www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Constrained Templates

reply Walter Bright <newshound1 digitalmars.com> writes:
http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html

Anyone want to do the honors and post to reddit, ycombinator, etc. ?
Jun 13 2010
next sibling parent reply "Yao G." <nospamyao gmail.com> writes:
http://www.reddit.com/r/programming/comments/cel4o/constrained_templates_in_the_d_programming/

On Sun, 13 Jun 2010 14:01:16 -0500, Walter Bright  
<newshound1 digitalmars.com> wrote:


 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html

 Anyone want to do the honors and post to reddit, ycombinator, etc. ?



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Jun 13 2010
parent reply Walter Bright <newshound1 digitalmars.com> writes:
Yao G. wrote:

 http://www.reddit.com/r/programming/comments/cel4o/constrained_templates_i
_the_d_programming/ 


Thanks!

Anyone know why this doesn't appear on the "new" page?

http://www.reddit.com/r/programming/new/
Jun 13 2010
parent reply "Yao G." <nospamyao gmail.com> writes:
Maybe I did something wrong at the moment I submitted the article.  :(

Somebody else should try to submit it too.

On Sun, 13 Jun 2010 15:15:25 -0500, Walter Bright  
<newshound1 digitalmars.com> wrote:


 Yao G. wrote:

 http://www.reddit.com/r/programming/comments/cel4o/constrained_templates_in_the_d_programming/


 Thanks!

 Anyone know why this doesn't appear on the "new" page?

 http://www.reddit.com/r/programming/new/



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Jun 13 2010
parent reply Walter Bright <newshound1 digitalmars.com> writes:
Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


I sent an email to the reddit moderators asking what's up with that.
Jun 13 2010
next sibling parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Walter Bright wrote:

 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 
 I sent an email to the reddit moderators asking what's up with that.


I posted this as a test:

http://www.reddit.com/r/reddit.com/comments/cemtc/tech_talk_tips/

Andrei
Jun 13 2010
parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Andrei Alexandrescu wrote:

 Walter Bright wrote:

 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 I sent an email to the reddit moderators asking what's up with that.


 
 I posted this as a test:
 
 http://www.reddit.com/r/reddit.com/comments/cemtc/tech_talk_tips/
 
 Andrei


Sorry I was on the plane and in a hurry so I posted to general reddit, 
not programming. I now deleted that (sorry BCS) and now reposted under 
programming. Vote me up please :o).

http://www.reddit.com/r/programming/comments/ceo9s/tech_talk_tips/

The point of the experiment was that Walter had hypothesized it's 
possible his name is on a scrutiny/spam list. I removed his name from 
the article and the results are mixed: when I posted in the general 
reddit community the article didn't make it on the new articles list, 
however this latest post in /programming/ was listed instantly. In the 
meantime I saw that Walter's constrained templates article appeared but 
with a delay. Go figure.


Andrei
Jun 13 2010
prev sibling parent reply "Yao G." <nospamyao gmail.com> writes:
It is now displayed in the front page.

http://www.reddit.com/r/programming/

On Sun, 13 Jun 2010 15:42:13 -0500, Walter Bright  
<newshound1 digitalmars.com> wrote:


 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 I sent an email to the reddit moderators asking what's up with that.



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Jun 13 2010
parent reply Walter Bright <newshound1 digitalmars.com> writes:
Yao G. wrote:

 It is now displayed in the front page.
 
 http://www.reddit.com/r/programming/
 
 On Sun, 13 Jun 2010 15:42:13 -0500, Walter Bright 
 <newshound1 digitalmars.com> wrote:
 

 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 I sent an email to the reddit moderators asking what's up with that.


 
 


Here's the email I received from the moderators: "Fixed, it was autobanned." It 
didn't say what rule was used to autoban it.
Jun 13 2010
next sibling parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Walter Bright wrote:

 Yao G. wrote:

 It is now displayed in the front page.

 http://www.reddit.com/r/programming/

 On Sun, 13 Jun 2010 15:42:13 -0500, Walter Bright 
 <newshound1 digitalmars.com> wrote:


 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 I sent an email to the reddit moderators asking what's up with that.




 
 Here's the email I received from the moderators: "Fixed, it was 
 autobanned." It didn't say what rule was used to autoban it.


This is odd. Now my post on tech talk tips has vanished from the new 
articles list.

Andrei
Jun 13 2010
prev sibling parent "Yao G." <nospamyao gmail.com> writes:
:D That's weird. I just filled some typical form controls and clicked  
submit. No "autoban" button was clicked.

Yao G.

On Sun, 13 Jun 2010 20:05:04 -0500, Walter Bright  
<newshound1 digitalmars.com> wrote:


 Yao G. wrote:

 It is now displayed in the front page.
  http://www.reddit.com/r/programming/
  On Sun, 13 Jun 2010 15:42:13 -0500, Walter Bright  
 <newshound1 digitalmars.com> wrote:


 Yao G. wrote:

 Maybe I did something wrong at the moment I submitted the article.  :(


 I sent an email to the reddit moderators asking what's up with that.




 Here's the email I received from the moderators: "Fixed, it was  
 autobanned." It didn't say what rule was used to autoban it.



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Jun 14 2010
prev sibling next sibling parent reply Leandro Lucarella <llucax gmail.com> writes:
Walter Bright, el 13 de junio a las 12:01 me escribiste:

 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html
 
 Anyone want to do the honors and post to reddit, ycombinator, etc. ?


Nice article, but when I read:

	T gcd(T)(T a, T b)
		if (is(typeof(a % b)))
	{
		...
	}

Under the presence of such beauty as template constraint syntax is,
"is(typeof(a % b))" makes my eyes hurt, and my brain wonder. When you
get used to this idiom, it might not look to bad, but I'm seeing people
new to D wonder "why in the hell was that syntax used?" while they try
to decode what is(typeof()) means.

I think is really a shame, all the beauty and clarity gained by the
simple "if" of template constraints, is lost with the cryptic
is(typeof()) (well, maybe not all, but a significant part).

I would love to see this fixed/simplified. For example, by applying the
changes in suggested in bug 3702 [1]. This looks much better, and is
pretty clear to anyone (even people that doesn't know D):

	T gcd(T)(T a, T b)
		if (meta.compiles(a % b))
	{
		...
	}

I guess at this point this is not going to happen for D2, a real shame
:S

The, the article made me think that, even when template constraints are
useful not only for improving error reporting, they are often used only
for that. Since the compiler knows all the operations a template
parameter will need to satisfy from the function body, it would be nice
to have some sort of way to tell the compiler to write the template
constraints for us (the obvious ones at least, there might be other
template constraints desired besides the ones the ones the compiler can
figure out). This way, the errors can be improved without user
intervention.

[1] http://d.puremagic.com/issues/show_bug.cgi?id=3702

-- 
Leandro Lucarella (AKA luca)                     http://llucax.com.ar/
----------------------------------------------------------------------
GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145  104C 949E BFB6 5F5A 8D05)
----------------------------------------------------------------------
Salvajes, de traje, me quieren enseñar
Salvajes, de traje, me quieren educar
Jun 13 2010
next sibling parent "Yao G." <nospamyao gmail.com> writes:
Andrei used that idiom quite extensive in Phobos, unfortunately, I dare to  
say. It makes the code a lot harder to read. But the alternative, using  
__traits(compiles, ... ) is even worse IMO, because even as is a little  
more obvious, is quite verbose.

Something I would like to see with template constraints, is the ability to  
show text messages, like when static if is used.

Yao G.

On Sun, 13 Jun 2010 14:36:14 -0500, Leandro Lucarella <llucax gmail.com>  
wrote:


 Walter Bright, el 13 de junio a las 12:01 me escribiste:

 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html

 Anyone want to do the honors and post to reddit, ycombinator, etc. ?


 Nice article, but when I read:

 	T gcd(T)(T a, T b)
 		if (is(typeof(a % b)))
 	{
 		...
 	}

 Under the presence of such beauty as template constraint syntax is,
 "is(typeof(a % b))" makes my eyes hurt, and my brain wonder. When you
 get used to this idiom, it might not look to bad, but I'm seeing people
 new to D wonder "why in the hell was that syntax used?" while they try
 to decode what is(typeof()) means.

 I think is really a shame, all the beauty and clarity gained by the
 simple "if" of template constraints, is lost with the cryptic
 is(typeof()) (well, maybe not all, but a significant part).

 I would love to see this fixed/simplified. For example, by applying the
 changes in suggested in bug 3702 [1]. This looks much better, and is
 pretty clear to anyone (even people that doesn't know D):

 	T gcd(T)(T a, T b)
 		if (meta.compiles(a % b))
 	{
 		...
 	}

 I guess at this point this is not going to happen for D2, a real shame
 :S

 The, the article made me think that, even when template constraints are
 useful not only for improving error reporting, they are often used only
 for that. Since the compiler knows all the operations a template
 parameter will need to satisfy from the function body, it would be nice
 to have some sort of way to tell the compiler to write the template
 constraints for us (the obvious ones at least, there might be other
 template constraints desired besides the ones the ones the compiler can
 figure out). This way, the errors can be improved without user
 intervention.

 [1] http://d.puremagic.com/issues/show_bug.cgi?id=3702



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Jun 13 2010
prev sibling next sibling parent reply =?UTF-8?B?QWxpIMOHZWhyZWxp?= <acehreli yahoo.com> writes:
Leandro Lucarella wrote:

 Walter Bright, el 13 de junio a las 12:01 me escribiste:

 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html

 Anyone want to do the honors and post to reddit, ycombinator, etc. ?


 Nice article, but when I read:

 	T gcd(T)(T a, T b)
 		if (is(typeof(a % b)))
 	{
 		...
 	}

 Under the presence of such beauty as template constraint syntax is,
 "is(typeof(a % b))" makes my eyes hurt, and my brain wonder.


The method that I learned from Phobos is not the best, but at least the 
function interface reads better:

T gcd(T)(T a, T b)
     if (supports_modulus!T)
{
     T result;
     // ...
     return result;
}

That makes it clear that T must support the modulus operation.

Here is how supports_modulus may be defined:

template supports_modulus(T)
{
     const bool supports_modulus = is (typeof(
     {
         T a;
         T b;
         T c = a % b;
     }()));
}

Here are what is at play to get that "named constraint":

- The curly braces after the typeof define a closure (delegate?) literal

- The body includes how the type should be used to satisfy the 
'supports_modulus' constraint

- The closure is "executed" with the empty parethesis at the end of it 
(actually, not executed at all; because the expression is a parameter to 
typeof, which never executes its argument)

- [This is my assumption] typeof produces an invalid type for invalid 
expressions

- One of the uses of the 'is' expression produces a bool result if the 
type that it receives is not valid

- When a template contains just one definition and that definition 
matches the name of the template, the template instantiation is the same 
as the definition that it contains. i.e. instead of writing

     (supports_modulus!T).supports_modulus

to mean the bool value, one merely writes

      supports_modulus!T

A lot of D features! :) But once that code structure is accepted as an 
idiom, it works and gives names to constraints.

Ali
Jun 13 2010
parent BCS <none anon.com> writes:
Hello Ali,



 The method that I learned from Phobos is not the best, but at least
 the function interface reads better:
 
 T gcd(T)(T a, T b)
 if (supports_modulus!T)
 {
 T result;
 // ...
 return result;
 }
 That makes it clear that T must support the modulus operation.
 
 Here is how supports_modulus may be defined:
 
 template supports_modulus(T)
 {
 const bool supports_modulus = is (typeof(
 {
 T a;
 T b;
 T c = a % b;
 }()));
 }


this shorter vertion also works:

template supports_modulus(T)
{
const bool supports_modulus = is (typeof(T.init%T.init) == T);
}

-- 
... <IXOYE><
Jun 13 2010
prev sibling next sibling parent "Steven E. Harris" <seh panix.com> writes:
Leandro Lucarella <llucax gmail.com> writes:


 When you get used to this idiom, it might not look to bad, but I'm
 seeing people new to D wonder "why in the hell was that syntax used?"
 while they try to decode what is(typeof()) means.


The syntax doesn't bother me as much as the suggestion to repeat details
of the template's implementation. C++'s concepts were going in this
direction too, adding code that suffers the same problem as
documentation: It can fall out of step with "the real code", and keeping
it aligned requires repeating details that should only be stated once --
in "the real code".

Here we're debating how to state that some type can participate in a
modulus operation, but that we even need to apply this operation to the
function's arguments is an internal detail. It could change after the
first pass at implementing the function, adopting a different algorithm,
and we could wind up continuing to impose requirements on the types that
are no longer relevant to the revised implementation. It looks like yet
another maintenance burden.

-- 
Steven E. Harris
Jun 13 2010
prev sibling parent reply "Simen kjaeraas" <simen.kjaras gmail.com> writes:
Leandro Lucarella <llucax gmail.com> wrote:

 it would be nice
 to have some sort of way to tell the compiler to write the template
 constraints for us (the obvious ones at least, there might be other
 template constraints desired besides the ones the ones the compiler can
 figure out). This way, the errors can be improved without user
 intervention.


How's about  optional? Marks an entire function as optional, i.e. will
not be included if it does not compile.

struct foo( T ) {
    optional
   void bar( ) { // Will not exist if T cannot be flabbergasted.
     T tmp;
     t.flabbergast( );
   }
}

-- 
Simen
Jun 14 2010
next sibling parent reply Jason House <jadon.james.house gmail.com> writes:
Simen kjaeraas Wrote:


 Leandro Lucarella <llucax gmail.com> wrote:

 it would be nice
 to have some sort of way to tell the compiler to write the template
 constraints for us (the obvious ones at least, there might be other
 template constraints desired besides the ones the ones the compiler can
 figure out). This way, the errors can be improved without user
 intervention.


 
 How's about  optional? Marks an entire function as optional, i.e. will
 not be included if it does not compile.
 
 struct foo( T ) {
     optional
    void bar( ) { // Will not exist if T cannot be flabbergasted.
      T tmp;
      t.flabbergast( );
    }
 }
 
 -- 
 Simen


Would it be a bug or a feature if bar was never included? As written, bar
should never compile...
Jun 14 2010
parent "Simen kjaeraas" <simen.kjaras gmail.com> writes:
Jason House <jadon.james.house gmail.com> wrote:


 Simen kjaeraas Wrote:


 Leandro Lucarella <llucax gmail.com> wrote:

 it would be nice
 to have some sort of way to tell the compiler to write the template
 constraints for us (the obvious ones at least, there might be other
 template constraints desired besides the ones the ones the compiler  


 can

 figure out). This way, the errors can be improved without user
 intervention.


 How's about  optional? Marks an entire function as optional, i.e. will
 not be included if it does not compile.

 struct foo( T ) {
     optional
    void bar( ) { // Will not exist if T cannot be flabbergasted.
      T tmp;
      t.flabbergast( );
    }
 }

 --
 Simen


 Would it be a bug or a feature if bar was never included? As written,  
 bar should never compile...


Well spotted. And a very good reason not to have this feature (or to
write unit tests, I guess :p )

-- 
Simen
Jun 14 2010
prev sibling parent reply Don <nospam nospam.com> writes:
Simen kjaeraas wrote:

 Leandro Lucarella <llucax gmail.com> wrote:

 it would be nice
 to have some sort of way to tell the compiler to write the template
 constraints for us (the obvious ones at least, there might be other
 template constraints desired besides the ones the ones the compiler can
 figure out). This way, the errors can be improved without user
 intervention.


 
 How's about  optional? Marks an entire function as optional, i.e. will
 not be included if it does not compile.
 
 struct foo( T ) {
    optional
   void bar( ) { // Will not exist if T cannot be flabbergasted.
     T tmp;
     t.flabbergast( );
   }
 }
 


This can be trivially shown to be NP-complete.

void bar(T)()
{
    static if ( big_function!T) {
      T t;
      t.flabbergast( );
    }
}

Compiler cannot determine if T needs flabbergast(), unless it evaluates 
big_function. Which can be arbitrarily complicated.

We're taking the approach of making the default error message in such 
cases as helpful as possible.
Jun 14 2010
parent "Simen kjaeraas" <simen.kjaras gmail.com> writes:
Don <nospam nospam.com> wrote:


 This can be trivially shown to be NP-complete.

 void bar(T)()
 {
     static if ( big_function!T) {
       T t;
       t.flabbergast( );
     }
 }

 Compiler cannot determine if T needs flabbergast(), unless it evaluates  
 big_function. Which can be arbitrarily complicated.


But how does the compiler decide whether or not to include that code
usually? Would it not then also have to evaluate big_function?



 We're taking the approach of making the default error message in such  
 cases as helpful as possible.


Which is likely a better choice. The idea just hit me, and I decided
to probe for reactions.

-- 
Simen
Jun 14 2010
prev sibling parent reply "Nick Sabalausky" <a a.a> writes:
"Walter Bright" <newshound1 digitalmars.com> wrote in message 
news:hv39tt$87s$1 digitalmars.com...

 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html

 Anyone want to do the honors and post to reddit, ycombinator, etc. ?


"and a line number pointing into the template body. The error is being 
reported as occurring somewhere other than where the actual problem is"

That would be a good place to point out that recent versions of DMD display 
template instantiation traces upon template instatiation errors (Does C++ do 
that?). Then, of course, it could go on to "but better yet...".

-------------------------------
Not sent from an iPhone.
Jun 13 2010
next sibling parent BCS <none anon.com> writes:
Hello Nick,


 "Walter Bright" <newshound1 digitalmars.com> wrote in message
 news:hv39tt$87s$1 digitalmars.com...
 

 http://www.drdobbs.com/blog/archives/2010/06/constrained_tem.html
 
 Anyone want to do the honors and post to reddit, ycombinator, etc. ?
 


 "and a line number pointing into the template body. The error is being
 reported as occurring somewhere other than where the actual problem
 is"
 
 That would be a good place to point out that recent versions of DMD
 display template instantiation traces upon template instatiation
 errors (Does C++ do that?).


I think GCC does. But IIRC it ends up being an almost unreadable one.


 -------------------------------
 Not sent from an iPhone.


-- 
... <IXOYE><
Jun 13 2010
prev sibling parent reply Walter Bright <newshound1 digitalmars.com> writes:
Nick Sabalausky wrote:

 "and a line number pointing into the template body. The error is being 
 reported as occurring somewhere other than where the actual problem is"
 
 That would be a good place to point out that recent versions of DMD display 
 template instantiation traces upon template instatiation errors (Does C++ do 
 that?). Then, of course, it could go on to "but better yet...".


I wished to avoid quality of implementation issues, and instead focus on 
characteristics of the language.
Jun 13 2010
parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Walter Bright wrote:

 Nick Sabalausky wrote:

 "and a line number pointing into the template body. The error is being 
 reported as occurring somewhere other than where the actual problem is"

 That would be a good place to point out that recent versions of DMD 
 display template instantiation traces upon template instatiation 
 errors (Does C++ do that?). Then, of course, it could go on to "but 
 better yet...".


 
 I wished to avoid quality of implementation issues, and instead focus on 
 characteristics of the language.


I think Nick does make a valid point: there is access to the error, it 
just takes some more info from the compiler.

Andrei
Jun 13 2010
parent "Nick Sabalausky" <a a.a> writes:
"Andrei Alexandrescu" <SeeWebsiteForEmail erdani.org> wrote in message 
news:hv46bv$1kpl$2 digitalmars.com...

 Walter Bright wrote:

 Nick Sabalausky wrote:

 "and a line number pointing into the template body. The error is being 
 reported as occurring somewhere other than where the actual problem is"

 That would be a good place to point out that recent versions of DMD 
 display template instantiation traces upon template instatiation errors 
 (Does C++ do that?). Then, of course, it could go on to "but better 
 yet...".


 I wished to avoid quality of implementation issues, and instead focus on 
 characteristics of the language.


 I think Nick does make a valid point: there is access to the error, it 
 just takes some more info from the compiler.


My concern was that C++ users (or other non-D-users) might get the mistaken 
impression that without constraints, *all* they would get is the unhelpful 
"Error: incompatible types for ((a) % (b)): 'int*' and 'int*'", and 
re-enforce their belief that templates have an innate tendency to give 
useless error messages, and that if you use D, constraints are the only help 
you get.

But I may be over-thinking it and worrying about nothing.
Jun 13 2010