Preserve information about used array syntax and number format.

[suralc]

Is it currently impossible to retrieve the syntax type used to define an array (array() vs []) or did I overlook something? (The grammar file already makes a difference, but creates the same Array_ instance, maybe pass an optional parameter if the new syntax is used?)

Also binary and octal numbers (0b1010 and 0777) are represented as LNumber. It may be useful to have the source format available there. (type attribute with a LNumber::TYPE_BINARY constant or similiar)

This may be used for some static analysis scenarios and detection of minimal php version requirements.

edit: I read about the second part in #26 It may still be nice to have atleast the used array notation as a flag on the node.

[suralc]

@nikic Both topics I raised here can be partially solved by extending the emulative Lexer and overriding getNextToken. Are you aware of any possible pitfalls when using the following code or should it be safe to use?

    public function getNextToken(&$value = null, &$startAttributes = null, &$endAttributes = null)
    {
        $tokenId = parent::getNextToken($value, $startAttributes, $endAttributes);
        // snipped....
        } elseif ($tokenId == Parser::T_ARRAY) {
            $startAttributes['traditionalArray'] = true;
        }

        return $tokenId;
    }

        elseif ($node instanceof Node\Expr\Array_) {
            if (!$node->hasAttribute('traditionalArray')) {
                $this->getResult()->addRequirement(RequirementReason::SHORT_ARRAY_DECLARATION, $node->getLine());
            }
        }

[nikic]

@suralc Nope, that's what I'd do as well. The lexer docs (https://github.com/nikic/PHP-Parser/blob/master/doc/component/Lexer.markdown) also suggest this in the lower part (the originalValue for literals). The top part of the page has a more general approach using token offsets. This can be used in some more complicated cases where just looking at the first or last token of a node is not enough.

[gonzaloserrano]

Hi, this addition would be great! Since we have part of our codebase that supports PHP 5.3 and sometimes our devs use short array syntax by mistake, we could write a script to analyze the changed files and check for this behaviour.

[suralc]

@gonzaloserrano Most of this should be managable to implement, I guess.
You could use http://php5.laurent-laville.org/compatinfo/blog/ for most circumstances (not short arrays, though). Some things are however not (yet) implemented there:
Take a look at the snippet above and the implementations(my project (some rewrites required), not related to the one linked above) here (Feel free to copy):

• Lexer
• PHP 5.4 Binary numbers
• PHP 5.4 Short array
• PHP 5.4 Always on - Short Echo

[gonzaloserrano]

Very cool projects @suralc, thanks!

[llaville]

@suralc Very interesting project (pvra). I've just take a look, and seems really promising.

As, I said, in an old version of my roadmap, I'd like to implement such PHP features analyser.
But my priorities changed. So I'm happy to see that an alternative exist !

Perharps, before the stable release of new major version 4.0, a similar analyser will be born.
Until this day come true, I've a big challenge, almost reach, detection of conditional code.

[nikic]

Added kind attribute for LNumber nodes in ae30f97.

[nikic]

And same is now also supported for array nodes: aa19912

As such closing this issue, finally :)