ircmaxell · July 25, 2014 21:35
diff --git a/gistfile1.txt b/gistfile1.txt
 13:11 <ajf> I’m mostly happy with it as-is, but given that I made int/float/numeric/string not accept NULL, I think booleans shouldn’t be accepted for them either.
 13:19  → arty joined  ⇐ DaveRand quit  
 13:49  → sonotos and lonnylot joined  ⇐ dmitry and RemiFedor quit  
 14:41 <ircmaxell> ajf: that sounds sane (consistency wise). So it means that `int/float/numeric` only accept int/float/valid-number-string, and string only accepts string/int/float
 14:43 <ajf> ircmaxell: Yeah
 14:43 <ajf> Also, we don’t accept int/float/string for bool, so we shouldn’t accept bool for int/float/string
 14:43 <ircmaxell> yeah, makes perfect sense
 14:44 <ajf> Another point someone raised was that FALSE is a common error value alongside NULL.
 14:44 <ajf> So, should I go ahead and amend the patch and RFC to do that?
 14:44 <ircmaxell> I think that's the direction we're going, so that's valid
 14:44 <ircmaxell> ++
 14:45 <ajf> :)
 14:54 ⇐ arty quit ([email protected]) Quit: My MacBook Pro has gone to sleep. ZZZzzz…
 15:00 <ajf> And done.
 15:00 <ajf> Oh wait, oops
 15:00 <ajf> OK, done
 15:01 <ajf> ircmaxell: https://wiki.php.net/rfc/scalar_type_hinting_with_cast#conversion_rules
 15:02 <ircmaxell> 12.5 -> int... I'm not 100% sold on that behavior...
 15:03 <ajf> ircmaxell: Can you think of any instances where you might end up with 12.5 yet need it for an integer parameter?
 15:03 <ircmaxell> division
 15:03 <ircmaxell> foo($a / 2) where foo expects an integer
 15:04 <ajf> ircmaxell: Cast explicitly or use intdiv()
 15:04 <ajf> Granted, we haven’t added that yet ;)
 15:04 <ircmaxell> well, you get what I mean though...
 15:05 <ircmaxell> when would it obviously be an error to pass a float to something expecting an integer
 15:05 <ajf> ircmaxell: It’s a bit of a minefield too because there are several approaches to conversion
 15:05 <ajf> We could round or we could truncate
 15:05 <ajf> Which does the user expect?
 15:05 <ircmaxell> well, C always truncates, and I think that's the only valid response if we do accept it...
 15:06 <ajf> ircmaxell: But then, what if I’ve ended up with 1.99999999999999999997 somehow?
 15:06 <ircmaxell> very true
 15:06 <ajf> I think in the cases where you might end up with a float, you should explicitly convert it to int yourself.
 15:07 <ircmaxell> well, is there a good argument for not though?
 15:07 <ircmaxell> meaning what cases would it cause serious issues?
 15:08 <ajf> Hmm
 15:08 <ircmaxell> because again, I think accepting "12.5" as int is a LOT safer than accepting "Apple". And saying "just cast" could introduce far more significant bugs... It's a tradeoff...
 15:09 <ajf> True, if we make people cast, they lose the overflow protection
 15:09 <ajf> Actually, PHP’s integer-float casting semantics are awful. It doesn’t handle NaN or Infinity right.
 15:09 <ircmaxell> can it be cleanded up?
 15:10 <ajf> It’s something my bigints RFC includes as a bit of a bonus, actually
 15:11 <ajf> ircmaxell: Thing is, I don’t think you’re going to end up doing foo_takes_int(12.5) if your code is well-written
 15:11 <ircmaxell> well, remember we're not targetting well-written code. We're targetting generic code
 15:11 <ircmaxell> the average PHP user doesn't produce well-written-code from a type perspective. They expect it to "just work".
 15:12 <ircmaxell> and that's something we need to balance. Making it "just work" where possible, and error where necessary. 
 15:12 <ajf> Right
 15:13 <ajf> Division is the obvious case where you might end up with a float, but if you want integer division, you should be casting or rounding, or using intdiv() if that gets in.
 15:13 <ircmaxell> and I'm not sure if 12.5 -> int conversion erroring is necessary enough to justify the error...
 15:13 <ircmaxell> ajf: the average user doesn't know what integer division is
 15:13 <ircmaxell> substr($foo, 0, strlen($foo) / 2) <-- 
 15:14 <ajf> ircmaxell: Maybe not, but they probably do know about floor(), ceil() and round(), *especially* if they’ve ever touched JavaScript, which I’d say is quite likely
 15:15 <ircmaxell> that works 100% today, and if we're talking about re-normalizing ZPP to these rules down the road, that's a problem...
 15:15 <bwoebi> Tyrael: I thought the bugs were fixed now?
 15:17 <ajf> ircmaxell: I wonder if allowing 12.5 might cause confusion. If the function just eats it without complaint, maybe someone might think the function actually accepts float parameters, and gets surprised later when they discover it doesn’t
 15:18 <ircmaxell> I don't know if that's true, as people know they can pass a float to `substr`, but that doesn't mean they think it accepts floats
 15:18 <ircmaxell> I mean it may cause some slight confusion, but nothing that isn't there today
 15:18 <ajf> ircmaxell: Do they actually know that? I’m not actually sure how well zpp’s behaviour is documented
 15:19 <bwoebi> I’m not sure if we even should make a difference between integer and float in the typehint?
 15:19 <ajf> bwoebi: ?
 15:19 <ircmaxell> bwoebi: that's what I'm thinking. Both should accept the same values, and what comes out on the other side is either an int or float (depending on what you asked for)
 15:19 <bwoebi> ircmaxell: this.
 15:20 <ircmaxell> so int/float/numeric both accept int/float/numeric-string, and return int/float/best-fit
 15:20 <ajf> Hmm
 15:21 <ajf> On the other hand, might that mask bugs? Bear in mind that not everything does treat floats and ints th
 15:21 <ajf> e same
 15:21 <ircmaxell> so really, it would look like: if (!is_numeric($foo)) { error("Expecting {type}, got $foo") } $foo = (type) $foo;
 15:21 <ircmaxell> ajf it may mask bugs, but people casting to int may mask more (as it's an explicit cast)
 15:22 <ajf> True
 15:22 <@Tyrael> bwoebi: nevermind, I've somehow missed https://github.com/php/php-src/commit/c49a06168ed3759779452e2b2a44da22e75a0702 and https://github.com/php/php-src/commit/d909b6330e314bb434350317fda5bec185ea12c0
 15:22 <bwoebi> good
 15:23 → arty joined ([email protected])
 15:24 <@Tyrael> on the scalar type coersion, I start to think that we should first unify the userland casting rules and the zpp arg parsing stuff (obviously in a major) to be able to introduce it
 15:24 <ircmaxell> ajf: also, hold off on sending a mail for every revision... that thread is already way too long to live
 15:25 <@Tyrael> currently whatever we pick, it will be inconsistent in some way or another with other parts
 15:25 <ircmaxell> Tyrael: I think the approach we were looking at, is come up with a sane set of consistent rules that make sense, get them in for userland, then refactor ZPP to match
 15:25 <ajf> ircmaxell: I’d rather keep internals informed, though :/
 15:26 <ircmaxell> ajf: yes, keep them informed by sending a mail with overall changes, in a new thread once the concept is in place
 15:26 <bwoebi> ajf: I agree with ircmaxell, you should now first work up every issue and then make a new thread, otherwise nobody will read it.
 15:26 <ircmaxell> What we're doing at this point is tweaking the philosophy for dealing with these changes, so piecemeal decisions (like disallowing bool for numerics) doesn't really communicate that
 15:27 <ircmaxell> instead, once the RFC is consistent and presents the clean story, share that story back (based off of feedback from the prior discussion)
 15:27 <ajf> Ah, I see what you’re saying
 15:28 <ircmaxell> Tyrael: the reason for userland-first, is that it avoids the bias of massive BC concerns, and let's us discuss what should happen in an ideal way, and then apply that ideal to zpp. The other way around *may* be much more difficult... Disagree? Or...?
 15:29 <@Tyrael> ircmaxell: didn't thought of it that way
 15:29 <@Tyrael> makes sense, but it changes the premise of the original rfc
 15:29 <@Tyrael> which was to copy/follow the otherwise known rules of typejuggling in php
 15:29 <ajf> Yeah, the introduction doesn’t make sense now
 15:29 <ircmaxell> well, copy/follow where it makes sense
 15:30 <ircmaxell> which is one reason for the numeric discussion (allowing 12.5 to an int typehint by truncating)
 15:30 <@Tyrael> instead of that, we would introduce a new set of rules to cast scalars which can be more sensible
 15:30 <ircmaxell> yeah, that may be better for the overall story
 15:30 <ajf> Part of the problem with type juggling is that “scalar” is a fairly broad descriptor
 15:31 <@Tyrael> but basically we would expect people to memorize yet another conversion table
 15:31 <@Tyrael> because $foo = (scalar_type)$foo; would still be there
 15:31 <ajf> Unlike with ==, these rules only cover a small set of useful cases, so I think they’re easier and more intuitive to learn
 15:32 <ircmaxell> Well, I think there should be 2 sets of rules (and only 2). Implicit (where we don't know intent) and Explicit (where we do)
 15:32 <ircmaxell> and none of the ZPP rules are documented... So the only people that *really* know them are internals developers (and those devs who know how to read the ZPP code)
 15:32 <@Tyrael> ircmaxell: as far as I can tell we already have 2 in a sense
 15:33 <@Tyrael> implicit cast array to string will produce a notice for example
 15:33 <ajf> Well… we have more than 2 sets
 15:33 <ajf> Numeric strings aren’t consistently interpreted.
 15:33 <@Tyrael> but the scalar typehints would be different from the currently existing implicit and explicit casting rules
 15:33 <ircmaxell> currently, we have 4 sets of rules. Explicit, ZPP, ==, and +
 15:34 <ajf> Then there are array indices
 15:34 <ajf> and functions which do it themselves
 15:34 <@Tyrael> yeah, so introducing yet another similar but still different ruleset
 15:34 <@Tyrael> which cannot be introduced with one short sentence
 15:35 <ircmaxell> Tyrael: would you feel better if ZPP was refactored to match within the RFC?
 15:35 <ircmaxell> as opposed to doing it in a later RFC?
 15:35 <@Tyrael> I think we can't do that without merging some of those 4 rulesets
 15:35 <@Tyrael> ircmaxell: yeah, this is what I was trying to say with "on the scalar type coersion, I start to think that we should first unify the userland casting rules and the zpp arg parsing stuff (obviously in a major) to be able to introduce it"
 15:35 <ajf> More than 4. There are several casting rulesets.
 15:36 <@Tyrael> but this is just my 2 cents
 15:36 <@Tyrael> I just don't think we can have consistency without following exactly one of the already existing rulesets
 15:36 <ircmaxell> well, I don't think Explicit and == can go away or change... + becomes a challenge, because context matters, so I don't think that'll go away. And I don't think ZPP should be merged into any of the 3
 15:37 <ajf> ZPP’s always been different to userland functions, which I don’t like, but it’s the reality of things
 15:37 <ircmaxell> Tyrael: if done correctly (forget what that is for a second), do you think you could support changing the ZPP rules to be more consistent?
 15:37 <@Tyrael> and I can see why do you think that there is neither of the current ones are suitable for an explicit typehint
 15:37 <ajf> Userland functions don’t enforce the number of parameters, userland functions error instead of returning NULL, etc.
 15:37 <@Tyrael> ircmaxell: in a major I think I could
 15:37 <@Tyrael> I mean we have more control over that
 15:38 <@Tyrael> it could cause less BC impact for userland
 15:38 <ircmaxell> yeah, talking major only at this point (since changing zpp is way out of the question for a minor)
 15:38 <@Tyrael> as extensions could chose to keep the old behavior
 15:38 <@Tyrael> via adjusting their zpp usage
 15:39 <ircmaxell> well, if we change the internal rules ZPP uses, they'd need to re-implement ZPP to match the old way (or partially at least)?
 15:39 <ajf> Just add more characters! /s
 15:39 <ircmaxell> and there's nothing stopping avoiding using ZPP (or with "z"), just like there's no avoiding not hinting (foo($int) { if (!is_int($int)) {...
 15:40 <ircmaxell> hmm...
 15:41 <ircmaxell> ajf: what do you think about changing ZPP in this RFC (or creating a new one, since it's a different tactic) to match the concept we've been working towards? Make it a "Scalar Typing" RFC rather than type hint (as it's both internal and external)?
 15:42 ⇐ arty quit ([email protected]) Quit: My MacBook Pro has gone to sleep. ZZZzzz…
 15:42 <ajf> I’m not sure
 15:42 <ajf> I wonder if fixing PHP’s inconsistent and varying type conversion rules is a lost cause
 15:43 <ircmaxell> it may be
 15:44 <@Tyrael> ajf: btw. I think Stas come to the same conclusion regarding the current ruleset as me
 15:44 <ircmaxell> but so far, in all the years I've been following internals, I can't remember any serious proposals to do so (there were a few that wanted to make everything purely strict), but nothing that tried to grasp the idealology of PHP...
 15:45 <ajf> The problem is PHP doesn’t have an ideology
 15:45 <ajf> I’m not sure anyone has exactly the same idea of what PHP is
 15:45 <ircmaxell> ajf: yes, it does (when it comes to typing at least). "12" **is** int(12)...
 15:46 <ajf> Except when it’s not
 15:46 <ajf> PHP is (sadly?) not Perl
 15:46 <ajf> In Perl you can’t easily distinguish “12” and 12 AFAIK. The same is hardly true of PHP, where they’re sometimes but not always equivalent
 15:46 <ircmaxell> there's only one case that I know of where "12" isn't replaceable for int(12) (and vise versa), and that's in ctype_digit()
 15:47 <ircmaxell> well, all the ctype apis...
 15:47 <ajf> andreas-air:php-src ajf$ sapi/cli/php -r 'var_dump(23 ^ 7); var_dump("23" ^ "7");'
 15:47 <ajf> int(16)
 15:47 <ajf> string(1) ""
 15:47 <ircmaxell> fair point...
 15:48 <ajf> Also, hexadecimal number strings are sometimes, but only sometimes, considered numbers
 15:48 <ajf> http://3v4l.org/GF9EI
 15:48 <ajf> Is NULL an empty array? Sometimes, maybe. Sometime it isn’t.
 15:48 <ircmaxell> well, that's fixable by making the string->int parsing rules consistent in all cases (I don't think there's justification for it changing, is there?)
 15:48 <ircmaxell> null an empty array?
 15:49 <ajf> $x = NULL; $x[2] = 3; var_dump($x); => array(1) {  [2]=>  int(3)}
 15:50 <ircmaxell> well, that's not really casting going on...
 15:50 <ircmaxell> it's transparent initialization, but I see the point
 15:52 <ajf> I’m not sure if the RFC is actually sane or not. In some respects I think its conversion rules are sane, but isn’t insane to have so many sets of conversion rules in PHP? Then again, isn’t it insane already? Is having a sane set and an insane set insane, or is it saner than just having an insane set? I’m not sure.
 15:53 <auroraeos> maybe we need a conversion rule rfc (slightly being a troll but serious as well) - another set will make the language even more ... schizophrenic
 15:53 <ircmaxell> which is why I'm wondering if doing ZPP at the same time as user-land hints, making them all sane conversions, would be the better way...
 15:54 <ajf> Yes, but then we just bikeshed about the right way to do userland hints.
 15:54 <ajf> er
 15:54 <ajf> ZPP hints
 15:55 <ircmaxell> Compromise isn't always a good thing, and sometimes let the bikeshed happen without getting sucked into it. If the rules are sane enough, the vote should be clear. If they are not, then they are not...
 15:55 <ajf> ircmaxell: Something I’d like to try is holding an online poll of sorts to see what people’s voting intentions are. I have no idea how people would vote atm.
 15:55 <ircmaxell> ajf: I don't think people really know what they would vote on
 15:56 <ircmaxell> The overall RFC has changed a number of times, and the discussion has been hard to follow
 15:59 <ajf> :/
 16:12 <ajf> I’m not sure where to go from here
 16:15 → druid joined ([email protected])
 16:15 <auroraeos> perhaps get it to the point where you're truly happy with it - and do a summary on list of the RFC in it's final form
 16:15 <auroraeos> then attempt a short discussion for minor things
 16:23 <ajf> I’m quite happy with it
 16:23 <ajf> I’m just not sure it could pass
 16:24 <bwoebi> quite. Yeah. But it still has its uncertainities…
 16:37 <ajf> ircmaxell: I wonder if it might be worthwhile just changing the RFC to do exactly what zpp does, no exceptions
 16:37 <ircmaxell> and then a later RFC to change the overall behavior?
 16:37 <ajf> Yeah.
 16:37 <ircmaxell> that's likely the most likely route to actually pass
 16:38 <ajf> RIght
 16:38 <ajf> Every little exception to and inconsistency with zpp is a problem
 16:38 <ajf> The best solution is to have absolutely none and exactly emulate - no, use the same code - as zpp
 16:38 <ajf> Obviously it’s an E_RECOVERABLE_ERROR and not a return NULL, but otherwise the same
 16:38 <ircmaxell> at this point, while I don't think it's good to ship those rules, I agree that's about the only way forward...
 16:39 <ajf> Right\
 16:39 <ircmaxell> so +1 from me
 16:39 <ajf> I don’t like zpp’s rules, but at least that would be consistent
 16:41 <ircmaxell> and it would cast...
 16:41 <ircmaxell> so it wouldn't be purely strict
 16:42 <ajf> Right
 16:43 <ajf> Actually, there’d need to be another difference from zpp: How do we handle nullability?
 16:43 <ircmaxell> how does ZPP handle it?
 16:43 <ajf> I’m not sure. Does it even have nullable types? *checks*
 16:43 <ircmaxell> by my original behavior
 16:44 <ircmaxell> if it's not nullable, you can never get null (null auto-casts), if it is nullable, then you can get null
 16:44 <ircmaxell> http://lxr.php.net/xref/PHP_5_2/Zend/zend_API.c#327
 16:44 <ircmaxell> at least depending on the type
 16:45 <bwoebi> 5.2?
 16:45 <ircmaxell> so it's only supported for strings
 16:45 <ircmaxell> numbers can't be nullable
 16:45 <ajf> PHP 5.2?!
 16:45 <ircmaxell> whoops: http://lxr.php.net/xref/PHP_TRUNK/Zend/zend_API.c#check_null
 16:46 <ircmaxell> I had referenced 5.2 yesterday, so search defaulted to 5.2 (even though I was browsing trunk)
 16:46 <ircmaxell> but the behavior is the same
 16:49 → cjones_ joined ([email protected])
 16:52 <ircmaxell> also: if you're going to do that (make scalar params follow zpp), I'd suggest explicitly targetting NEXT.0.0, as opposed to 5.NEXT...
 16:53 <ajf> hmm
 16:53 <bwoebi> agree. It’d be a too big BC break for 5.NEXT (at least if people consider engine_exceptions too big for 5.NEXT too)
 16:56 <ajf> There’d be no BC break
 16:57 <ajf> The problem is that you now have even more code that’ll break if you change zpp’s behaviour
 16:58 <ircmaxell> exactly
 16:58 <ircmaxell> if you release it in 5.NEXT, it basically prevents ZPP refactors down the road in NEXT
 17:00 <ajf> Welp.
 17:01  → EstaTiC joined  ↔ cjones_ and domeh nipped out  
 17:05 <ajf> ircmaxell: Should I just withdraw the RFC as it stands? Amending it to instead propose doing the ZPP thing in 6 might be bad.
 17:05 <ajf> Er, NEXT :)
 17:05 <ircmaxell> yeah, that's fair, it preserves history. 
 17:05 <ircmaxell> HOWEVER, I woulnd't officially withdraw it
 17:05 <ircmaxell> until you have the other one drafted
 17:05 <ajf> Ah, I see
 17:06 <ircmaxell> that way it isn't "we're giving up" but "Rather than do it that way, let's take a different approach"
 17:06 <auroraeos> +1
 17:06 <auroraeos> and make sure you link to the old from the new :)
 17:06 <cjones_> and vice versa
 17:07 <cjones_> And include the correct URL in ALL email about the RFC
 17:07 <ircmaxell> ++
 17:08 <ircmaxell> and title it so it's dead obvious "User Hints - ZPP Style" or something along those lines
 17:08 <ajf> Right, this all sounds workable
 17:08 <ajf> However I’m a little fed up of scalar type hints for today
 17:08 <ajf> Maybe later.
 17:09 <@bjori> parameter casting
 17:10 <ircmaxell> ajf: I think this was quite productive. Thank you for taking point on it, you've done an amazing job on it so far. :-)
 17:10 <ajf> Thanks ^^
 17:10 <ajf> bjori: internal-style scalar parameters?
	13:11 <ajf> I’m mostly happy with it as-is, but given that I made int/float/numeric/string not accept NULL, I think booleans shouldn’t be accepted for them either.
	13:19 → arty joined ⇐ DaveRand quit
	13:49 → sonotos and lonnylot joined ⇐ dmitry and RemiFedor quit
	14:41 <ircmaxell> ajf: that sounds sane (consistency wise). So it means that `int/float/numeric` only accept int/float/valid-number-string, and string only accepts string/int/float
	14:43 <ajf> ircmaxell: Yeah
	14:43 <ajf> Also, we don’t accept int/float/string for bool, so we shouldn’t accept bool for int/float/string
	14:43 <ircmaxell> yeah, makes perfect sense
	14:44 <ajf> Another point someone raised was that FALSE is a common error value alongside NULL.
	14:44 <ajf> So, should I go ahead and amend the patch and RFC to do that?
	14:44 <ircmaxell> I think that's the direction we're going, so that's valid
	14:44 <ircmaxell> ++
	14:45 <ajf> :)
	14:54 ⇐ arty quit ([email protected]) Quit: My MacBook Pro has gone to sleep. ZZZzzz…
	15:00 <ajf> And done.
	15:00 <ajf> Oh wait, oops
	15:00 <ajf> OK, done
	15:01 <ajf> ircmaxell: https://wiki.php.net/rfc/scalar_type_hinting_with_cast#conversion_rules
	15:02 <ircmaxell> 12.5 -> int... I'm not 100% sold on that behavior...
	15:03 <ajf> ircmaxell: Can you think of any instances where you might end up with 12.5 yet need it for an integer parameter?
	15:03 <ircmaxell> division
	15:03 <ircmaxell> foo($a / 2) where foo expects an integer
	15:04 <ajf> ircmaxell: Cast explicitly or use intdiv()
	15:04 <ajf> Granted, we haven’t added that yet ;)
	15:04 <ircmaxell> well, you get what I mean though...
	15:05 <ircmaxell> when would it obviously be an error to pass a float to something expecting an integer
	15:05 <ajf> ircmaxell: It’s a bit of a minefield too because there are several approaches to conversion
	15:05 <ajf> We could round or we could truncate
	15:05 <ajf> Which does the user expect?
	15:05 <ircmaxell> well, C always truncates, and I think that's the only valid response if we do accept it...
	15:06 <ajf> ircmaxell: But then, what if I’ve ended up with 1.99999999999999999997 somehow?
	15:06 <ircmaxell> very true
	15:06 <ajf> I think in the cases where you might end up with a float, you should explicitly convert it to int yourself.
	15:07 <ircmaxell> well, is there a good argument for not though?
	15:07 <ircmaxell> meaning what cases would it cause serious issues?
	15:08 <ajf> Hmm
	15:08 <ircmaxell> because again, I think accepting "12.5" as int is a LOT safer than accepting "Apple". And saying "just cast" could introduce far more significant bugs... It's a tradeoff...
	15:09 <ajf> True, if we make people cast, they lose the overflow protection
	15:09 <ajf> Actually, PHP’s integer-float casting semantics are awful. It doesn’t handle NaN or Infinity right.
	15:09 <ircmaxell> can it be cleanded up?
	15:10 <ajf> It’s something my bigints RFC includes as a bit of a bonus, actually
	15:11 <ajf> ircmaxell: Thing is, I don’t think you’re going to end up doing foo_takes_int(12.5) if your code is well-written
	15:11 <ircmaxell> well, remember we're not targetting well-written code. We're targetting generic code
	15:11 <ircmaxell> the average PHP user doesn't produce well-written-code from a type perspective. They expect it to "just work".
	15:12 <ircmaxell> and that's something we need to balance. Making it "just work" where possible, and error where necessary.
	15:12 <ajf> Right
	15:13 <ajf> Division is the obvious case where you might end up with a float, but if you want integer division, you should be casting or rounding, or using intdiv() if that gets in.
	15:13 <ircmaxell> and I'm not sure if 12.5 -> int conversion erroring is necessary enough to justify the error...
	15:13 <ircmaxell> ajf: the average user doesn't know what integer division is
	15:13 <ircmaxell> substr($foo, 0, strlen($foo) / 2) <--
	15:14 <ajf> ircmaxell: Maybe not, but they probably do know about floor(), ceil() and round(), especially if they’ve ever touched JavaScript, which I’d say is quite likely
	15:15 <ircmaxell> that works 100% today, and if we're talking about re-normalizing ZPP to these rules down the road, that's a problem...
	15:15 <bwoebi> Tyrael: I thought the bugs were fixed now?
	15:17 <ajf> ircmaxell: I wonder if allowing 12.5 might cause confusion. If the function just eats it without complaint, maybe someone might think the function actually accepts float parameters, and gets surprised later when they discover it doesn’t
	15:18 <ircmaxell> I don't know if that's true, as people know they can pass a float to `substr`, but that doesn't mean they think it accepts floats
	15:18 <ircmaxell> I mean it may cause some slight confusion, but nothing that isn't there today
	15:18 <ajf> ircmaxell: Do they actually know that? I’m not actually sure how well zpp’s behaviour is documented
	15:19 <bwoebi> I’m not sure if we even should make a difference between integer and float in the typehint?
	15:19 <ajf> bwoebi: ?
	15:19 <ircmaxell> bwoebi: that's what I'm thinking. Both should accept the same values, and what comes out on the other side is either an int or float (depending on what you asked for)
	15:19 <bwoebi> ircmaxell: this.
	15:20 <ircmaxell> so int/float/numeric both accept int/float/numeric-string, and return int/float/best-fit
	15:20 <ajf> Hmm
	15:21 <ajf> On the other hand, might that mask bugs? Bear in mind that not everything does treat floats and ints th
	15:21 <ajf> e same
	15:21 <ircmaxell> so really, it would look like: if (!is_numeric($foo)) { error("Expecting {type}, got $foo") } $foo = (type) $foo;
	15:21 <ircmaxell> ajf it may mask bugs, but people casting to int may mask more (as it's an explicit cast)
	15:22 <ajf> True
	15:22 <@Tyrael> bwoebi: nevermind, I've somehow missed https://github.com/php/php-src/commit/c49a06168ed3759779452e2b2a44da22e75a0702 and https://github.com/php/php-src/commit/d909b6330e314bb434350317fda5bec185ea12c0
	15:22 <bwoebi> good
	15:23 → arty joined ([email protected])
	15:24 <@Tyrael> on the scalar type coersion, I start to think that we should first unify the userland casting rules and the zpp arg parsing stuff (obviously in a major) to be able to introduce it
	15:24 <ircmaxell> ajf: also, hold off on sending a mail for every revision... that thread is already way too long to live
	15:25 <@Tyrael> currently whatever we pick, it will be inconsistent in some way or another with other parts
	15:25 <ircmaxell> Tyrael: I think the approach we were looking at, is come up with a sane set of consistent rules that make sense, get them in for userland, then refactor ZPP to match
	15:25 <ajf> ircmaxell: I’d rather keep internals informed, though :/
	15:26 <ircmaxell> ajf: yes, keep them informed by sending a mail with overall changes, in a new thread once the concept is in place
	15:26 <bwoebi> ajf: I agree with ircmaxell, you should now first work up every issue and then make a new thread, otherwise nobody will read it.
	15:26 <ircmaxell> What we're doing at this point is tweaking the philosophy for dealing with these changes, so piecemeal decisions (like disallowing bool for numerics) doesn't really communicate that
	15:27 <ircmaxell> instead, once the RFC is consistent and presents the clean story, share that story back (based off of feedback from the prior discussion)
	15:27 <ajf> Ah, I see what you’re saying
	15:28 <ircmaxell> Tyrael: the reason for userland-first, is that it avoids the bias of massive BC concerns, and let's us discuss what should happen in an ideal way, and then apply that ideal to zpp. The other way around may be much more difficult... Disagree? Or...?
	15:29 <@Tyrael> ircmaxell: didn't thought of it that way
	15:29 <@Tyrael> makes sense, but it changes the premise of the original rfc
	15:29 <@Tyrael> which was to copy/follow the otherwise known rules of typejuggling in php
	15:29 <ajf> Yeah, the introduction doesn’t make sense now
	15:29 <ircmaxell> well, copy/follow where it makes sense
	15:30 <ircmaxell> which is one reason for the numeric discussion (allowing 12.5 to an int typehint by truncating)
	15:30 <@Tyrael> instead of that, we would introduce a new set of rules to cast scalars which can be more sensible
	15:30 <ircmaxell> yeah, that may be better for the overall story
	15:30 <ajf> Part of the problem with type juggling is that “scalar” is a fairly broad descriptor
	15:31 <@Tyrael> but basically we would expect people to memorize yet another conversion table
	15:31 <@Tyrael> because $foo = (scalar_type)$foo; would still be there
	15:31 <ajf> Unlike with ==, these rules only cover a small set of useful cases, so I think they’re easier and more intuitive to learn
	15:32 <ircmaxell> Well, I think there should be 2 sets of rules (and only 2). Implicit (where we don't know intent) and Explicit (where we do)
	15:32 <ircmaxell> and none of the ZPP rules are documented... So the only people that really know them are internals developers (and those devs who know how to read the ZPP code)
	15:32 <@Tyrael> ircmaxell: as far as I can tell we already have 2 in a sense
	15:33 <@Tyrael> implicit cast array to string will produce a notice for example
	15:33 <ajf> Well… we have more than 2 sets
	15:33 <ajf> Numeric strings aren’t consistently interpreted.
	15:33 <@Tyrael> but the scalar typehints would be different from the currently existing implicit and explicit casting rules
	15:33 <ircmaxell> currently, we have 4 sets of rules. Explicit, ZPP, ==, and +
	15:34 <ajf> Then there are array indices
	15:34 <ajf> and functions which do it themselves
	15:34 <@Tyrael> yeah, so introducing yet another similar but still different ruleset
	15:34 <@Tyrael> which cannot be introduced with one short sentence
	15:35 <ircmaxell> Tyrael: would you feel better if ZPP was refactored to match within the RFC?
	15:35 <ircmaxell> as opposed to doing it in a later RFC?
	15:35 <@Tyrael> I think we can't do that without merging some of those 4 rulesets
	15:35 <@Tyrael> ircmaxell: yeah, this is what I was trying to say with "on the scalar type coersion, I start to think that we should first unify the userland casting rules and the zpp arg parsing stuff (obviously in a major) to be able to introduce it"
	15:35 <ajf> More than 4. There are several casting rulesets.
	15:36 <@Tyrael> but this is just my 2 cents
	15:36 <@Tyrael> I just don't think we can have consistency without following exactly one of the already existing rulesets
	15:36 <ircmaxell> well, I don't think Explicit and == can go away or change... + becomes a challenge, because context matters, so I don't think that'll go away. And I don't think ZPP should be merged into any of the 3
	15:37 <ajf> ZPP’s always been different to userland functions, which I don’t like, but it’s the reality of things
	15:37 <ircmaxell> Tyrael: if done correctly (forget what that is for a second), do you think you could support changing the ZPP rules to be more consistent?
	15:37 <@Tyrael> and I can see why do you think that there is neither of the current ones are suitable for an explicit typehint
	15:37 <ajf> Userland functions don’t enforce the number of parameters, userland functions error instead of returning NULL, etc.
	15:37 <@Tyrael> ircmaxell: in a major I think I could
	15:37 <@Tyrael> I mean we have more control over that
	15:38 <@Tyrael> it could cause less BC impact for userland
	15:38 <ircmaxell> yeah, talking major only at this point (since changing zpp is way out of the question for a minor)
	15:38 <@Tyrael> as extensions could chose to keep the old behavior
	15:38 <@Tyrael> via adjusting their zpp usage
	15:39 <ircmaxell> well, if we change the internal rules ZPP uses, they'd need to re-implement ZPP to match the old way (or partially at least)?
	15:39 <ajf> Just add more characters! /s
	15:39 <ircmaxell> and there's nothing stopping avoiding using ZPP (or with "z"), just like there's no avoiding not hinting (foo($int) { if (!is_int($int)) {...
	15:40 <ircmaxell> hmm...
	15:41 <ircmaxell> ajf: what do you think about changing ZPP in this RFC (or creating a new one, since it's a different tactic) to match the concept we've been working towards? Make it a "Scalar Typing" RFC rather than type hint (as it's both internal and external)?
	15:42 ⇐ arty quit ([email protected]) Quit: My MacBook Pro has gone to sleep. ZZZzzz…
	15:42 <ajf> I’m not sure
	15:42 <ajf> I wonder if fixing PHP’s inconsistent and varying type conversion rules is a lost cause
	15:43 <ircmaxell> it may be
	15:44 <@Tyrael> ajf: btw. I think Stas come to the same conclusion regarding the current ruleset as me
	15:44 <ircmaxell> but so far, in all the years I've been following internals, I can't remember any serious proposals to do so (there were a few that wanted to make everything purely strict), but nothing that tried to grasp the idealology of PHP...
	15:45 <ajf> The problem is PHP doesn’t have an ideology
	15:45 <ajf> I’m not sure anyone has exactly the same idea of what PHP is
	15:45 <ircmaxell> ajf: yes, it does (when it comes to typing at least). "12" is int(12)...
	15:46 <ajf> Except when it’s not
	15:46 <ajf> PHP is (sadly?) not Perl
	15:46 <ajf> In Perl you can’t easily distinguish “12” and 12 AFAIK. The same is hardly true of PHP, where they’re sometimes but not always equivalent
	15:46 <ircmaxell> there's only one case that I know of where "12" isn't replaceable for int(12) (and vise versa), and that's in ctype_digit()
	15:47 <ircmaxell> well, all the ctype apis...
	15:47 <ajf> andreas-air:php-src ajf$ sapi/cli/php -r 'var_dump(23 ^ 7); var_dump("23" ^ "7");'
	15:47 <ajf> int(16)
	15:47 <ajf> string(1) ""
	15:47 <ircmaxell> fair point...
	15:48 <ajf> Also, hexadecimal number strings are sometimes, but only sometimes, considered numbers
	15:48 <ajf> http://3v4l.org/GF9EI
	15:48 <ajf> Is NULL an empty array? Sometimes, maybe. Sometime it isn’t.
	15:48 <ircmaxell> well, that's fixable by making the string->int parsing rules consistent in all cases (I don't think there's justification for it changing, is there?)
	15:48 <ircmaxell> null an empty array?
	15:49 <ajf> $x = NULL; $x[2] = 3; var_dump($x); => array(1) { [2]=> int(3)}
	15:50 <ircmaxell> well, that's not really casting going on...
	15:50 <ircmaxell> it's transparent initialization, but I see the point
	15:52 <ajf> I’m not sure if the RFC is actually sane or not. In some respects I think its conversion rules are sane, but isn’t insane to have so many sets of conversion rules in PHP? Then again, isn’t it insane already? Is having a sane set and an insane set insane, or is it saner than just having an insane set? I’m not sure.
	15:53 <auroraeos> maybe we need a conversion rule rfc (slightly being a troll but serious as well) - another set will make the language even more ... schizophrenic
	15:53 <ircmaxell> which is why I'm wondering if doing ZPP at the same time as user-land hints, making them all sane conversions, would be the better way...
	15:54 <ajf> Yes, but then we just bikeshed about the right way to do userland hints.
	15:54 <ajf> er
	15:54 <ajf> ZPP hints
	15:55 <ircmaxell> Compromise isn't always a good thing, and sometimes let the bikeshed happen without getting sucked into it. If the rules are sane enough, the vote should be clear. If they are not, then they are not...
	15:55 <ajf> ircmaxell: Something I’d like to try is holding an online poll of sorts to see what people’s voting intentions are. I have no idea how people would vote atm.
	15:55 <ircmaxell> ajf: I don't think people really know what they would vote on
	15:56 <ircmaxell> The overall RFC has changed a number of times, and the discussion has been hard to follow
	15:59 <ajf> :/
	16:12 <ajf> I’m not sure where to go from here
	16:15 → druid joined ([email protected])
	16:15 <auroraeos> perhaps get it to the point where you're truly happy with it - and do a summary on list of the RFC in it's final form
	16:15 <auroraeos> then attempt a short discussion for minor things
	16:23 <ajf> I’m quite happy with it
	16:23 <ajf> I’m just not sure it could pass
	16:24 <bwoebi> quite. Yeah. But it still has its uncertainities…
	16:37 <ajf> ircmaxell: I wonder if it might be worthwhile just changing the RFC to do exactly what zpp does, no exceptions
	16:37 <ircmaxell> and then a later RFC to change the overall behavior?
	16:37 <ajf> Yeah.
	16:37 <ircmaxell> that's likely the most likely route to actually pass
	16:38 <ajf> RIght
	16:38 <ajf> Every little exception to and inconsistency with zpp is a problem
	16:38 <ajf> The best solution is to have absolutely none and exactly emulate - no, use the same code - as zpp
	16:38 <ajf> Obviously it’s an E_RECOVERABLE_ERROR and not a return NULL, but otherwise the same
	16:38 <ircmaxell> at this point, while I don't think it's good to ship those rules, I agree that's about the only way forward...
	16:39 <ajf> Right\
	16:39 <ircmaxell> so +1 from me
	16:39 <ajf> I don’t like zpp’s rules, but at least that would be consistent
	16:41 <ircmaxell> and it would cast...
	16:41 <ircmaxell> so it wouldn't be purely strict
	16:42 <ajf> Right
	16:43 <ajf> Actually, there’d need to be another difference from zpp: How do we handle nullability?
	16:43 <ircmaxell> how does ZPP handle it?
	16:43 <ajf> I’m not sure. Does it even have nullable types? checks
	16:43 <ircmaxell> by my original behavior
	16:44 <ircmaxell> if it's not nullable, you can never get null (null auto-casts), if it is nullable, then you can get null
	16:44 <ircmaxell> http://lxr.php.net/xref/PHP_5_2/Zend/zend_API.c#327
	16:44 <ircmaxell> at least depending on the type
	16:45 <bwoebi> 5.2?
	16:45 <ircmaxell> so it's only supported for strings
	16:45 <ircmaxell> numbers can't be nullable
	16:45 <ajf> PHP 5.2?!
	16:45 <ircmaxell> whoops: http://lxr.php.net/xref/PHP_TRUNK/Zend/zend_API.c#check_null
	16:46 <ircmaxell> I had referenced 5.2 yesterday, so search defaulted to 5.2 (even though I was browsing trunk)
	16:46 <ircmaxell> but the behavior is the same
	16:49 → cjones_ joined ([email protected])
	16:52 <ircmaxell> also: if you're going to do that (make scalar params follow zpp), I'd suggest explicitly targetting NEXT.0.0, as opposed to 5.NEXT...
	16:53 <ajf> hmm
	16:53 <bwoebi> agree. It’d be a too big BC break for 5.NEXT (at least if people consider engine_exceptions too big for 5.NEXT too)
	16:56 <ajf> There’d be no BC break
	16:57 <ajf> The problem is that you now have even more code that’ll break if you change zpp’s behaviour
	16:58 <ircmaxell> exactly
	16:58 <ircmaxell> if you release it in 5.NEXT, it basically prevents ZPP refactors down the road in NEXT
	17:00 <ajf> Welp.
	17:01 → EstaTiC joined ↔ cjones_ and domeh nipped out
	17:05 <ajf> ircmaxell: Should I just withdraw the RFC as it stands? Amending it to instead propose doing the ZPP thing in 6 might be bad.
	17:05 <ajf> Er, NEXT :)
	17:05 <ircmaxell> yeah, that's fair, it preserves history.
	17:05 <ircmaxell> HOWEVER, I woulnd't officially withdraw it
	17:05 <ircmaxell> until you have the other one drafted
	17:05 <ajf> Ah, I see
	17:06 <ircmaxell> that way it isn't "we're giving up" but "Rather than do it that way, let's take a different approach"
	17:06 <auroraeos> +1
	17:06 <auroraeos> and make sure you link to the old from the new :)
	17:06 <cjones_> and vice versa
	17:07 <cjones_> And include the correct URL in ALL email about the RFC
	17:07 <ircmaxell> ++
	17:08 <ircmaxell> and title it so it's dead obvious "User Hints - ZPP Style" or something along those lines
	17:08 <ajf> Right, this all sounds workable
	17:08 <ajf> However I’m a little fed up of scalar type hints for today
	17:08 <ajf> Maybe later.
	17:09 <@bjori> parameter casting
	17:10 <ircmaxell> ajf: I think this was quite productive. Thank you for taking point on it, you've done an amazing job on it so far. :-)
	17:10 <ajf> Thanks ^^
	17:10 <ajf> bjori: internal-style scalar parameters?