, it is > very limited. _Too_ limited to be truly useful in my opinion. It is simple > statistics; there is more need for specific types than scalars in general, > when using type hinting. > > I do agree that it is a good idea to allow for type hinting scalars in > general, but it should not be considered a solution. It is more like a > comfortable add-on. > > And to be frank, this really feels like a secondary price (bronze at most). If > pressed I would go bold say that this really feels like a screw. A way of > saying that we went out of our way to fix this feature that was limited by > adding a seemingly useful attribute to it, but in the end still leaving the > feature limited and not really fixing anything.
A useful addon is better than none though, regardless. I agree this is not a full solution, but I'd rather have 3 type hints than 2.

Lukas Kahwe Smith

18 years ago
Hi, I am still wondering if we ever going to get a summary of this "discussion". regards, Lukas