Re: caller trouble

2012-07-14 Thread Jonathan M Davis
On Saturday, July 14, 2012 01:53:34 captaindet wrote:
 i need a discreet handle on the calling/instantiating source file (module).
 using __FILE__, it is surprisingly easy for functions (via argument) and
 templated functions (via template parameter) but i cannot get it working
 for templated classes. how can i make them aware of the calling module?
 
 thx, det
 
 
 module other;
 
 string fun(string infile = __FILE__){ return infile; }
 string tfun(string infile = __FILE__)(){ return infile; }
 class tclass(string infile = __FILE__){ string from = infile; }
 
 //...//
 
 module main;
 import other;
 
 void main(){
  auto _fun = fun();   //_fun == main.d
  auto _tfun = tfun(); //_tfun == main.d
  auto _tclass = new tclass!();//_tclass.from == other.d !!!
 
  //this works but i do not want to provide __FILE__ explicitly:
  auto _tclassx = new tclass!(__FILE__)(); //_tclass.from == main.d
  //and why do i get 2 different results for the last 2 cases?
 }

I believe that __FILE__ and __LINE__ are treated specially with functions in 
order for them to be filled in at the call site rather than the declaration 
site. If it's not working with classes, then that probably means that whatever 
special logic was done to make them work with functions was only done for 
functions. If you want, you can open an enhancement request:

http://d.puremagic.com/issues

However, I would point out that that would mean that every single module which 
uses tclass would end up with a different and incompatible instantiations of 
tclass, so if __FILE__ worked with class templates like you want it to, then 
any such class wouldn't work outside of the module that it was constructed in 
save for code that uses auto and typeof to get the type. And I have hard time 
believing that that would be a good idea.

With a templated function, it doesn't really matter if two different call 
points end up with different instantiations. They'll both work just fine. But 
having a class which only works in the module where it's instantiated would be 
_far_ more limiting.

However, if you _really_ want something like this, you can always just wrap 
the creation of the object in a function:

auto create(string infile = __FILE__, Args...)(Args args) { return new 
tclass!infile(args); }

Args isn't necessary if there are no arguments for the constructor, but it 
_is_ if there are, and it'll work if there aren't.

- Jonathan M Davis


Re: caller trouble

2012-07-14 Thread captaindet

On 2012-07-14 02:12, Jonathan M Davis wrote:
[..]

I believe that __FILE__ and __LINE__ are treated specially with
functions in order for them to be filled in at the call site rather
than the declaration site. If it's not working with classes, then
that probably means that whatever special logic was done to make them
work with functions was only done for functions.


IIRC, __FILE__ is a template itself. so it must be about the very moment it is resolved. 
apparently, for function (templates) default parameters are treated as if given on the 
caller side. maybe even for __FILE__ and __LINE__ as main usage case? in any case, there 
would be a consistent rule resolve all templates in argument lists on caller side, 
including default parameters if it were not broken for class templates. moreover, i 
am puzzled why the small variation in usage rusults in an inconsistent outcome:

auto _tclass = new tclass!(); //_tclass.from == other.d !!!
auto _tclassx = new tclass!(__FILE__)();  //_tclass.from == main.d



However, I would point out that that would mean that every single
module which uses tclass would end up with a different and
incompatible instantiations of tclass


yes, but not an issue in my case. there would be only one object, created from 
the main module. background: i want to do some compile time magic including 
parsing information from a file that always has the same name as the module 
containing main(){...}.



However, if you _really_ want something like this, you can always
just wrap the creation of the object in a function:


good idea, i will try this next.

thanks for your help!

/det


Re: caller trouble

2012-07-14 Thread Jonathan M Davis
On Saturday, July 14, 2012 13:30:03 captaindet wrote:
 IIRC, __FILE__ is a template itself

It's not. It's a special symbol recognized by the compiler, and ends up in 
several places in the grammar.

- Jonathan M Davis


Re: caller trouble

2012-07-14 Thread Timon Gehr

On 07/14/2012 08:53 AM, captaindet wrote:

i need a discreet handle on the calling/instantiating source file
(module). using __FILE__, it is surprisingly easy for functions (via
argument) and templated functions (via template parameter) but i cannot
get it working for templated classes. how can i make them aware of the
calling module?

thx, det


module other;

string fun(string infile = __FILE__){ return infile; }
string tfun(string infile = __FILE__)(){ return infile; }
class tclass(string infile = __FILE__){ string from = infile; }

//...//

module main;
import other;

void main(){
 auto _fun = fun();//_fun == main.d
 auto _tfun = tfun();//_tfun == main.d
 auto _tclass = new tclass!();//_tclass.from == other.d !!!

 //this works but i do not want to provide __FILE__ explicitly:
 auto _tclassx = new tclass!(__FILE__)();//_tclass.from == main.d
 //and why do i get 2 different results for the last 2 cases?
}


This is a bug.

dlang.org agrees: http://dlang.org/template.html#TemplateValueParameter


Re: caller trouble

2012-07-14 Thread captaindet

auto _fun = fun(); //_fun == main.d
auto _tfun = tfun(); //_tfun == main.d
auto _tclass = new tclass!(); //_tclass.from == other.d !!!

//this works but i do not want to provide __FILE__ explicitly:
auto _tclassx = new tclass!(__FILE__)(); //_tclass.from == main.d
//and why do i get 2 different results for the last 2 cases?
}


This is a bug.

dlang.org agrees: http://dlang.org/template.html#TemplateValueParameter


right, i missed this in the specs. there seem to be several bug reports on this 
and related issues, eg:

http://d.puremagic.com/issues/show_bug.cgi?id=4018
http://d.puremagic.com/issues/show_bug.cgi?id=5686

apparently and old issue that was never addressed...too bad