# RFC - Array Of for PHP 7

- Thread: https://externals.io/message/100946
- Original archive: https://news-web.php.net/php.internals/100946
- Started by harezlak.dev@gmail.com on 2017-10-25T20:41:29+00:00
- Messages: 37
- RFC: Array Of (Declined) — https://externals.io/rfc/arrayof.md
- Page 1 of 2: messages 1 to 25 (25 per page, add `?page=N` to the URL)
- Next page: https://externals.io/message/100946.md?page=2
- Reply by email to internals@lists.php.net
- More for agents: search with https://externals.io/search.md?q={query} · RFCs at https://externals.io/rfc.md · JSON API at https://externals.io/api · documentation at https://externals.io/llms.txt

---

## #100946 — harezlak.dev@gmail.com — 2017-10-25T20:41:29+00:00

URL: https://externals.io/message/100946#100946

Hallo,

I would like to create a RFC for PHP 7, but the same RFC was created and
declined 3 years ago for PHP 5.4. PHP 7 support much better type hinging so
I think this RFC should be voted again.

What should I do? Should I create the RPF in common way?

Link to mentioned RFC:
https://wiki.php.net/rfc/arrayof

---

## #100947 — Sara Golemon — 2017-10-25T20:53:23+00:00

URL: https://externals.io/message/100946#100947 · In reply to: #100946

> I would like to create a RFC for PHP 7, but the same RFC was created and
> declined 3 years ago for PHP 5.4. PHP 7 support much better type hinging so
> I think this RFC should be voted again.
>
> What should I do? Should I create the RPF in common way?
>
> Link to mentioned RFC:
> https://wiki.php.net/rfc/arrayof
>
1. Approach the original author(s) directly. You'll find Joe is much
more active in PHP atm (he's also an RM for 7.1), though Phil is
decently responsive via Twitter.
2. If that fails, go ahead and create your own RFC which references
their original version.

Given that the previous version *did* fail, you should take extra care
to identify and address the reasons why it failed.  The lack of scalar
type hinting was likely a component, but I doubt it was the overriding
reason for most.

-Sara

---

## #100949 — Dan Ackroyd — 2017-10-25T21:52:37+00:00

URL: https://externals.io/message/100946#100949 · In reply to: #100946

> Hallo,
>

> I think this RFC should be voted again.

Although 'array of' as well as full blown generics are popular ideas,
I'm pretty certain the RFC failed due to the type check on the array
happening every time the 'array of' was passed from one function to
another, which would make it too slow to actually use.

An 'array of' RFC almost certainly needs a different programming
approach to have a change of succeeding.

> What should I do?

If you can, either learn a lot of C, or hire a world class C
programmer who is familiar with PHP internals for several months, to
make a new implementation.

Joe might have some thoughts on alternative approaches.

cheers
Dan

---

## #100965 — Rowan Collins — 2017-10-27T12:35:56+00:00

URL: https://externals.io/message/100946#100965 · In reply to: #100949

>Although 'array of' as well as full blown generics are popular ideas,
>I'm pretty certain the RFC failed due to the type check on the array
>happening every time the 'array of' was passed from one function to
>another, which would make it too slow to actually use.

The fundamental problem with extending PHP's type declarations - either to more complex types, or to other code elements such as properties - is that they are currently always checked at run-time, not with a separate static analyser. It's notable that in Go, the runtime type checks are considered a "development only" option, and strongly discouraged in production code due to performance impact.

I wrote about this a few months back here: http://rwec.co.uk/q/php-type-system

One possibility if we do want to keep the current run-time approach is to cache type information against values (zvals) so subsequent checks become quicker. The problem is working out exactly what to store and when to invalidate it - if you append 42 to an array that passed int[] on the last check, the ideal would be for us to know the type is still valid.

Regards,

-- 
Rowan Collins
[IMSoP]

---

## #100999 — Tony Marston — 2017-10-31T10:35:36+00:00

URL: https://externals.io/message/100946#100999 · In reply to: #100946

""Michal Harezlak""  wrote in message 
news:CALTzvrKFLnCW4+D-FU4cp2rmOScF77vZFz4f1cuVP3VyMzkkeg@mail.gmail.com...
>
>Hallo,
>
>I would like to create a RFC for PHP 7, but the same RFC was created and
>declined 3 years ago for PHP 5.4. PHP 7 support much better type hinging so
>I think this RFC should be voted again.
>
>What should I do? Should I create the RPF in common way?
>
>Link to mentioned RFC:
>https://wiki.php.net/rfc/arrayof

This strikes me as being nothing more than a micro-optimisation that does 
nothing but pander to the laziness of certain programmers. Instead of having 
to write a few lines of code to validate something they want the language to 
do it for them. It may come as a surprise to some people, but being a 
programmer actually involves the writing of program code. It is not 
sufficient to express an idea and have the language fill in all the details 
as that forces the language to have to detect and deal with a myriad of 
possibilities.

I would evaluate each proposed change to the language with a simple 
question - does it provide the greatest good to the greatest number? 
Considering the fact that this RFC will only benefit a miniscule minority of 
developers yet make the language more complicated, slower to run, and more 
difficult to maintain as more and more edge cases are identified as "bugs", 
it offers negative benefits to the vast number of programmers who are happy 
with the language as it currently exists. As such it fails that test and 
should be rejected.

-- 
Tony Marston

---

## #101000 — Thomas Hruska — 2017-10-31T14:39:39+00:00

URL: https://externals.io/message/100946#101000 · In reply to: #100999

> ""Michal Harezlak""  wrote in message 
> news:CALTzvrKFLnCW4+D-FU4cp2rmOScF77vZFz4f1cuVP3VyMzkkeg@mail.gmail.com...
> > […]
> 
> This strikes me as being nothing more than a micro-optimisation that 
> does nothing but pander to the laziness of certain programmers. Instead 
> of having to write a few lines of code to validate something they want 
> the language to do it for them. It may come as a surprise to some 
> people, but being a programmer actually involves the writing of program 
> code. It is not sufficient to express an idea and have the language fill 
> in all the details as that forces the language to have to detect and 
> deal with a myriad of possibilities.
> 
> I would evaluate each proposed change to the language with a simple 
> question - does it provide the greatest good to the greatest number? 
> Considering the fact that this RFC will only benefit a miniscule 
> minority of developers yet make the language more complicated, slower to 
> run, and more difficult to maintain as more and more edge cases are 
> identified as "bugs", it offers negative benefits to the vast number of 
> programmers who are happy with the language as it currently exists. As 
> such it fails that test and should be rejected.

I agree.

Features that result in significant improvements to algorithmic 
complexity but can't be implemented in userland are the sort of features 
that should be priority for core.  That's another way to look at the 
question that Tony put forth but in more concrete terms.  Internals 
already tends to adhere to that sort of policy anyway even if it isn't 
formally written anywhere.

-- 
Thomas Hruska
CubicleSoft President

I've got great, time saving software that you will find useful.

http://cubiclesoft.com/

And once you find my software useful:

http://cubiclesoft.com/donate/

---

## #101001 — Kalle Sommer Nielsen — 2017-10-31T15:20:14+00:00

URL: https://externals.io/message/100946#101001 · In reply to: #100999

Hi Tony

2017-10-31 11:35 GMT+01:00 Tony Marston <TonyMarston@hotmail.com>:
> This strikes me as being nothing more than a micro-optimisation that does
> nothing but pander to the laziness of certain programmers. Instead of having
> to write a few lines of code to validate something they want the language to
> do it for them. It may come as a surprise to some people, but being a
> programmer actually involves the writing of program code. It is not
> sufficient to express an idea and have the language fill in all the details
> as that forces the language to have to detect and deal with a myriad of
> possibilities.

I do understand where you are coming from, but I don't necessarily
agree on this topic. We can (hopefully) agree that programming
language design is hard, because we need to determine how fine a line
we should have between things thats an integral part of the language,
its standard library or its extensions and how much power the
programmer has in their arsenal to do crazy things.

If we boil things down, then we didn't really need the scalar type
hints, PHP had been working perfectly fine for 20 years without it and
while it does not add anything but a couple of checks at
compile/runtime, its essentially "laziness of certain programmers" it
becomes useful to. Another example is constant visibility modifiers in
PHP 7.1.

I think one of the advocates for features that are within that
category you mention can sometimes be productivity and rid of
boilerplate code. For this case with 'Array Of', I think it makes
perfect sense to add with PHP7's improved type system on that regard,
but thats my personal opinion.

> I would evaluate each proposed change to the language with a simple question
> - does it provide the greatest good to the greatest number? Considering the
> fact that this RFC will only benefit a miniscule minority of developers yet
> make the language more complicated, slower to run, and more difficult to
> maintain as more and more edge cases are identified as "bugs", it offers
> negative benefits to the vast number of programmers who are happy with the
> language as it currently exists. As such it fails that test and should be
> rejected.

Tho you said its a micro optimization, would argue that (see [1]), it
far from makes the code complicated, internally it doesn't add any
complexity and only adds a member to the arg_info, which is an
unsigned char, it wouldn't do anything unless a type is specified
anyway and the slower to run argument above is pretty void, sure it
adds a few CPU instructions but its not something you will feel unless
you are Facebook, in which case you already re-implemented the
language on your own.

I fail to see how it offers "negative benefits to the vast number of
programmers who are happy with the language as it currently exists", I
myself don't like PDO, so I just use mysqli instead, great. If its not
something that affects the programmer and the programmers code
continue to run, I fail to see how it negatively impacts the vast
majority. If I asked you how you feel about the exif extension now
supports streams as arguments instead of only file names, would you
care much unless you are actively using the exif extension? Probably
not.


[1] https://gist.github.com/KalleZ/c7ba4f78314c989e27710e4fa14e2f3e



-- 
regards,

Kalle Sommer Nielsen
kalle@php.net

---

## #101002 — Levi Morrison — 2017-10-31T17:37:17+00:00

URL: https://externals.io/message/100946#101002 · In reply to: #100949

My thoughts on the RFC:

  - The syntax `typename[]` is a poor. Consider how nullability might
bind `?int[]`. Is that `(?int)[]` or `?(int[])` ? Here are some
variants that do not have this issue:
    [?int] and ?[int]
    array<?int> and ?array<int>
  - Our current infrastructure requires us to check every element of
the array for conformance. While undesirable I believe this can be
optimized away in the future if we care enough, and therefore not a
show-stopper.

---

## #101005 — Andrea Faulds — 2017-11-01T01:36:40+00:00

URL: https://externals.io/message/100946#101005 · In reply to: #101002

Hi Levi,

Levi Morrison wrote:
> My thoughts on the RFC:
>
>   - The syntax `typename[]` is a poor. Consider how nullability might
> bind `?int[]`. Is that `(?int)[]` or `?(int[])` ? Here are some
> variants that do not have this issue:
>     [?int] and ?[int]
>     array<?int> and ?array<int>

Thank you for bringing this up. The introduction of nullables means that 
the type[] syntax is problematic and should probably be avoided now. As 
you say, there is an issue of ambiguity as to whether it would be 
interpreted as (?int)[] or ?(int[]). Unfortunately, both of these are 
reasonable interpretations and might be desired. So if we use that 
syntax, we'd have to add brackets. It would be simpler and prevent any 
misunderstandings if we use a syntax that is unambiguous in the first place.

Personally I might like generics-esque angular brackets, but a problem 
there is that << and >> are operators with their own tokens, so 
supporting nested arrays (array<array<array>>) would require some sort 
of lexer hack.

>   - Our current infrastructure requires us to check every element of
> the array for conformance. While undesirable I believe this can be
> optimized away in the future if we care enough, and therefore not a
> show-stopper.
>

Something I've thought about but not gotten round to implementing is a 
typed array value. That is, you could write `$x = array<int>(1, 2, 3);` 
and `$x` would contain a typed array that would enforce the types of 
values added to it. This feels like a cleaner solution, but it 
introduces a new issue: does an array<int> type declaration require an 
array<int> value, or will it implicitly cast from vanilla array?

Thanks.
-- 
Andrea Faulds
https://ajf.me/

---

## #101016 — Rowan Collins — 2017-11-01T14:40:55+00:00

URL: https://externals.io/message/100946#101016 · In reply to: #101002

>  - Our current infrastructure requires us to check every element of
>the array for conformance. While undesirable I believe this can be
>optimized away in the future if we care enough, and therefore not a
>show-stopper.

Do you have a mechanism in mind for "optimising this away"? I think releasing the feature without any optimisation will just lead to painful slowdowns when people use it without realising the implications.

As I said in my previous message, it seems to be a fundamental issue with PHP's current type declarations that they are a) runtime, and b) asserted at arbitrary points, not at assignment. So even if you optimise the check slightly, it's still a separate check every time you enter a new function, which puts a limit on optimisation.

The only way I've thought of to improve things is to cache the result of previous type checks, but this comes with all the normal challenges of caching. For instance, all assignments into the array would need to invalidate these cached checks, or check the new value against each. References would be particularly tricky, as they were with the property type hint proposal.

If we selectively invalidated cached checks, we'd be running specific type checks at every variable assignment, but not letting the user actually assert on those checks. At which point, should we approach from the other direction, and let the user declare variable types, rather than assert value types? 

This is kind of implied if we implement generics, too, since explicitly constructing an instance of list<int> is asking the runtime to check future assignments to members of that list. In effect, could declaring "int $foo = 42;" be considered equivalent to "$foo = new checkedScalar<int>(42);"?

Regards,

-- 
Rowan Collins
[IMSoP]

---

## #101026 — Michael Morris — 2017-11-02T04:31:59+00:00

URL: https://externals.io/message/100946#101026 · In reply to: #101016

Drupal 8 accomplishes this through `assert()` and a helper class.

function foo ( array $a ) {
  assert( Inspector::assertAllStrings( $a ));
}

This could be improved by having an collectionof operator similar to the
instanceof operator.

function ( array $a ) {
  assert( $a collectionof string );
}

I say "collectionof" because while arrays are the most common traversable
objects, they aren't the only ones.  The above approach, combined with the
existing assert structure, can provide the dev time checking of the code
while in under development, and it can be turned off in production. Or it
can be used outside of assert to be checked at all times.

Since this invokes a new keyword I think that would mean this solution
would be PHP 8.

Brackets might be used with instance of to notify it to traverse down one
level maybe??

function ( array $a ) {
  assert( $a instanceof [string] );
}

This avoids any BC issues, but it looks odd.  Not as odd as \ for a
namespace operation :P  But odd.


On Wed, Nov 1, 2017 at 10:40 AM, Rowan Collins <rowan.collins@gmail.com>
wrote:

> On 31 October 2017 17:37:17 GMT+00:00, Levi Morrison <levim@php.net>
> wrote:
> > […]
>
> Do you have a mechanism in mind for "optimising this away"? I think
> releasing the feature without any optimisation will just lead to painful
> slowdowns when people use it without realising the implications.
>
> As I said in my previous message, it seems to be a fundamental issue with
> PHP's current type declarations that they are a) runtime, and b) asserted
> at arbitrary points, not at assignment. So even if you optimise the check
> slightly, it's still a separate check every time you enter a new function,
> which puts a limit on optimisation.
>
> The only way I've thought of to improve things is to cache the result of
> previous type checks, but this comes with all the normal challenges of
> caching. For instance, all assignments into the array would need to
> invalidate these cached checks, or check the new value against each.
> References would be particularly tricky, as they were with the property
> type hint proposal.
>
> If we selectively invalidated cached checks, we'd be running specific type
> checks at every variable assignment, but not letting the user actually
> assert on those checks. At which point, should we approach from the other
> direction, and let the user declare variable types, rather than assert
> value types?
>
> This is kind of implied if we implement generics, too, since explicitly
> constructing an instance of list<int> is asking the runtime to check future
> assignments to members of that list. In effect, could declaring "int $foo =
> 42;" be considered equivalent to "$foo = new checkedScalar<int>(42);"?
>
> Regards,
>
> --
> Rowan Collins
> [IMSoP]

---

## #101027 — Tony Marston — 2017-11-02T09:55:59+00:00

URL: https://externals.io/message/100946#101027 · In reply to: #101001

"Kalle Sommer Nielsen"  wrote in message 
news:CAJW__o3QJOe6G3ybmBcoCU=FCAdjZAcgbKTiJRBM3RS8Q0hBNQ@mail.gmail.com...
>
>Hi Tony
>
>2017-10-31 11:35 GMT+01:00 Tony Marston <TonyMarston@hotmail.com>:
> > […]
>
>I do understand where you are coming from, but I don't necessarily
>agree on this topic. We can (hopefully) agree that programming
>language design is hard,

I totally agree that compiler writing is hard, which is why I believe that 
only those things which are hard to do in userland code should be added to 
the language.

>because we need to determine how fine a line
>we should have between things thats an integral part of the language,
>its standard library or its extensions and how much power the
>programmer has in their arsenal to do crazy things.

For me the rule is simple - if something can already be done quite easily in 
userland code then it should not be forced into the language just because a 
small number of developers wnat to save themselves a few keystrokes. Making 
the language more complicated than it need be leads to a maintenance burden 
both for the core developers and the application developers who now have to 
deal with a growing array of indeciferable shortcuts.

>If we boil things down, then we didn't really need the scalar type
>hints, PHP had been working perfectly fine for 20 years without it and
>while it does not add anything but a couple of checks at
>compile/runtime, its essentially "laziness of certain programmers" it
>becomes useful to. Another example is constant visibility modifiers in
>PHP 7.1.
>
>I think one of the advocates for features that are within that
>category you mention can sometimes be productivity and rid of
>boilerplate code. For this case with 'Array Of', I think it makes
>perfect sense to add with PHP7's improved type system on that regard,
>but thats my personal opinion.

This type of checking can quite easily be done in userland code, so I see no 
reason why it should be built into the language. What percentage of userland 
developers would actually use such a feature? I wouldn't as I never pass 
around arrays of single types. I regularly use the $_POST array and a 
database record array which are both of mixed types.

> > […]
>
>Tho you said its a micro optimization, would argue that (see [1]), it
>far from makes the code complicated,

>internally it doesn't add any complexity

I disagree. The language has to be changed to recognise that type of 
argument, then it has to iterate through the array checking that every 
member is of the designated type. In other words the langage now has to do 
what you are doing with a few lines of userland code.

>and only adds a member to the arg_info, which is an
>unsigned char, it wouldn't do anything unless a type is specified
>anyway and the slower to run argument above is pretty void, sure it
>adds a few CPU instructions but its not something you will feel unless
>you are Facebook, in which case you already re-implemented the
>language on your own.
>
>I fail to see how it offers "negative benefits to the vast number of
>programmers who are happy with the language as it currently exists", I

If it's put into the language then it affects 100% of the users, but what 
percentage of the user base would actually take advantage of this feature? 
If it's only 1% then for the other 99% it's a complete waste of time.

>myself don't like PDO, so I just use mysqli instead, great. If its not
>something that affects the programmer and the programmers code
>continue to run, I fail to see how it negatively impacts the vast
>majority.

It has complications in the language that are not used. Have you ever heard 
o the 80-20 rule? This is where 80% of the usage is simple and 20% is 
complicated, but that complicated 20% takes up 80% of the programming 
effort.  I was around in the 1980s during the switch to RISC (Reduced 
Instruction Set Computers) where it was found that by removing the 
complicated instructions which could be executed in a single cycle and 
replacing them with a series of simple instructions there were enormous 
benefits all round - the chips became smaller, were cheaper to manufacture, 
the cycles faster, the energy consumption smaller and with less heat 
generated. It was a win-win all round.

By keeping out the frivolous froth from the language and restricting it to a 
set of really useful features you end up with a smaller language which is 
faster to run and easier to maintain. IMHO this RFC does not benefit the 
greater PHP community, it just saves a few keystrokes for the lazy 
developer.

>If I asked you how you feel about the exif extension now
>supports streams as arguments instead of only file names, would you
>care much unless you are actively using the exif extension? Probably
>not.

There is a great deal of difference between an extension which does 
something complicated which cannot easily be done in userland code and 
changing the core language just to save a few keystrokes. There are dozens 
of extensions that I don't use, but I would never call for them to be 
scrapped. Changing the core language is something else altogether. It's 
there whether I want it or not, and it affects the execution of every 
function or method call.

>[1] https://gist.github.com/KalleZ/c7ba4f78314c989e27710e4fa14e2f3e
>

-- 
Tony Marston

---

## #101028 — lists@rhsoft.net — 2017-11-02T10:13:45+00:00

URL: https://externals.io/message/100946#101028 · In reply to: #101027

Am 02.11.2017 um 10:55 schrieb Tony Marston:
> "Kalle Sommer Nielsen"  wrote in message 
> > […]
> 
> If it's put into the language then it affects 100% of the users, but 
> what percentage of the user base would actually take advantage of this 
> feature? If it's only 1% then for the other 99% it's a complete waste of 
> time

how does any feature you don't use affect you?

i don't care about pdo and many other core extensions nor about 
namespaces, traits and so on - but they don't affect me at all

---

## #101032 — Mark Randall — 2017-11-02T12:09:50+00:00

URL: https://externals.io/message/100946#101032 · In reply to: #101005

> Thank you for bringing this up. The introduction of nullables means that 
> the type[] syntax is problematic and should probably be avoided now. As 
> you say, there is an issue of ambiguity as to whether it would be 
> interpreted as (?int)[] or ?(int[]). 

That would be a shame, TypeName[] is a defacto standard already for PHP 
docs (not to mention a absolute standard for many programming languages) 
and if it can be made to work it would be one less piece of syntax.

IMHO ?(int[]) would be the way to go staying with the current trend, or 
alternatively bite the bullet and specific both like we do in docblocs:

function fn(null | int[] $x) { ... }

Generics gives the less ambiguous option:

function fn(?array<int> $x) { ... }

This would become a lot less of an issue if we had function overloading 
in future *crosses fingers*.

> > […]

Which brings up perhaps a longer term point, does the language need an 
inbuilt alternative to "array" that implements ArrayAccess and provides 
for an "Operator array" that will allow it to be cast to an native array 
just as __tostring does for objects that implement it.

The benefit would be that type checking could be performed on 
initialization and assignment, and with "operator array" it would be 
transparent.

The downside I see? Well, being a generic object it would not exhibit 
traditional PHP copy-on-write behaviour, unless that was somehow baked 
into the language as part of the class definition (in a similar manner 
to "abstract" or "implements CopyOnWrite").

> Something I've thought about but not gotten round to implementing is a 
> typed array value. That is, you could write `$x = array<int>(1, 2, 3);` 
> and `$x` would contain a typed array that would enforce the types of 
> values added to it. This feels like a cleaner solution, but it 
> introduces a new issue: does an array<int> type declaration require an 
> array<int> value, or will it implicitly cast from vanilla array?

Would an "operator array" not solve this in the same manner that 
__tostring does?

I think once we have operator overloading (and supported by native 
types) then we should be able to do the following:

function fn(MyCollection<int> $y) { ... }

generic class MyCollection<T1> {
   public function __construct(array $in) {
     foreach ($in as $x) {
       if (!($x typeof T1)) {
         throw new InvalidArgumentException('...');
       }
     }
   }
}

fn(['hello', 'world']);

And the engine should be smart enough to create a new instance of 
MyCollection<int> with the defined constructor. I say overloading would 
be necessary as several different constructors may be required.

Equally:

generic class SomeArray<T1> {
    ...
    public function __operator_array(): array {
       return $this->privateInternalArray;
    }
    ...
}

function fn(array $x) { ... }
fn(new SomeArray<int>(1, 2, 3));

Admittedly I do not know the PHP core with any degre of intimacy, but 
from a language perspective I can't think of a better way than generics, 
casting operators, and overloaded function constructors...

... and yes, that did include almost every major language wishlist item 
in a single go. Chalk it up to me romancing my long lost love of C++.

> Thanks.

--
Mark Randall

---

## #101033 — Larry Garfield — 2017-11-02T19:21:09+00:00

URL: https://externals.io/message/100946#101033 · In reply to: #101026

> Drupal 8 accomplishes this through `assert()` and a helper class.
>
> function foo ( array $a ) {
>   assert( Inspector::assertAllStrings( $a ));
> }
>
> This could be improved by having an collectionof operator similar to the
> instanceof operator.
>
> function ( array $a ) {
>   assert( $a collectionof string );
> }
>
> I say "collectionof" because while arrays are the most common traversable
> objects, they aren't the only ones.  The above approach, combined with the
> existing assert structure, can provide the dev time checking of the code
> while in under development, and it can be turned off in production. Or it
> can be used outside of assert to be checked at all times.
>
> Since this invokes a new keyword I think that would mean this solution
> would be PHP 8.
>
> Brackets might be used with instance of to notify it to traverse down one
> level maybe??
>
> function ( array $a ) {
>   assert( $a instanceof [string] );
> }
>
> This avoids any BC issues, but it looks odd.  Not as odd as \ for a
> namespace operation :P  But odd.

While I normally strongly agree with supporting all traversables, not
just arrays, in this case I don't think it works.  The whole point of
using a traversable is that you don't have all of the values up front. 
If you did... you'd just have an array.  So checking "all values" of a
traversable in one point in time is a destructive operation as it runs
out the iterator, which if it's an infinite iterator could, erm, be bad.

Rather, arrays can/should be checked at once (as above), whereas a
traversable should be able to declare that it only returns a given type,
and if it ever tries to return a different type then the traversable
itself type-errors.

--Larry Garfield

---

## #101035 — Michael Morris — 2017-11-02T22:22:46+00:00

URL: https://externals.io/message/100946#101035 · In reply to: #101033

On Thu, Nov 2, 2017 at 3:21 PM, Larry Garfield <larry@garfieldtech.com>
wrote:

>
> While I normally strongly agree with supporting all traversables, not
> just arrays, in this case I don't think it works.


Hmm.. You're right. Traversables would be best served by having another
method to indicate their return type as part of the interface. Adding this
to the existing traversable interface would be BC breaking. A new interface
would be best and having another interface isn't without precedent -
countable already exists.

This new interface would the PHP engine on what to do.

So, what about this syntax

$a instanceof string[]

If the engine sees this and it's an array, all the values are checked to
see if it's a string.

Else if $a references a generator, the return type definition of the
generator will be checked, and a failure will occur if the generator
doesn't declare its return type.

Else if $a references an object and that object implements the
SingleReturnType* interface then the method of that interface,
getReturnType() is queried and it's return is checked to see if it matches.
In the above case 'string' would need to be returned.

I leave it to people smarter than me to find the holes in the above, but
this moron thinks it might work.

* Someone smarter than me can come up with a better name too :P

---

## #101036 — Andrea Faulds — 2017-11-03T02:27:02+00:00

URL: https://externals.io/message/100946#101036 · In reply to: #101032

Hi,

Mark Randall wrote:
> 
> > […]
>
> That would be a shame, TypeName[] is a defacto standard already for PHP
> docs (not to mention a absolute standard for many programming languages)
> and if it can be made to work it would be one less piece of syntax.
>
> IMHO ?(int[]) would be the way to go staying with the current trend,

Yes, it is a bit unfortunate to deviate from the docs. But array<> is 
also a familiar syntax, yet it's unambiguous and wouldn't require the 
new complication of parentheses. For that reason I don't find [] a 
compelling option. Syntax æsthetics are nice but in the end it's 
readability that matters.

  or
> alternatively bite the bullet and specific both like we do in docblocs:
>
> function fn(null | int[] $x) { ... }
>
> Generics gives the less ambiguous option:
>
> function fn(?array<int> $x) { ... }

This makes things even more complicated.

>
> This would become a lot less of an issue if we had function overloading
> in future *crosses fingers*.
>
> > […]
>
> Which brings up perhaps a longer term point, does the language need an
> inbuilt alternative to "array" that implements ArrayAccess and provides
> for an "Operator array" that will allow it to be cast to an native array
> just as __tostring does for objects that implement it.
>
> The benefit would be that type checking could be performed on
> initialization and assignment, and with "operator array" it would be
> transparent.
>
> The downside I see? Well, being a generic object it would not exhibit
> traditional PHP copy-on-write behaviour, unless that was somehow baked
> into the language as part of the class definition (in a similar manner
> to "abstract" or "implements CopyOnWrite").

Why create an almost-array that works almost-everywhere when you can 
improve the actual array that works actually everywhere?

>
> > […]
>
> Would an "operator array" not solve this in the same manner that
> __tostring does?

Your proposed objects would not be usable everywhere an array is, 
because they're not arrays, and by converting to an array you lose the 
type info, so we still have to iterate over the whole thing to type 
check. This would be significantly less useful than an actually typed 
array. I can see some benefit to it, but I'm not sure it's worth the effort.

>
> I think once we have operator overloading (and supported by native
> types) then we should be able to do the following:
>
> function fn(MyCollection<int> $y) { ... }
>
> generic class MyCollection<T1> {
>   public function __construct(array $in) {
>     foreach ($in as $x) {
>       if (!($x typeof T1)) {
>         throw new InvalidArgumentException('...');
>       }
>     }
>   }
> }
>
> fn(['hello', 'world']);
>
> And the engine should be smart enough to create a new instance of
> MyCollection<int> with the defined constructor. I say overloading would
> be necessary as several different constructors may be required.
>
> Equally:
>
> generic class SomeArray<T1> {
>    ...
>    public function __operator_array(): array {
>       return $this->privateInternalArray;
>    }
>    ...
> }
>
> function fn(array $x) { ... }
> fn(new SomeArray<int>(1, 2, 3));

A problem here is that we'd need to implement generics! That's quite the 
complex task. I do like the idea of generics, but I don't really expect 
them to get implemented (let alone accepted) any time soon. Maybe if I 
suddenly decide to take all my yearly paid holiday at once and am 
unusually determined. But there are much nicer things to go on holiday for…

>
> Admittedly I do not know the PHP core with any degre of intimacy, but
> from a language perspective I can't think of a better way than generics,
> casting operators, and overloaded function constructors...
>
> ... and yes, that did include almost every major language wishlist item
> in a single go. Chalk it up to me romancing my long lost love of C++.

We all have our wishlist features. :)

I think I share some of yours.

Thanks for your reply.

>
> > […]
>
> --
> Mark Randall


-- 
Andrea Faulds
https://ajf.me/

---

## #101037 — Mark Randall — 2017-11-03T05:22:08+00:00

URL: https://externals.io/message/100946#101037 · In reply to: #101036

> Your proposed objects would not be usable everywhere an array is, 
> because they're not arrays, and by converting to an array you lose the 
> type info, so we still have to iterate over the whole thing to type 
> check. This would be significantly less useful than an actually typed 
> array. I can see some benefit to it, but I'm not sure it's worth the 
> effort.

Thanks for the reply.

So if I am following right, would your goal in the short-to-mid term at 
least be to attach type information inside the HashTable, or as part of 
the zval value?

I may be pulling this out our my arse of course, but does the nature of 
an array not mean that the individual values inside can be used as 
references and potentially break the type restrictions and require 
all-item checking again?

$t1 = array<string>([ 'hello', 'goodbye']);
$val = &$t1[1];
$val = 1;
-> [ 'hello', 1]

I did take a look in zend_types.h half expecting to see a pointer to an 
optional structure defining an indirect function call + argument to be 
called whenever the zval_value was about to be changed, but if it's in 
there, I missed it.

--
Mark Randall

---

## #101038 — Tony Marston — 2017-11-03T10:33:04+00:00

URL: https://externals.io/message/100946#101038 · In reply to: #101028

wrote in message news:6643d10b-8703-693c-15c2-da338022ef41@rhsoft.net...
>
>
>
>Am 02.11.2017 um 10:55 schrieb Tony Marston:
> > […]
>
>how does any feature you don't use affect you?

Because the language itself becomes bloated with the capabilities it has to 
offer, look for and deal with. This makes it bigger and slower. For example, 
if you have a vehicle which is capable of going anywhere, over every types 
of terrain, in every climate from sub zero to blisteringly hot then at any 
one time you are not using all the capabilities, yet you are still carrying 
the around.

The core language should be kept lean and mean, and should only have new 
instructions added when (a) something cannot be done easily in userland 
code, and (b) when that something is of genuine use to a significant number 
of people. Complicating the language with something that can already be done 
with a few lines of userland code and which is desired by only a miniscule 
number of people does not fall into this category.

>i don't care about pdo and many other core extensions nor about namespaces, 
>traits and so on - but they don't affect me at all

-- 
Tony Marston

---

## #101039 — lists@rhsoft.net — 2017-11-03T10:48:23+00:00

URL: https://externals.io/message/100946#101039 · In reply to: #101038

Am 03.11.2017 um 11:33 schrieb Tony Marston:
> wrote in message news:6643d10b-8703-693c-15c2-da338022ef41@rhsoft.net...
> > […]
> 
> Because the language itself becomes bloated with the capabilities it has 
> to offer, look for and deal with. This makes it bigger and slower

unproven claim!

or why did 5.3, 5.4, 5.5 and 5.6 not speaking about 7.0/7.1 *all* have 
new features and where *faster* then the previous version - frankly you 
are raising alarm for no reason

---

## #101047 — Tony Marston — 2017-11-04T09:18:10+00:00

URL: https://externals.io/message/100946#101047 · In reply to: #101039

wrote in message news:941fd347-4a17-78b6-1bd7-4a5519aa722b@rhsoft.net...
>
>
>
>Am 03.11.2017 um 11:33 schrieb Tony Marston:
> > […]
>
>unproven claim!

It's pure common sense! You have to carry around the capability of doing 
something, then have tests everywhere to see if that capability is actually 
required or not at run-time.

There is a big difference between adding something to the language core 
which everyone has to load into memory, and having something in an extension 
which is entirely optional.

>or why did 5.3, 5.4, 5.5 and 5.6 not speaking about 7.0/7.1 *all* have new 
>features and where *faster* then the previous version - frankly you are 
>raising alarm for no reason

PHP 7 is faster than PHP 5 for various reasons, such as it being 64bit 
instead of 32bit, and improvements made to the engine itself, such as the 
AST. I submit that it would be smaller and faster if it did not have to 
carry around so much dross. Adding something to the core language just to 
save a few keystrokes for a small number of lazy developers falls into the 
category of dross.

-- 
Tony Marston

---

## #101048 — lists@rhsoft.net — 2017-11-04T10:41:36+00:00

URL: https://externals.io/message/100946#101048 · In reply to: #101047

Am 04.11.2017 um 10:18 schrieb Tony Marston:
> wrote in message news:941fd347-4a17-78b6-1bd7-4a5519aa722b@rhsoft.net...
> > […]
> 
> It's pure common sense! You have to carry around the capability of doing 
> something, then have tests everywhere to see if that capability is 
> actually required or not at run-time.

it depends on the implementation and just beause you say so does not 
prove anything and even if you need to measure, optimize and make 
decisions based on technical facts - what you do is "mimimi i say"

> There is a big difference between adding something to the language core 
> which everyone has to load into memory, and having something in an 
> extension which is entirely optional.
> 
> > […]
> 
> PHP 7 is faster than PHP 5 for various reasons, such as it being 64bit 
> instead of 32bit

WTF, only in your windows world which don't matter that much, everywhere 
else x86_64 is normal for many years and each software

> and improvements made to the engine itself, such as 
> the AST. I submit that it would be smaller and faster if it did not have 
> to carry around so much dross. Adding something to the core language 
> just to save a few keystrokes for a small number of lazy developers 
> falls into the category of dross

you ignored that practicaly *every* PHP version before PHP/ was faster 
*and* had new features compared to the previous one

---

## #101056 — Andrea Faulds — 2017-11-04T22:45:05+00:00

URL: https://externals.io/message/100946#101056 · In reply to: #101037

Hi Mark,

Mark Randall wrote:
> 
> > […]
>
> Thanks for the reply.
>
> So if I am following right, would your goal in the short-to-mid term at
> least be to attach type information inside the HashTable, or as part of
> the zval value?

Inside the HashTable, yes. There's no space to hide it in the zval, and 
it's not the zval it's specific to anyway.

>
> I may be pulling this out our my arse of course, but does the nature of
> an array not mean that the individual values inside can be used as
> references and potentially break the type restrictions and require
> all-item checking again?
>
> $t1 = array<string>([ 'hello', 'goodbye']);
> $val = &$t1[1];
> $val = 1;
> -> [ 'hello', 1]

Yes. There's a similar problem for implementing typed property. I think 
Bob Weinand might have been working on a solution to this (typed 
references), but I have no idea where that went, if anywhere.

Thanks.
-- 
Andrea Faulds
https://ajf.me/

---

## #101061 — Tony Marston — 2017-11-05T10:24:52+00:00

URL: https://externals.io/message/100946#101061 · In reply to: #101048

wrote in message news:d70cc49d-c397-3f09-d08d-b79b31014271@rhsoft.net...
>
>
>
>Am 04.11.2017 um 10:18 schrieb Tony Marston:
> > […]
>
>it depends on the implementation and just beause you say so does not prove 
>anything and even if you need to measure, optimize and make decisions based 
>on technical facts - what you do is "mimimi i say"

I have worked on software which provided lots of different options, which 
means that you have to keep testing if an option is being used or not. This 
is an overhead whether you like it or not.

> > […]

Can you prove that each new version was faster? Where is your evidence?

> > […]
>
>WTF, only in your windows world which don't matter that much, everywhere 
>else x86_64 is normal for many years and each software

Excuse me! Some of the major clients who use my ERP application only use 
Windows servers, so your claim that Windows does matter is completely bogus.

> > […]
>
>you ignored that practicaly *every* PHP version before PHP/ was faster 
>*and* had new features compared to the previous one

Just think how much faster and easier to maintain it would be if all this 
save-a-few-keystrokes dross had not been added in the first place.

-- 
Tony Marston

---

## #101063 — lists@rhsoft.net — 2017-11-05T12:25:43+00:00

URL: https://externals.io/message/100946#101063 · In reply to: #101061

Am 05.11.2017 um 11:24 schrieb Tony Marston:
> wrote in message news:d70cc49d-c397-3f09-d08d-b79b31014271@rhsoft.net...
> > […]
> 
> I have worked on software which provided lots of different options, 
> which means that you have to keep testing if an option is being used or 
> not. This is an overhead whether you like it or not.

maybe your implementation was bad

> > […]
> 
> Can you prove that each new version was faster? Where is your evidence?

everbody knows that and can benchmark it at any time, but if it makes 
you happy that others are doing your homework

https://www.phpclasses.org/blog/post/493-php-performance-evolution.html

> > […]
> 
> Excuse me! Some of the major clients who use my ERP application only use 
> Windows servers, so your claim that Windows does matter is completely 
> bogus.

how does that change the fact that your claim "such as it being 64bit 
instead of 32bit" is nonsense when most of the benchamrks and production 
servers out there are running PHP on x86_64 with 86_64 builds for a 
decade now?

> > […]
> 
> Just think how much faster and easier to maintain it would be if all 
> this save-a-few-keystrokes dross had not been added in the first place.

again: unproven claim, but in your own world a hashtable probably is 
also not O(1) or you are just not capable to optimize software at all 
but then stop claim others aren't too

---

Next page: https://externals.io/message/100946.md?page=2