www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Concerns about using struct initializer in UDA?

reply Andre Pany <andre s-e-a-p.de> writes:
Hi,

I know there are concerns about struct initialization in method 
calls but what is about struct initializer in UDA?

Scenario: I want to set several UDA values. At the moment I have 
to create for each value a structure with exactly 1 field. But it 
would be quite nice if I could use struct initialization to group 
these values:

struct Field
{
	string location;
	string locationName;
}

struct Foo
{
	 A = {locationName: "B"} int c; // <------------------
}

void main() {}

Of course the syntax is questionable, it is just a proposal.
What do you think?

Kind regards
André
May 11 2017
next sibling parent reply Stefan Koch <uplink.coder googlemail.com> writes:
On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 Hi,

 I know there are concerns about struct initialization in method 
 calls but what is about struct initializer in UDA?

 Scenario: I want to set several UDA values. At the moment I 
 have to create for each value a structure with exactly 1 field. 
 But it would be quite nice if I could use struct initialization 
 to group these values:

 struct Field
 {
 	string location;
 	string locationName;
 }

 struct Foo
 {
 	 A = {locationName: "B"} int c; // <------------------
 }

 void main() {}

 Of course the syntax is questionable, it is just a proposal.
 What do you think?

 Kind regards
 André


We have that syntax already.
May 11 2017
parent reply Andre Pany <andre s-e-a-p.de> writes:
On Thursday, 11 May 2017 at 10:51:09 UTC, Stefan Koch wrote:

 On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 Hi,

 I know there are concerns about struct initialization in 
 method calls but what is about struct initializer in UDA?

 Scenario: I want to set several UDA values. At the moment I 
 have to create for each value a structure with exactly 1 
 field. But it would be quite nice if I could use struct 
 initialization to group these values:

 struct Field
 {
 	string location;
 	string locationName;
 }

 struct Foo
 {
 	 A = {locationName: "B"} int c; // <------------------
 }

 void main() {}

 Of course the syntax is questionable, it is just a proposal.
 What do you think?

 Kind regards
 André


 We have that syntax already.


I do not understand. Should the syntax I have written already 
work as I expect or do you mean my proposal is not possible as 
the syntax is ambiguous?

Kind regards
André
May 11 2017
parent reply Stefan Koch <uplink.coder googlemail.com> writes:
On Thursday, 11 May 2017 at 11:36:17 UTC, Andre Pany wrote:

 On Thursday, 11 May 2017 at 10:51:09 UTC, Stefan Koch wrote:

 On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 [...]


 We have that syntax already.


 I do not understand. Should the syntax I have written already 
 work as I expect or do you mean my proposal is not possible as 
 the syntax is ambiguous?

 Kind regards
 André


I thought it should have worked already.

My apologies the struct literal initialization syntax is 
unsupported because of the parser implementation.

I don't know if you would introduce new ambiguities; I suspect 
that you wouldn't.
May 11 2017
parent reply Andre Pany <andre s-e-a-p.de> writes:
On Thursday, 11 May 2017 at 11:57:01 UTC, Stefan Koch wrote:

 On Thursday, 11 May 2017 at 11:36:17 UTC, Andre Pany wrote:

 On Thursday, 11 May 2017 at 10:51:09 UTC, Stefan Koch wrote:

 On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 [...]


 We have that syntax already.


 I do not understand. Should the syntax I have written already 
 work as I expect or do you mean my proposal is not possible as 
 the syntax is ambiguous?

 Kind regards
 André


 I thought it should have worked already.

 My apologies the struct literal initialization syntax is 
 unsupported because of the parser implementation.

 I don't know if you would introduce new ambiguities; I suspect 
 that you wouldn't.


Does it make sense to open a DIP? It would make UDA quite nice to 
use if you can group several attributes into a struct.

Or someone know whether there are already known issues why it 
isn't possible?

Kind regards
André
May 12 2017
parent reply Seb <seb wilzba.ch> writes:
On Friday, 12 May 2017 at 10:13:02 UTC, Andre Pany wrote:

 Does it make sense to open a DIP? It would make UDA quite nice 
 to use if you can group several attributes into a struct.

 Or someone know whether there are already known issues why it 
 isn't possible?


The entire point of a DIP is to do this research for everyone and 
summarize it an easy-to-access and easy-to-find fashion ;-)
FWIW there is an abandoned DIP for in-place struct initialization 
(https://github.com/dlang/DIPs/pull/22) - it might make sense to 
invest the effort to revive it from the dead...
May 12 2017
parent Andre Pany <andre s-e-a-p.de> writes:
On Friday, 12 May 2017 at 10:25:07 UTC, Seb wrote:

 On Friday, 12 May 2017 at 10:13:02 UTC, Andre Pany wrote:

 Does it make sense to open a DIP? It would make UDA quite nice 
 to use if you can group several attributes into a struct.

 Or someone know whether there are already known issues why it 
 isn't possible?


 The entire point of a DIP is to do this research for everyone 
 and summarize it an easy-to-access and easy-to-find fashion ;-)
 FWIW there is an abandoned DIP for in-place struct 
 initialization (https://github.com/dlang/DIPs/pull/22) - it 
 might make sense to invest the effort to revive it from the 
 dead...


If I find time I will create a new dip. The dip you mentioned is 
about a different usage scenario. My proposal is only about 
sructures used as UDA.
As far as I remember the old dip was not possible due to comma 
syntax which is now in deprecation phase.

Kund regards
André
May 12 2017
prev sibling parent reply Nicholas Wilson <iamthewilsonator hotmail.com> writes:
On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 Hi,

 I know there are concerns about struct initialization in method 
 calls but what is about struct initializer in UDA?

 Scenario: I want to set several UDA values. At the moment I 
 have to create for each value a structure with exactly 1 field. 
 But it would be quite nice if I could use struct initialization 
 to group these values:

 struct Field
 {
 	string location;
 	string locationName;
 }

 struct Foo
 {
 	 A = {locationName: "B"} int c; // <------------------
 }

 void main() {}

 Of course the syntax is questionable, it is just a proposal.
 What do you think?

 Kind regards
 André


I'm not sure if it isi what you're looking for but you can do

  struct Field
  {
  	string location;
  	string locationName;
  }

  struct Foo
  {
  	 Field("","B") int c;
  }

We use this for our compiler recognised attributes in LDC, see 
llvmAttr for an example
(https://github.com/ldc-developers/druntime/blob/ldc/src/ldc/attributes.d#L108)
May 12 2017
parent reply Andre Pany <andre s-e-a-p.de> writes:
On Friday, 12 May 2017 at 10:23:34 UTC, Nicholas Wilson wrote:

 On Thursday, 11 May 2017 at 10:49:58 UTC, Andre Pany wrote:

 Hi,

 I know there are concerns about struct initialization in 
 method calls but what is about struct initializer in UDA?

 Scenario: I want to set several UDA values. At the moment I 
 have to create for each value a structure with exactly 1 
 field. But it would be quite nice if I could use struct 
 initialization to group these values:

 struct Field
 {
 	string location;
 	string locationName;
 }

 struct Foo
 {
 	 A = {locationName: "B"} int c; // <------------------
 }

 void main() {}

 Of course the syntax is questionable, it is just a proposal.
 What do you think?

 Kind regards
 André


 I'm not sure if it isi what you're looking for but you can do

  struct Field
  {
  	string location;
  	string locationName;
  }

  struct Foo
  {
  	 Field("","B") int c;
  }

 We use this for our compiler recognised attributes in LDC, see 
 llvmAttr for an example
 (https://github.com/ldc-developers/druntime/blob/ldc/src/ldc/attributes.d#L108)


I think there are cases where this syntax is more readable and 
there are cases where struct initialization is more readable. In 
the example you gave, I do not know what is the second field 
about and having to mention the first field feels like a burden. 
As developer I want the choice to decide between these 2 syntax 
options.

Kind regards
André
May 12 2017
parent reply ag0aep6g <anonymous example.com> writes:
On 05/12/2017 09:01 PM, Andre Pany wrote:

 On Friday, 12 May 2017 at 10:23:34 UTC, Nicholas Wilson wrote:


[...]

 I'm not sure if it isi what you're looking for but you can do

  struct Field
  {
      string location;
      string locationName;
  }

  struct Foo
  {
       Field("","B") int c;
  }




[...]

 I think there are cases where this syntax is more readable and there are
 cases where struct initialization is more readable. In the example you
 gave, I do not know what is the second field about and having to mention
 the first field feels like a burden. As developer I want the choice to
 decide between these 2 syntax options.


You can create the attribute separately:

     enum Field a = { locationName: "B" };
      a int c;

Or if avoiding the extra symbol is more important than beauty, call a 
function literal:

      ((){ Field a = { locationName: "B" }; return a; }()) int c;

That's not as succinct as the syntax you propose, of course.

But your syntax (` A = {locationName: "B"} int c;`) misses an important 
detail: There's no indication what the type of the attribute is (or is 
`A` supposed to be the type?). To make it work, you will have to add 
that. Something like ` Field(locationName: "B")` or 
` Field{locationName: "B"}` or whatever works.

At that point, why limit it to attributes? Constructors like that would 
be nice everywhere. I remember such syntax being discussed repeatedly, 
but I don't know where we stand. Maybe there's a DIP already?
May 12 2017
parent Andre Pany <andre s-e-a-p.de> writes:
On Friday, 12 May 2017 at 19:33:52 UTC, ag0aep6g wrote:

 On 05/12/2017 09:01 PM, Andre Pany wrote:

 [...]


 [...]

  [...]




 [...]

 [...]


 You can create the attribute separately:

     enum Field a = { locationName: "B" };
      a int c;

 Or if avoiding the extra symbol is more important than beauty, 
 call a function literal:

      ((){ Field a = { locationName: "B" }; return a; }()) int c;

 That's not as succinct as the syntax you propose, of course.

 But your syntax (` A = {locationName: "B"} int c;`) misses an 
 important detail: There's no indication what the type of the 
 attribute is (or is `A` supposed to be the type?). To make it 
 work, you will have to add that. Something like 
 ` Field(locationName: "B")` or ` Field{locationName: "B"}` or 
 whatever works.

 At that point, why limit it to attributes? Constructors like 
 that would be nice everywhere. I remember such syntax being 
 discussed repeatedly, but I don't know where we stand. Maybe 
 there's a DIP already?


Yes, A is supposed to be the type. I have included the equal sign 
to to match the existing struct initializer as far as possible.

I would like to limit the dip to increase the chances the dip is 
accepted.

Kind regards
André
May 12 2017