Package is too large

[JakeQZ]

The package is currently 6.31 MB, with 573 files.

This arises from the following:

• The entire PHP documenation is copied to the DocBlock for each function, when only the type annotations are needed;
• A separate implementation is provided for each supported PHP version, when in most cases that is not necessary.

It can and should be slimmed down.

See MyIntervals/PHP-CSS-Parser#1482 for more info and background.

[shish]

This is a side effect of prioritising correctness, then maintainability, and then efficiency - I would love more efficiency too, but don't currently have time to focus on it D:

I think by far the biggest package-size win without affecting correctness would be being able to detect when the implementation of a function is identical across PHP versions, and stick that in a "common" library, so that only the functions which actually differ per-version have separate implementations per-version -- with quick napkin math I think that'd reduce package size by around 80%

(That said, doing that would reduce download size a lot, and increase CPU / memory use a little - probably not much, but it's a tradeoff worth being aware of)

Removing docstrings (except for type annotations) might be fairly simple - I'd consider that a minor loss for correctness, so I'd want another maintainer to give a second opinion before going ahead with that, but I'm personally ok with it (I'd also be curious to see what difference it makes in practice)

[JakeQZ]

Removing docstrings (except for type annotations) might be fairly simple - I'd consider that a minor loss for correctness, so I'd want another maintainer to give a second opinion before going ahead with that, but I'm personally ok with it (I'd also be curious to see what difference it makes in practice)

I've created #707 for this. It reduces the size by more than half.

As stated there:

Developers are unlikely to refer to [the DocBlocks] for the PHP documentation, and they can still inspect the return type to confirm that, e.g., false will not be returned.

[JakeQZ]

I think by far the biggest package-size win without affecting correctness would be being able to detect when the implementation of a function is identical across PHP versions, and stick that in a "common" library, so that only the functions which actually differ per-version have separate implementations per-version

A simple way of getting most of the way might be on a per-file basis:

• Generate files for PHP 8.1;
• When generating files for PHP 8.2, see if the file contents are the same as the PHP 8.1 version, and if so don't write that file;
• When generating files for PHP 8.3, see if the file contents are the same as the PHP 8.2 version (if it exists) or the PHP 8.1 version (if there is no version for PHP 8.2), and if so don't write that file;
• Etc.

Then in the version-loading bootstrapper logic we go in the other direction, e.g.

// Use `call_user_func` to avoid polluting the global namespace.
\call_user_func(
    function () {
        foreach (['8.5', '8.4', '8.3', '8.2', '8.1'] as $phpVersion) {
            if (\version_compare(PHP_VERSION, $phpVersion) >= 0) {
                $includeFilePath = __DIR__ . '/' . $phpVersion . '/apache.php';
                if (\file_exists($includeFilePath)) {
                    require_once $includeFilePath;
                    break;
                }
            }
        }
    }
);

(Not tested.)

Are these bootstrappers also auto-generated?

with quick napkin math I think that'd reduce package size by around 80%

80% is the maximum achievable; that won't be possible; but maybe somewhere around 60-70% would be.

increase CPU / memory use a little - probably not much, but it's a tradeoff worth being aware of)

I'm not seeing how it would increase CPU/memory use at all significantly. There might be an alternative to using version_compare() that is more efficient.

[JakeQZ]

A simple way of getting most of the way might be on a per-file basis:

Then in the version-loading bootstrapper logic we go in the other direction, e.g.

#709 for the bootstrapper.