www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Replace druntime Hooks with Templates: Milestone 1, Week 2

reply Teodor Dutu <teodor.dutu gmail.com> writes:
Hi,

This week I faced a few setbacks with `_d_arraycatnTX` due to 
type mismatches.
Due to the hustle of moving house, I haven't made much progress 
on this front.

To take a break from this, I temporarily moved on to the 
`_d_newitem{T,iT,U}` hooks.
I implemented templated one templated hook for `_d_newitemT` and 
added unittests for it.
The difference will now be made by the compiler during the 
lowering, by calling the constructor, copying the default 
initialiser, or doing nothing (`void` initialisation).

This week I am looking to make some progress with this hook and 
come up with the aforementioned lowerings.

Thanks,\
Teodor
Oct 03 2022
parent reply Johan <j j.nl> writes:
On Monday, 3 October 2022 at 20:28:19 UTC, Teodor Dutu wrote:

 Hi,

 This week I faced a few setbacks with `_d_arraycatnTX` due to 
 type mismatches.
 Due to the hustle of moving house, I haven't made much progress 
 on this front.

 To take a break from this, I temporarily moved on to the 
 `_d_newitem{T,iT,U}` hooks.
 I implemented templated one templated hook for `_d_newitemT` 
 and added unittests for it.
 The difference will now be made by the compiler during the 
 lowering, by calling the constructor, copying the default 
 initialiser, or doing nothing (`void` initialisation).


Hi Teodor,
   Are you aware about the capability to override these 
implementations by the user? DMD itself is using this, see for 
example: 
https://github.com/dlang/dmd/blob/16fad687b0e894879564864d7eb75747fb41a057/compiler/src/dmd/root/rmem.d#L257-L262

Is this still possible after your work?

regards,
   Johan
Oct 03 2022
next sibling parent reply Teodor Dutu <teodor.dutu gmail.com> writes:
On Monday, 3 October 2022 at 22:28:51 UTC, Johan wrote:

 On Monday, 3 October 2022 at 20:28:19 UTC, Teodor Dutu wrote:

 Hi,

 This week I faced a few setbacks with `_d_arraycatnTX` due to 
 type mismatches.
 Due to the hustle of moving house, I haven't made much 
 progress on this front.

 To take a break from this, I temporarily moved on to the 
 `_d_newitem{T,iT,U}` hooks.
 I implemented templated one templated hook for `_d_newitemT` 
 and added unittests for it.
 The difference will now be made by the compiler during the 
 lowering, by calling the constructor, copying the default 
 initialiser, or doing nothing (`void` initialisation).


 Hi Teodor,
   Are you aware about the capability to override these 
 implementations by the user? DMD itself is using this, see for 
 example: 
 https://github.com/dlang/dmd/blob/16fad687b0e894879564864d7eb75747fb41a057/compiler/src/dmd/root/rmem.d#L257-L262

 Is this still possible after your work?

 regards,
   Johan


Hi Johan,

I am aware of this and it will still be possible to override 
these implementations by writing your own `object.d` module, 
which is where they are instantiated from. I will also look into 
the link you gave and update the definitions there to conform to 
the new template.

Thanks,
Teodor
Oct 07 2022
parent reply Johan <j j.nl> writes:
On Friday, 7 October 2022 at 08:25:25 UTC, Teodor Dutu wrote:

 On Monday, 3 October 2022 at 22:28:51 UTC, Johan wrote:

 On Monday, 3 October 2022 at 20:28:19 UTC, Teodor Dutu wrote:

 Hi,

 This week I faced a few setbacks with `_d_arraycatnTX` due to 
 type mismatches.
 Due to the hustle of moving house, I haven't made much 
 progress on this front.

 To take a break from this, I temporarily moved on to the 
 `_d_newitem{T,iT,U}` hooks.
 I implemented templated one templated hook for `_d_newitemT` 
 and added unittests for it.
 The difference will now be made by the compiler during the 
 lowering, by calling the constructor, copying the default 
 initialiser, or doing nothing (`void` initialisation).


 Hi Teodor,
   Are you aware about the capability to override these 
 implementations by the user? DMD itself is using this, see for 
 example: 
 https://github.com/dlang/dmd/blob/16fad687b0e894879564864d7eb75747fb41a057/compiler/src/dmd/root/rmem.d#L257-L262

 Is this still possible after your work?

 regards,
   Johan


 Hi Johan,

 I am aware of this and it will still be possible to override 
 these implementations by writing your own `object.d` module, 
 which is where they are instantiated from.


Of course you can override anything by writing your own 
`object.d` module = writing your own druntime. The question is, 
can the user still override it using the current way. This 
impacts DMD itself. What DMD wants to override is the memory 
allocation itself, not necessarily the initialization of the 
memory. If the templates invoke an ` weak` **non-template** 
function for allocation, then there is still the possibility of 
overriding that in user code. Please think about that.

regards,
   Johan
Oct 07 2022
parent max haughton <maxhaton gmail.com> writes:
On Friday, 7 October 2022 at 10:16:25 UTC, Johan wrote:

 On Friday, 7 October 2022 at 08:25:25 UTC, Teodor Dutu wrote:

 On Monday, 3 October 2022 at 22:28:51 UTC, Johan wrote:

   [...]


 Hi Johan,

 I am aware of this and it will still be possible to override 
 these implementations by writing your own `object.d` module, 
 which is where they are instantiated from.


 Of course you can override anything by writing your own 
 `object.d` module = writing your own druntime. The question is, 
 can the user still override it using the current way. This 
 impacts DMD itself. What DMD wants to override is the memory 
 allocation itself, not necessarily the initialization of the 
 memory. If the templates invoke an ` weak` **non-template** 
 function for allocation, then there is still the possibility of 
 overriding that in user code. Please think about that.

 regards,
   Johan


You can also override the GC using a custom GC class.

The way dmd does memory allocation is quite dumb anyway I don't 
think it would be missed. If needed the GC's collections could 
just be disabled
Oct 07 2022
prev sibling parent Steven Schveighoffer <schveiguy gmail.com> writes:
On 10/3/22 6:28 PM, Johan wrote:

 On Monday, 3 October 2022 at 20:28:19 UTC, Teodor Dutu wrote:

 Hi,

 This week I faced a few setbacks with `_d_arraycatnTX` due to type 
 mismatches.
 Due to the hustle of moving house, I haven't made much progress on 
 this front.

 To take a break from this, I temporarily moved on to the 
 `_d_newitem{T,iT,U}` hooks.
 I implemented templated one templated hook for `_d_newitemT` and added 
 unittests for it.
 The difference will now be made by the compiler during the lowering, 
 by calling the constructor, copying the default initialiser, or doing 
 nothing (`void` initialisation).


 
 Hi Teodor,
    Are you aware about the capability to override these implementations 
 by the user? DMD itself is using this, see for example: 
 https://github.com/dlang/dmd/blob/16fad687b0e894879564864d7eb75747fb41a057/compiler/src/dmd/r
ot/rmem.d#L257-L262 
 
 
 Is this still possible after your work?


Given that one of the purposes is to remove the dependence on `TypeInfo` 
for such things, is there a way to do this override without it?

Otherwise, you could potentially provide a function pointer that you can 
set if you want to override the hook, and the hook can first check that 
pointer to see if it should call that first.

Another idea is provide an easier way to override object things, like 
maybe if you plop a `customobject.d` into the dmd import path, it can 
provide overrides of things that `object` looks for.

-Steve
Oct 07 2022