www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Changeing return type of struct.toString()

reply Benjamin Thaut <code benjamin-thaut.de> writes:
I'm currently trying to use D without a gc and didn't have major 
problems so far. A minor issues is that struct.toString() returns a 
string and this will almost allways leak memory when this is called.
The question now is, can I change the return type of struct.toString() 
somewhere inside druntime or is it hardcoded in the compiler?

It was possible without any problems to change the return type of 
Object.toString() and therefore every class in D because Object is the 
baseclass for every class. Is there something similar for structs?

Kind Regards
Benjamin Thaut
Feb 12 2012
next sibling parent reply Stewart Gordon <smjg_1998 yahoo.com> writes:
On 12/02/2012 18:36, Benjamin Thaut wrote:

 I'm currently trying to use D without a gc and didn't have major problems so
far. A minor
 issues is that struct.toString() returns a string and this will almost allways
leak memory
 when this is called.
 The question now is, can I change the return type of struct.toString()
somewhere inside
 druntime or is it hardcoded in the compiler?


<snip>

If you create a struct, you can declare a method called toString with whatever
return type 
you want.  Though it usually isn't a good idea.

What, exactly, are you planning to do with this non-string string
representation of a struct?

Stewart.
Feb 12 2012
parent Timon Gehr <timon.gehr gmx.ch> writes:
On 02/13/2012 01:21 AM, Stewart Gordon wrote:

 On 12/02/2012 18:36, Benjamin Thaut wrote:

 I'm currently trying to use D without a gc and didn't have major
 problems so far. A minor
 issues is that struct.toString() returns a string and this will almost
 allways leak memory
 when this is called.
 The question now is, can I change the return type of struct.toString()
 somewhere inside
 druntime or is it hardcoded in the compiler?


 <snip>

 If you create a struct, you can declare a method called toString with
 whatever return type you want. Though it usually isn't a good idea.

 What, exactly, are you planning to do with this non-string string
 representation of a struct?

 Stewart.


I suppose he wants to use a custom string type with deterministic memory 
management.
Feb 12 2012
prev sibling next sibling parent "Robert Jacques" <sandford jhu.edu> writes:
On Sun, 12 Feb 2012 12:36:47 -0600, Benjamin Thaut <code benjamin-thaut.de>
wrote:

 I'm currently trying to use D without a gc and didn't have major
 problems so far. A minor issues is that struct.toString() returns a
 string and this will almost allways leak memory when this is called.
 The question now is, can I change the return type of struct.toString()
 somewhere inside druntime or is it hardcoded in the compiler?

 It was possible without any problems to change the return type of
 Object.toString() and therefore every class in D because Object is the
 baseclass for every class. Is there something similar for structs?

 Kind Regards
 Benjamin Thaut


The primary users of toString() methods (struct or otherwise) are in std.format
and std.conv, IIRC. I'd imagine that you'd be able to catch most uses via a
correctly placed static assert.
Feb 12 2012
prev sibling parent reply Don <nospam nospam.com> writes:
On 12.02.2012 19:36, Benjamin Thaut wrote:

 I'm currently trying to use D without a gc and didn't have major
 problems so far. A minor issues is that struct.toString() returns a
 string and this will almost allways leak memory when this is called.
 The question now is, can I change the return type of struct.toString()
 somewhere inside druntime or is it hardcoded in the compiler?

 It was possible without any problems to change the return type of
 Object.toString() and therefore every class in D because Object is the
 baseclass for every class. Is there something similar for structs?

 Kind Regards
 Benjamin Thaut


I don't know why struct toString() still exists. It's a legacy from the 
very beginning of D, back when the language didn't even have templates.
std.format doesn't even require it, any more.
Feb 12 2012
next sibling parent Benjamin Thaut <code benjamin-thaut.de> writes:
Am 13.02.2012 03:21, schrieb Don:

 On 12.02.2012 19:36, Benjamin Thaut wrote:

 I'm currently trying to use D without a gc and didn't have major
 problems so far. A minor issues is that struct.toString() returns a
 string and this will almost allways leak memory when this is called.
 The question now is, can I change the return type of struct.toString()
 somewhere inside druntime or is it hardcoded in the compiler?

 It was possible without any problems to change the return type of
 Object.toString() and therefore every class in D because Object is the
 baseclass for every class. Is there something similar for structs?

 Kind Regards
 Benjamin Thaut


 I don't know why struct toString() still exists. It's a legacy from the
 very beginning of D, back when the language didn't even have templates.
 std.format doesn't even require it, any more.


Well TypeInfo_struct declarers a delegate for the xtostring member which 
returns a char[]. So I'm curiours if the compiler fills that member, and 
if it does any checking on the toString() method while doing so.

Basically I have 2 string types now:
1) string = garantueed to be either in static memory or to outlive every 
object that has a reference to it
2) rcstring = reference counted string (you have to use the most ugly 
casts you can think of to actually make reference counting possible with 
the D typesystem)
Feb 12 2012
prev sibling parent reply Stewart Gordon <smjg_1998 yahoo.com> writes:
On 13/02/2012 02:21, Don wrote:
<snip>

 I don't know why struct toString() still exists.


<snip>

What are you talking about?  If you define a struct, it doesn't have any
methods other 
than the ones you put into it.

Stewart.
Feb 13 2012
parent reply Benjamin Thaut <code benjamin-thaut.de> writes:
Am 13.02.2012 13:26, schrieb Stewart Gordon:

 On 13/02/2012 02:21, Don wrote:
 <snip>

 I don't know why struct toString() still exists.


 <snip>

 What are you talking about? If you define a struct, it doesn't have any
 methods other than the ones you put into it.

 Stewart.


import std.stdio;

struct fun
{
   string toString()
   {
     return "fun";
   }
}

void main(string[] args)
{
   auto ti = cast(TypeInfo_Struct)typeid(fun);
   fun gun;
   writefln("%d",cast(void*)ti.xtoString);
   writefln("%s",ti.xtoString(&gun));
}

If you change the return type of toString() to int for example the 
program will crash because ti.xtoString will be null. Therefore the 
compiler seems to check what type toString() does return and only fills 
the type info if it actually does return a string.

-- 
Kind Regards
Benjamin Thaut
Feb 13 2012
parent reply "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Mon, 13 Feb 2012 08:40:20 -0500, Benjamin Thaut  
<code benjamin-thaut.de> wrote:


 Am 13.02.2012 13:26, schrieb Stewart Gordon:

 On 13/02/2012 02:21, Don wrote:
 <snip>

 I don't know why struct toString() still exists.


 <snip>

 What are you talking about? If you define a struct, it doesn't have any
 methods other than the ones you put into it.

 Stewart.


 import std.stdio;

 struct fun
 {
    string toString()
    {
      return "fun";
    }
 }

 void main(string[] args)
 {
    auto ti = cast(TypeInfo_Struct)typeid(fun);
    fun gun;
    writefln("%d",cast(void*)ti.xtoString);
    writefln("%s",ti.xtoString(&gun));
 }

 If you change the return type of toString() to int for example the  
 program will crash because ti.xtoString will be null. Therefore the  
 compiler seems to check what type toString() does return and only fills  
 the type info if it actually does return a string.


That is a legacy feature, back when writefln was a D variadic function  
(not a template), and all you had was the TypeInfo.

It's essentially a hack that the compiler puts several "special" function  
pointers into the typeinfo to give structs a sort of runtime interface.  I  
think we can get rid of most, if not all, of those x-functions.  But then  
we need to get rid of all the usages of those.  It shouldn't be too  
difficult at this point, it just needs to be done.

-Steve
Feb 13 2012
parent reply Benjamin Thaut <code benjamin-thaut.de> writes:
Am 13.02.2012 15:23, schrieb Steven Schveighoffer:

 On Mon, 13 Feb 2012 08:40:20 -0500, Benjamin Thaut
 <code benjamin-thaut.de> wrote:


 Am 13.02.2012 13:26, schrieb Stewart Gordon:

 On 13/02/2012 02:21, Don wrote:
 <snip>

 I don't know why struct toString() still exists.


 <snip>

 What are you talking about? If you define a struct, it doesn't have any
 methods other than the ones you put into it.

 Stewart.


 import std.stdio;

 struct fun
 {
 string toString()
 {
 return "fun";
 }
 }

 void main(string[] args)
 {
 auto ti = cast(TypeInfo_Struct)typeid(fun);
 fun gun;
 writefln("%d",cast(void*)ti.xtoString);
 writefln("%s",ti.xtoString(&gun));
 }

 If you change the return type of toString() to int for example the
 program will crash because ti.xtoString will be null. Therefore the
 compiler seems to check what type toString() does return and only
 fills the type info if it actually does return a string.


 That is a legacy feature, back when writefln was a D variadic function
 (not a template), and all you had was the TypeInfo.

 It's essentially a hack that the compiler puts several "special"
 function pointers into the typeinfo to give structs a sort of runtime
 interface. I think we can get rid of most, if not all, of those
 x-functions. But then we need to get rid of all the usages of those. It
 shouldn't be too difficult at this point, it just needs to be done.

 -Steve


So you want every format like function to be a template instead of a 
variadic function? I actually found it pretty nice that printing structs 
is possible without templates. I don't want to imagine how big the code 
bload would be if you actually make format a template instead of a 
variadic function.

-- 
Kind Regards
Benjamin Thaut
Feb 13 2012
parent "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Mon, 13 Feb 2012 09:41:14 -0500, Benjamin Thaut  
<code benjamin-thaut.de> wrote:


 Am 13.02.2012 15:23, schrieb Steven Schveighoffer:

 That is a legacy feature, back when writefln was a D variadic function
 (not a template), and all you had was the TypeInfo.

 It's essentially a hack that the compiler puts several "special"
 function pointers into the typeinfo to give structs a sort of runtime
 interface. I think we can get rid of most, if not all, of those
 x-functions. But then we need to get rid of all the usages of those. It
 shouldn't be too difficult at this point, it just needs to be done.

 -Steve


 So you want every format like function to be a template instead of a  
 variadic function? I actually found it pretty nice that printing structs  
 is possible without templates. I don't want to imagine how big the code  
 bload would be if you actually make format a template instead of a  
 variadic function.


No need to imagine, writefln does not use xtoString or the TypeInfo AFAIK.

But in any case, the template can just be glue code.  All the TypeInfo  
does is store a function pointer to the toString.  If you took that  
delegate from each item, and then passed it into a non-template function,  
the bulk of the format code would not be repeated in the template.

And it would be more robust -- toString wouldn't have to necessarily be  
always the same signature.

All this is not the right way to do formatting anyway, see DIP9  
http://prowiki.org/wiki4d/wiki.cgi?LanguageDevel/DIPs/DIP9

-Steve
Feb 13 2012