www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - [NOTABLE PR] First step from traditional to generic runtime

reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Worth a look: https://github.com/dlang/druntime/pull/1781. This moves 
comparison code away from tedious runtime-introspected routines to nice 
templates. -- Andrei
Mar 02 2017
next sibling parent sarn <sarn theartofmachinery.com> writes:
On Thursday, 2 March 2017 at 19:32:23 UTC, Andrei Alexandrescu 
wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This 
 moves comparison code away from tedious runtime-introspected 
 routines to nice templates. -- Andrei


Great news :)
Mar 02 2017
prev sibling next sibling parent reply rikki cattermole <rikki cattermole.co.nz> writes:
On 03/03/2017 8:32 AM, Andrei Alexandrescu wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This moves
 comparison code away from tedious runtime-introspected routines to nice
 templates. -- Andrei


If we end up full on templated TypeInfo that would be lovely.

TypeInfo_Class clasz = TypeInfo_ClassT!Foo;

class TypeInfo_ClassT(T) if(is(T == class)) {
	version(FullReflection) {
		string[] methodNames() { ... }
	}
}

But just a thought.
Mar 02 2017
parent reply Adam D. Ruppe <destructionator gmail.com> writes:
On Friday, 3 March 2017 at 02:59:57 UTC, rikki cattermole wrote:

 If we end up full on templated TypeInfo that would be lovely.

 TypeInfo_Class clasz = TypeInfo_ClassT!Foo;

 class TypeInfo_ClassT(T) if(is(T == class)) {
 	version(FullReflection) {
 		string[] methodNames() { ... }
 	}
 }



We already have the capability to do that with the existing dmd 
and druntime - RTInfo is a template instantiated for each type, 
linked into TypeInfo, so we can extend it... if we are willing to 
edit object.d.

What is really nice about templating TypeInfo though is removing 
some annoying code from druntime and also being able to tweak 
that in constrained environments.... we can already add anything 
we want, being able to *remove* is what's going to be new.
Mar 02 2017
parent rikki cattermole <rikki cattermole.co.nz> writes:
On 03/03/2017 4:06 PM, Adam D. Ruppe wrote:

 On Friday, 3 March 2017 at 02:59:57 UTC, rikki cattermole wrote:

 If we end up full on templated TypeInfo that would be lovely.

 TypeInfo_Class clasz = TypeInfo_ClassT!Foo;

 class TypeInfo_ClassT(T) if(is(T == class)) {
     version(FullReflection) {
         string[] methodNames() { ... }
     }
 }



 We already have the capability to do that with the existing dmd and
 druntime - RTInfo is a template instantiated for each type, linked into
 TypeInfo, so we can extend it... if we are willing to edit object.d.

 What is really nice about templating TypeInfo though is removing some
 annoying code from druntime and also being able to tweak that in
 constrained environments.... we can already add anything we want, being
 able to *remove* is what's going to be new.


Yeah, once it is fully template based it gives us the ability to 
actually play with TypeInfo and not have to go modify the compiler on 
any set of changes. So while we do have RTInfo, we don't have this 
freedom to change and adapt it like we could and this PR moves us 
towards that hopefully.
Mar 02 2017
prev sibling next sibling parent Walter Bright <newshound2 digitalmars.com> writes:
On 3/2/2017 11:32 AM, Andrei Alexandrescu wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This moves
comparison
 code away from tedious runtime-introspected routines to nice templates. --
Andrei


It also moves logic from the compiler to druntime, where it becomes much more 
easily adapted to users' requirements.
Mar 02 2017
prev sibling next sibling parent reply Dukc <ajieskola gmail.com> writes:
On Thursday, 2 March 2017 at 19:32:23 UTC, Andrei Alexandrescu 
wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This 
 moves comparison code away from tedious runtime-introspected 
 routines to nice templates. -- Andrei


This means that if I don't use the templated features, they don't 
affect my program size nor prevent compilation for platforms 
where they aren't implemented or have errors.

Did I understand right?
Mar 02 2017
parent Walter Bright <newshound2 digitalmars.com> writes:
On 3/2/2017 11:04 PM, Dukc wrote:

 On Thursday, 2 March 2017 at 19:32:23 UTC, Andrei Alexandrescu wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This moves
 comparison code away from tedious runtime-introspected routines to nice
 templates. -- Andrei


 This means that if I don't use the templated features, they don't affect my
 program size nor prevent compilation for platforms where they aren't
implemented
 or have errors.

 Did I understand right?


That's right. Of course, that was true before, too. What the templates do is
put 
the ABI under user control.
Mar 03 2017
prev sibling next sibling parent reply Kagamin <spam here.lot> writes:
Nitpick: object.d is for symbols visible to user code, but it's 
not necessary to provide these helper functions to used code, so 
they should be in a different module known to the compiler.
Mar 03 2017
next sibling parent Jacob Carlborg <doob me.com> writes:
On 2017-03-03 16:16, Kagamin wrote:

 Nitpick: object.d is for symbols visible to user code, but it's not
 necessary to provide these helper functions to used code, so they should
 be in a different module known to the compiler.


Exactly https://github.com/dlang/druntime/pull/1781#issuecomment-283904287

-- 
/Jacob Carlborg
Mar 03 2017
prev sibling parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 3/3/17 10:16 AM, Kagamin wrote:

 Nitpick: object.d is for symbols visible to user code, but it's not
 necessary to provide these helper functions to used code, so they should
 be in a different module known to the compiler.


Fundamentally the code resulting from lowering must be in the standard 
language, i.e. no "magic" artifacts, no hidden symbols, no special 
rules. The lowering should be as if the user replaced the code with the 
lowered version by hand. -- Andrei
Mar 03 2017
parent reply Kagamin <spam here.lot> writes:
v1==v2;

can be lowered as

{
   import rthelpers:cmp;
   cmp(v1,v2);
}

or something like that
Mar 06 2017
parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 3/6/17 8:42 AM, Kagamin wrote:

 v1==v2;

 can be lowered as

 {
   import rthelpers:cmp;
   cmp(v1,v2);
 }

 or something like that


Interesting idea, will keep it in mind. Thanks! -- Andrei
Mar 06 2017
parent reply Kagamin <spam here.lot> writes:
On Monday, 6 March 2017 at 18:06:27 UTC, Andrei Alexandrescu 
wrote:

 On 3/6/17 8:42 AM, Kagamin wrote:

 v1==v2;

 can be lowered as

 {
   import rthelpers:cmp;
   cmp(v1,v2);
 }

 or something like that


 Interesting idea, will keep it in mind. Thanks! -- Andrei


Even better:

in object.d:
---
public import _d_helpers=core.helpers;
---

lowering:
---
v1==v2
to
_d_helpers.cmp(v1,v2)
---

This has the best scalability and zero possibility of code 
breakage and probably can be done right now. Also since the 
renamed import must be referenced explicitly, it doesn't need to 
be processed until used.
Mar 16 2017
parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 3/16/17 10:52 AM, Kagamin wrote:

 On Monday, 6 March 2017 at 18:06:27 UTC, Andrei Alexandrescu wrote:

 On 3/6/17 8:42 AM, Kagamin wrote:

 v1==v2;

 can be lowered as

 {
   import rthelpers:cmp;
   cmp(v1,v2);
 }

 or something like that


 Interesting idea, will keep it in mind. Thanks! -- Andrei


 Even better:

 in object.d:
 ---
 public import _d_helpers=core.helpers;
 ---

 lowering:
 ---
 v1==v2
 to
 _d_helpers.cmp(v1,v2)
 ---

 This has the best scalability and zero possibility of code breakage and
 probably can be done right now. Also since the renamed import must be
 referenced explicitly, it doesn't need to be processed until used.


A public import inside object.d is not useful because essentially it 
opens more files to load the same amount of code. Generally, there are a 
few interesting tradeoffs that govern handling of modularity in 
object.d. -- Andrei
Mar 16 2017
prev sibling parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 03/02/2017 02:32 PM, Andrei Alexandrescu wrote:

 Worth a look: https://github.com/dlang/druntime/pull/1781. This moves
 comparison code away from tedious runtime-introspected routines to nice
 templates. -- Andrei


Another notable PR in the same vein: 
https://github.com/dlang/druntime/pull/1792. -- Andrei
Mar 15 2017