Nightmares on type hints, annotations, and aop... View as Markdown

Hi Bishop,

De : bishop.bettini@gmail.com [mailto:bishop.bettini@gmail.com] De la part de Bishop Bettini

THING 1: Hints are just is_* wrappers

Wrong. Only strict types are.

THING 2: Hints are truthy callables

function my_mixed($m) {
return (null === $m || is_string($m));
}
function f(is_scalar $s, is_int $i, my_mixed $m) {};

Not sure I understand 'truthy callables'. Your example is a definition of 'strict-only' union types. It's a planned feature.

THING 3: Pre- and post- blocks to define contracts, establish formal join
points, with or without hints from above

Contracts are fine but cannot completely replace type hints. I initially followed this road but reverted to thinking that both tools have their use. Contracts can implement performance-intensive checks but, because of this, cannot be executed in production. Type hints are performance-constrained but always executed. Another benefit of type hints is that PHP does not provide a consistent set of 'weak' is_xxx() functions yet. Checking for an integer in PHP, for instance, when you're ready to accept 31, 31.0, or '31', is relatively complex and unintuitive, while 'int $arg' is readable. If is_int() did a intuitive 'PHP way' check, contracts would be an alternative, but it is too counter-intuitive as it exists now. And BC makes changing is_xxx() behavior impossible, of course.

I had to get these off my chest. Forgive me their implementation ignorance.
I am yet tired and uncaffeinated.

No problem. Every goodwill is welcome.

To the void I commit their bodies...

Sorry, my english is too poor for that. Can you explain?

Regards

François

Bishop,

Pardon me for saying so, but I don't think you're on to a huge scoop here.
Scalar type hints - of all the kinds we've been talking about here, be them
strict, weak or coercive - can be easily emulated with a couple of lines of
code, we all know that. It's been known for years that strict hints are
is_*() else error equivalent, that weak hints are (mostly) just casts, etc.
Saying the class type hints can be emulated with is_a() calls wouldn't shock
anybody either.

However, the conclusion you seem to draw from that, few language designers
would agree with. Adding first-class, reflectable, analyzable syntax - is
worlds apart from the situation today - regardless of what type of scalar
hinting we talk about. Once we introduce language-level syntax for this
pattern, usage will explode compared to what it is today. Explode, not
increase. Tools would be updated to support (and take advantage) of this
new data - in a way they can't practically do with userland 'emulation' of
this behavior. Best practices will be updated. Doc formats will be
changed. Etc etc.

Last, note that you only focused on userland functions. Things are a bit
different with internal functions. The different proposals on the table all
propose to change - in one way or another - the way internal functions
behave. This is much less intuitive to emulate using userland code, for
obvious reasons.

So no, I don't think we're making a bigger deal out of it than it is.
Scalar type hints are a huge deal, and the fact you can emulate them in
userland code today doesn't change that in any way.

Thanks,

Zeev

-----Original Message-----
From: bishop.bettini@gmail.com [mailto:bishop.bettini@gmail.com] On
Behalf Of Bishop Bettini
Sent: Wednesday, February 18, 2015 2:18 PM
To: PHP internals
Subject: [PHP-DEV] Nightmares on type hints, annotations, and aop...

After a spate of literal coding nightmares, I woke needing to commit these
thoughts to the list. These may have been raised before, I can't
remember,
these debate has been raging for so long.

THING 1: Hints are just is_* wrappers

function f(scalar $s, int $i) { }

effectively is sugar for:

function f($s, $i) {
if (! is_scalar($s)) throw new \InvalidArgumentException();
if (! is_int($i)) throw new \InvalidArgumentException(); }

THING 2: Hints are truthy callables

function my_mixed($m) {
return (null === $m || is_string($m)); } function f(is_scalar $s,
is_int $i,
my_mixed $m) {};

THING 3: Pre- and post- blocks to define contracts, establish formal join
points, with or without hints from above

function f($s, $i, $m)
before {
if (! is_scalar($s)) throw new \InvalidArgumentException; }, inside {
$fp = fopen($s, 'r');
return [ $i, $m ];
},
after ($retval) {
assert(is_array($retval) && 2 === count($retval)); }, finally
($retval) {
fclose($fp);
if ($retval instanceof \Exception) {
// I got here by an unhandled exception
}
}

weave('before', 'f', function ($s, $i, $m) {
syslog(LOG_DEBUG, POINTCUT); // is 'f'
}

I had to get these off my chest. Forgive me their implementation
ignorance.
I am yet tired and uncaffeinated. To the void I commit their bodies...

bishop

Hi Bishop,

THING 2: Hints are truthy callables

function my_mixed($m) {
return (null === $m || is_string($m));
}
function f(is_scalar $s, is_int $i, my_mixed $m) {};

This is interesting idea. I'm sure it will be much slower than type hints
as it
calls functions, but it's still interesting idea. i.e. Cost of calling
functions are
not cheap. Inline optimization will help, but it increases byte code size.

Type hint implementation is a lot simpler than this.

THING 3: Pre- and post- blocks to define contracts, establish formal join

points, with or without hints from above

function f($s, $i, $m)
before {
if (! is_scalar($s)) throw new \InvalidArgumentException;
},
inside {
$fp = fopen($s, 'r');
return [ $i, $m ];
},
after ($retval) {
assert(is_array($retval) && 2 === count($retval));
},
finally ($retval) {
fclose($fp);
if ($retval instanceof \Exception) {
// I got here by an unhandled exception
}
}

weave('before', 'f', function ($s, $i, $m) {
syslog(LOG_DEBUG, POINTCUT); // is 'f'
}

Although it has similarities, AOP and DbC cannot share blocks/etc. AOP is
for production codes, DbC is not for production codes, but only for
development
and testing. These are 2 distinct features.

Regards,

--
Yasuo Ohgaki
yohgaki@ohgaki.net