Tuesday, June 18, 2024
Google search engine
HomeUncategorizedMy PHP Problems

My PHP Problems

For the past months I’ve been keeping a list of things I encountered in PHP
that I’ve found to be problematic, or in other words things that annoy

me having problems and being annoyed

This is not my definitive list and things that annoy me in PHP largely depend
on the things I’m working on, and for the past month I’ve been working on:

  • Phpactor: PHP language server
  • PHP-TUI TUI framework and port of Rust’s Ratatui.
  • CLI Parser: me messing about creating a new CLI argument parser
  • Work Project: large E-Commerce project based on Spryker – my current day contract.

There are lots of things I like in PHP and some of the points that follow
would annoy me in other languages too!


This is a strange one, yet one that bothers me. I clearly see the need for
static constructors, but I also cringe when using them unnecessarily. Should I
use static constructors for everything, a specific subset of objects or
introduce them only when required?

What’s the big deal you ask? Consistency is the deal. I don’t want to have
to type new only to realise there the class has a private constructor or that there are
static constructors which I should be using, but I also don’t want to
introduce pointless indirection fo the sake of consistency:


// instantiation with the new keyword
new Foobar(['foo1', 'foo2']);

// static new
Foobar::new(['foo1', 'foo2']);

// dedicated constructor with variadic

Langauges such as Rust and Go do not have this problem, mainly because they
don’t have the new keyword!

Both languages feature “structs” which can be created directly without a
constructor and both have unwritten conventions on using constructor functions
(in Rust they are conventionally attached to the struct – similarly to static
constructors in PHP).

The disadvantage of bypassing (or simply not having) a constructor is that you permit the
“unsupervised” creation of the data structure – you can’t control and enforce
the business invariants. However this is mitigated in both languages as they
both have package level visibility and a strong type system.

Am I suggesting we abolish the new keyword and adopt better types and
package level visibility? Yes? No? Maybe? I don’t know. The truth is it’s
just something that bugs me.

Annotations vs. Attributes

Our static analysis tools use annotations:

class Foobar
     * @var Foobar[]
    public array $foobars;

This is painful when you need to use this metadata in other contexts:

class FoobarDTO
     * @var Foobar[]
    #[Collection(type: "Foobar")]
    public array $foobars;

So our tooling can switch to attributes? Let’s look at a generic type defined
with an annotation:


class FoobarDTO
     * @var Foobar
    public Foobar $foobar;

In attribute land this becomes:


use PhpType{GenericClassType,StringType,IntType};

class FoobarDTO
    #[GenericClassType('Foobar', [
        new StringType(),
        new ClosureType(new StringType(), new IntType()),
    public Foobar $foobar;

Is that better? Of course not, it’s HORRIBLE. We are importing types for
types. We could also imagine:


use PhpType;

class FoobarDTO
    public Foobar $foobar;

This would at least enable lower the barrier for sharing this metadata,
although from a usage point of view it’s arguably more cumbersome than an
annotation, it’s still annoying.

Generics would solve much of this pain, but it is tricky.
One solution that has been discussed is extending the PHP parser to accept
(but ignore) generic annotations
purely for the sake of static analysis tools, for example:

class Hello
    public Foobar<string,Closure> $foobar;

This would allow the array syntax, and maybe we can even get away with other exotic types like Closure:


use PhpType;

class FoobarDTO
    public Foobar<string,Closure(string):int> $foobar;

I like this! The PHP engine at runtime will only see
Foobar but the Reflection API will provide access to the “rich” types
facilitating static analysis tools and helping to eliminate many of the
incidental problems we have in the ecosystem.

No Nested Attributes

While working on a prototype for
PHPBench I was experimenting with
allowing users to compose benchmarking pipelines.

PHPBench needs to analyse files which may not even have the same autoloader as
the main process:


use PhpbenchxInstructionsIterations;
use PhpbenchxInstructionsPhpSampler;

final class TimeBench
    #[Iterations(10, new PhpSampler(reps: 10, warmup: 1))]
    public function benchTime(): void
        foreach ([1, 2, 3, 4] as $b) {
            foreach ([1, 2, 3, 4] as $b) {

Nice! But there’s a catch. The Iterations attribute is just a class name. We
can reflect the name using native reflection because, it’s just a
name. The new PhpSampler however is a value and will invoke the
autoloader and fail because PHPBench doesn’t necessarily exist in that

Nested attributes would look something like this I guess:

final class TimeBench
    #[Iterations(10, PhpSampler(reps: 10, warmup: 1))]
    public function benchTime(): void
        // ...

This would allow PHPBench to read the attributes even if they did not exist in
the other process.

what’s that? you say this approach is flawed?, and you’re probably right, but
it would still be nice if refecting “nested” attributes didn’t require the

See the
RFC for the reasons why
nested were excluded from the final implementation.


This is something that didn’t really bother me until I used Go and Rust.

Deserializing byte streams to objects is our daily bread. Whether it be HTTP
requests or RPC messages. We need to ingest bytes and map them to data

In PHP we start with:


$userData = $_POST['user'];
$user = new User(

If you’re lucky there may be even be some if (!array_key_exists or even
Assert::arrayHasKey but more often than not we see people living on a
and just assuming that
everything will kinda work out.

We then have
serialization libraries such as JMS
and later the Symfony
. This
is a huge improvement, but both libraries are complex.

Maybe I was burned by JMS serializer earlier in my career, and I
still have nightmares about debugging the Serializer stack in API Platform.
I don’t instinctively reach for these tools when I’m writing a tool and
instead wrote my own simple library to deserialize into
because I wanted to do this:

$config = Invoke::new(Config::class, $config);

My library has no other API. It maps to an object field-for-field via.
the constructor and throws useful exceptions if values have the wrong
types, are missing or if there are extra fields. (it has some serious
limitations too, and I wouldn’t recommend using it in your projects

PROTIP: Map to DTOs. Don’t use serialization groups to “reuse” objects for multiple use cases. Don’t map to entities. DTOs are the correct targets for deserialization. This is not controversial.

Even more recently we have Valinor which
parses type annotations used by PHPStan and Psalm, including generics. Even
more recently we have Serde which has been
created by somebody who obviously feels my pain.

Valinor is probably my favourite library as it doesn’t require you to
duplicate your type definitions with annotations and your DTOs can be
completely agnostic of the serialization library.

Let’s look at deserializing a Strava Activity in Rust:

#[derive(Serialize, Deserialize)]
pub struct Activity {
    pub id: i64,
    pub title: String,
    pub activity_type: String,
    pub description: String,
    pub distance: f64,
    pub average_speed: Option<f64>,
    pub moving_time: i64,
    pub elapsed_time: i64,
let activity: Activity = serde_json::from_str("/** json payload */".as_str())?;

With Valinor:


class Activity {
    public int $id;
    public string $title;
    public string $activity_type;
    public string $description;
    public float $distance;
    public ?float $average_speed;
    public int $moving_time;
    public int $elapsed_time;

$mapper = (new MapperBuilder())->mapper();
$activity = $mapper->map(Activity::class, Source::json('// json payload')); 

Not a bad comparison! In fact it’s even possible that Valinor, since it now also supports
SOLVES this issue for me. But until I can prove it otherwise,
serialization/deserialization in PHP still annoys me but hey, at least
it’s not Node.

Promoted properties are nice, let’s use one!

class Foobar {
    public function __construct(private Foobar ...$foobars) {}

Oops, can’t use variadics in promoted properties. Why!?! See generics.

Iterator to Array: Preserve Keys

I get bitten by this over and again, yield from to yield from
another generator:


function one() {
    yield 'bar';
    yield from two();
function two() {
    yield 'bar';

$bars = iterator_to_array(one());


array(1) {
  string(3) "bar"

And then passing the second argument as false:


// ...
$bars = iterator_to_array(one(), false);

I get answer I was expecting:

array(2) {
  string(3) "bar"
  string(3) "bar"

Why? because false is preserve_keys.

Why does this bother me? Because over the years I assume the wrong default
behavior, and after realising my error I pass true here.

Am I saying this is wrong? Are my instincts wrong? Am I the problem? I don’t know. It just
annoyed me.

Iterators vs. Arrays

Why do we even need to call iterator_to_array! Why can’t array_map and
friends accept an iterator?


$collection = fetch_a_penguin_collection();
// PenguinCollection

$collection = array_map(iterator_to_array($foobars, true|false), function (Penguin $penguin) {
    return $penguin;
// array

Well, it would seem that there is more than one way to skin an iterator and
implicitly mapping an iterator to an array doesn’t really make much sense.

Should it be allowed to pass iterators to array functions? No, probably
. Has it bugged 🐞 me repeatedly? Yes it has. Am I wrong to be bugged? 🤷

Short closures cannot have statements

I like short closures! But I find my self converting my beloved short closures back to long closures

  • I need to add a statement
  • I need to debug it[*]

I’d much prefer to enjoy the short syntax while also being able to have
multiple statements:

$func = fn($foo) => {
    echo 'hello';
    echo 'world';
    return 'goodbye';

This would be better, and yes it can capture variables automatically and no
that’s not confusing.

[*] yes I am one of those primitive developers that doesn’t use a step
debugger all the time

There is another great atricle here which broadly argues for multi-line

Statement Blocks in General

And why stop there?


$foo = match ($bar) {
    'foo' => {
        $a = 1;
        return $a;

or even arbitrary scoping like in Rust?


$foo = 0;
    $bar = 1;
    $foo += $bar;

assert(false === isset($bar));

Functions that return false

Now we’re getting to the good old stuff.


$value = json_decode('this is not valid json');
$value = json_decode('null');

So … json_decode returns NULL if there is an error, but it also returns
NULL if the value is err. null (a valid JSON string).

We can pass flags: JSON_THROW_ON_ERROR to both, and get a really great an
informative error:

Fatal error: Uncaught JsonException: Syntax error

What about file_get_contents?


// false

I love the smell of false in the morning, but despite that I do wish that
all PHP’s built-in functions threw exceptions. We have the famous
safe library which does just that!
But if you’re like me then you don’t like coupling huge amounts of code in
perpetuity to an external library.

Is there anything we can do about this without breaking all the code?
declare(throw_exceptions=true) maybe? probably not 🤷. If all functions
threw exceptions however I would be less annoyed.

Inline Classes

In Go you can efficiently declare structs within structs to create
deep data structures:

struct Foobar {
   Foobar struct {
      Foo string
      Bar string

In PHP you can’t and we need to declare them separately:

class FooAndBar {
    public string $foo;
    public string $bar;

class Foobar {
   public FooAndBar;

Would this be a good idea?

class Foobar {
   public class {
      public string $foo;
      public string $bar;
   } $foobar;

I don’t know. But it would sure make some things easier.


This is a subjective post about things that annoy me, some of the
points are invalid and for sure people with far more context and brain power
than I have have considered them. It is also to be expected that I take for
granted things that would annoy other people.

If I had to choose one thing to fix in PHP it would be generics support.
Of the 11 annoyances 3 of them would be solved by generics. Generics support,
even by type erasure, would, I think, take the language to the next level.

I still enjoy PHP in comparison to some other languages, and it certainly has
practical some advantages over Rust and Go and I’m excited to see it evolve

Read More



Please enter your comment!
Please enter your name here

- Advertisment -
Google search engine

Most Popular

Recent Comments