Hi Niels,

On 16-10-2023 23:19, Niels Dossche wrote:
Sorry for the resend... I accidentally replied to you only without including 
the list the first time.

On 15/10/2023 21:37, Frederik Bosch wrote:
Dear Niels,

First of all, thanks for all your hard work already on the DOM and SimpleXML 
extensions. I have been following your work in PHP-SRC, great! I am the author 
of this XSL 2.0 Transpiler in PHP package (https://github.com/genkgo/xsl). It 
is indeed possible to use workarounds for closures or object methods. I am 
relying on them in my package.

My suggestion for the future would be to add the following method.

public XSLTProcessor::registerFunctionsNS(string $namespace, array|ArrayAccess 
$functions): void

Then a user can register functions like this.

$xsltProcessorOrDomXpath->registerFunctionsNS('urn:my.namespace', array('upper-case', 
'strtoupper', 'pow' => fn ($a, $b) => $a[0]->textContent ** $b[0]->textContent, 
'other' => [$obj, 'method']);

Interesting suggestion. So you want to be able to use something like 
`my.namespace:function(...)` as I understand it.

Sorry I missed these messages completely. Unfortunately internals ends up in my spam. Sorry that I did not reply. What I want to use is this.

$processor->registerFunctionNS(
    'urn:my.ns',
    [
        'round-up' => fn ($arg1) => ceil($arg1),
        'round-down' => fn ($arg1) => floor($arg1)
    ]
);

<xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform"; xmlns:prefix="urn:my.ns">
    <xsl:value-of select="prefix:round-up(5.6)" />
    <xsl:value-of select="prefix:round-down(5.6)" />
</xsl:stylesheet>

While it follows the logic of XSL setParameter which also allows a namespace and DOM setAttributeNS, createElementNS et al methods, it also allows you to keep registerPhpFunctions method as is. Because basically, it can then become shorthand for registerFunctionNS (simplified):

$processor->registerFunctionNS(
    'http://php.net/xsl',
    [
        'functionString' => fn ($func, ...$args) => call_user_func_array($func, $args),         'function' => fn ($func, ...$args) => call_user_func_array($func, $args)
    ]
);

I'm not sure that adding this is that much more beneficial though (complexity 
vs benefit trade-off).
I assume this is motivated by the fact that you can then use third party 
libraries while having to worry less about name clashes?
It allows me to use namespaces as intended.
Let's say we add non-namespace `registerFunction(string $name, callable 
$callback): void`, you can then still use a convention of using a prefix, thus 
_kinda_ achieving the same.
I don't understand this one. Can I use a prefix without namespace?

In any case, this is going to be hard to support in combination with the 
underlying library (libxslt).
That's because the function namespace registration is process-wide, so this 
cannot be changed at runtime and certainly not for ZTS SAPIs.
So namespaces http://php.net/xsl and http://php.net/xpath are already process-wide, right? I do not see why more custom namespaces then would be a problem, but I have very little knowledge of libxslt library.

The registered functions should use the same methodology as php:function(). 
Hence, string casting of arguments is something the library user should do. I 
would leave registerPHPFunctions as is, and maybe discourage it in favor of the 
method above. What if both are called? I think it would be most clear if the 
registerFunctionsNS method throws InvalidArgumentException when 
http://php.net/xsl or http://php.net/xpath is passed as namespace.

Cheers,
Frederik
Cheers
Niels


On 13-10-2023 00:39, Niels Dossche wrote:
I'm looking to extend the functionality of calling PHP functions from within 
the DOMXPath or XSLTProcessor classes.

In case you're unfamiliar here's a quick rundown.
The DOMXPath class allows you to execute XPath queries on a DOM tree to lookup 
certain nodes satisfying a filter.
PHP allows the user to execute function callbacks within these. For example 
(from the manual):
    $xpath->query('//book[php:functionString("substr", title, 0, 3) = "PHP"]');
This will read the title element's text content, call substr on it, and then compare the 
output against "PHP".
You can not only call builtin functions, but also user functions.

To be able to call PHP functions, you need to use 
DOMXPath::registerPhpFunctions() 
(https://www.php.net/manual/en/domxpath.registerphpfunctions.php).
You either pass in NULL to allow all functions, or pass in which function names 
are allowed to be called.

Similarly, XSLTProcessor has the same registerPhpFunctions() method.
For XSLT it's mostly used for performing arbitrary manipulations on input data.
Normally the output of the function is put into the resulting document.


So what's the problem?
The current system doesn't allow you to call closures or object methods.
There are tricks you can do with global variables and global functions to try 
to work around this, but that's quite cumbersome.

There are two feature requests for this on the old bugtracker:
    - https://bugs.php.net/bug.php?id=38595
    - https://bugs.php.net/bug.php?id=49567

It's not hard to implement support for this, the question is just what API we 
should go with.
Based on what I've read, there are at least two obvious options:


OPTION 1) Extend registerPHPFunctions() such that you can pass in callables

```
// Adapted from https://bugs.php.net/bug.php?id=38595
$xslt->registerPHPFunctions(array(
     'functionblah', // Like we used to
     'func2' => fn ($x) => ...,
     'func3' => array($obj, 'method'), // etc
));
```

Example: Using php:function("func3") inside XPath/XSLT in this case will result 
in calling method on $obj.
Similarly func2 will call the closure, and functionblah in the snippet just 
allowlists calling functionblah.

It's a backwards compatible solution and a natural extension to the current 
method.
It may be hard to discover this feature compared to having a new API though.

Furthermore, once you pass in function names to registerPHPFunctions(), you're 
restricting what can be called.
For example: imagine you want to call both ucfirst() and $obj->method(), so you 
pass in an entry like func3 in the above example.
Now you have to pass in ucfirst to registerPHPFunctions() too, because 
registerPHPFunctions() acts as an allowlist. May be a bit inconvenient.


OPTION 2) Add new methods to register / unregister callables

This may be the cleaner way to go about it on first sight, but there's a 
potential BC break when new methods clash in user-defined subclasses.

Question here is: what about the interaction with registerPHPFunction?
What if both registerPHPFunction() and the register method add something with 
the same name?
What if registerPHPFunction() didn't allowlist a function but the register 
method added it, may be a bit confusing for users.
The interaction may be surprising.



Please let me know your thoughts.

Cheers
Niels


Again, whatever solution you choose, I am happy with all your contributions.

Cheers,
Frederik

--
PHP Internals - PHP Runtime Development Mailing List
To unsubscribe, visit: https://www.php.net/unsub.php

Reply via email to