Fix T_*_CAST syntax issues that caused argument list edge cases

[marcioAlmada]

Second, and more complete, attempt after #1629. Example of edge case: https://3v4l.org/eE5bq

<?php

const INT = 1;

function foo(){};

foo(INT);

Parse error: syntax error, unexpected '(INT)' (int) (T_INT_CAST) in /in/eE5bq on line 7

Relates to #1633, in some way

[marcioAlmada]

Hi! @nikic, @TazeTSchnitzel do you think this fix is doable?

[hikari-no-yume]

This issue had been bothering me for a long time. While I'd rather we just drop the cast tokens entirely and make the type names reserved words, we probably can't do that any time soon. This at least fixes the var_dump(int); thing, so I'm in favour of it.

[krakjoe]

@marcioAlmada there are conflicts, and this should target 7.0 ... if it will rebase clean then please just fix the conflicts ;)

ping me directly when you have done it ...

ta

[nikic]

@krakjoe As this is a major change to lexer output, I don't think it can target a patch release. For example, it will break PHP-Parser.

[krakjoe]

okay then

@marcioAlmada make that 7.1 ...

[nikic]

This doesn't take into account possible whitespace around the cast name.

[nikic]

This function should be in a .c file.

[nikic]

Not sure about this change in general. For example, doesn't the same issue also apply to a number of other places using parenthesized expressions? This PR would allow writing call(INT), but something like if (INT) {} would still fail.

I remember there is some reason for it, but doesn't come immediately to mind. Why is it that we handle this in the lexer instead of parsing "(" T_STRING ")" (where T_STRING can also be T_ARRAY and T_UNSET, of course)?

[nikic]

I guess the reason is that we'd have a shift/reduce conflict on '(' expression ')' vs. '(' T_STRING ')' expression.

[hikari-no-yume]

Ideally we'd reserve these words properly and then add '(' T_INT ')' etc to the parser. But I was cautious with my scalar types RFC about minimising BC impact, so I didn't do this. Alas.

[marcioAlmada]

@krakjoe this PR was created because of the callable types RFC, that if passed would have made it mandatory to fix the following edge case:

function x(callable(int):int $callback) {
#                  ^^^^^
}

But the callable types RFC was rejected. Considering important tools like PHP-Parser will break, this is the kind of technical debt we can live with until the next major release IMMO, specially if it doesn't get in the way of some important language feature until there.

I'd suggest to close this PR for now. If there is any urge to address this issue it will be trivial to resume. I just don't feel very comfortable to deliver a BC break now "just" for the sake of correctness.

/pings @nikic @Andrea

[coveralls]

Changes Unknown when pulling 466159a on marcioAlmada:cleanup/T_CAST into ** on php:master**.