digitalmars.D.learn - How to check member function for being disable?
- Uranuz (51/51) Sep 12 2016 In my code I iterate in CT over class methods marked as @property
- Jonathan M Davis via Digitalmars-d-learn (14/23) Sep 13 2016 I really don't think that it's going to scale properly to check whether
- Jonathan M Davis via Digitalmars-d-learn (10/32) Sep 13 2016 wrote:
- Uranuz (11/49) Sep 13 2016 OK. Seems that there is nothing that I could do more about my
- Jonathan M Davis via Digitalmars-d-learn (16/25) Sep 13 2016 The main places that I can think of at the moment where @disable makes s...
- Adam D. Ruppe (6/9) Sep 13 2016 Consider the case of `alias this` or a mixin template. You might
- Jonathan M Davis via Digitalmars-d-learn (10/19) Sep 13 2016 Ah. That makes sense. Thanks for pointing out that use case.
In my code I iterate in CT over class methods marked as property and I have a probleme that one of methods is disable. So I just want to skip disable members. I found possible solution, but it's interesting to we if we have more clear and obvious way to test for disable without using __traits( compile ) for it? disable "looks" like attribute but seems that I cant't get it through __traits( getAttributes ) or __traits( getFunctionAttributes ). Maybe we could add something to test for disable if it's not already exists? void fillProperties(Ctrl)(Ctrl ctrl, IDirectiveStatement statement, Interpreter interp) { import std.meta: Alias, AliasSeq; import std.traits; import std.conv; TDataNode[string] attrDict = getNamedAttributesValues(statement, interp); foreach( memberName; AliasSeq!( __traits( allMembers, Ctrl ) ) ) { static if( __traits( compiles, __traits(getOverloads, Ctrl.init, memberName) ) ) { alias Overloads = AliasSeq!(__traits( getOverloads, Ctrl.init, memberName )); foreach( overload; Overloads ) { static if( (functionAttributes!(overload) & FunctionAttribute.property) ) { alias params = Parameters!(overload); static if( params.length == 1 ) { alias propType = params[0]; static if( __traits(compiles, mixin("{ ctrl."~memberName~" = propType.init; }") ) ) { pragma( msg, memberName, ": ", typeof(&overload) ); TDataNode attrValue = attrDict.get( memberName, TDataNode(null) ); if( !attrValue.empty ) { mixin( "ctrl." ~ memberName ~ " = nodeToDValue!(propType)(attrValue);" ); } } } } } } } }
Sep 12 2016
On Tuesday, September 13, 2016 04:58:38 Uranuz via Digitalmars-d-learn wrote:In my code I iterate in CT over class methods marked as property and I have a probleme that one of methods is disable. So I just want to skip disable members. I found possible solution, but it's interesting to we if we have more clear and obvious way to test for disable without using __traits( compile ) for it? disable "looks" like attribute but seems that I cant't get it through __traits( getAttributes ) or __traits( getFunctionAttributes ). Maybe we could add something to test for disable if it's not already exists?I really don't think that it's going to scale properly to check whether something is marked with disable. The problem is that it propagates. For instance, if a struct has a member variable that has default initialization disabled via disable this(); then that struct effectively has disable this(); too even though it doesn't have it explicitly. So, ultimately what needs to be tested for is the behavior and not the presence of disable, and that means testing with __traits(compiles, ...). And I would point out that most traits test via __traits(compiles, ...) or is(typeof(...)) rather than checking for something like an attribute. So, if don't like using __traits(compiles, ...) in metaprogramming, your going to get frustrated quickly. A large portion of the time, it's exactly the solution to the problem. - Jonathan M Davis
Sep 13 2016
On Tuesday, September 13, 2016 08:28:10 Jonathan M Davis via Digitalmars-d- learn wrote:On Tuesday, September 13, 2016 04:58:38 Uranuz via Digitalmars-d-learnwrote:What would make sense would be creating a trait to test for the disabled functionality in queston - e.g. there could be an eponymous template named something like hasDefaultInitializer (though that name is a bit long) which indicated whether a type had disabled this(); or not. Then you can use that trait in your code rather than using __traits(compiles, ...) all over the place. - Jonthan M DavisIn my code I iterate in CT over class methods marked as property and I have a probleme that one of methods is disable. So I just want to skip disable members. I found possible solution, but it's interesting to we if we have more clear and obvious way to test for disable without using __traits( compile ) for it? disable "looks" like attribute but seems that I cant't get it through __traits( getAttributes ) or __traits( getFunctionAttributes ). Maybe we could add something to test for disable if it's not already exists?I really don't think that it's going to scale properly to check whether something is marked with disable. The problem is that it propagates. For instance, if a struct has a member variable that has default initialization disabled via disable this(); then that struct effectively has disable this(); too even though it doesn't have it explicitly. So, ultimately what needs to be tested for is the behavior and not the presence of disable, and that means testing with __traits(compiles, ...). And I would point out that most traits test via __traits(compiles, ...) or is(typeof(...)) rather than checking for something like an attribute. So, if don't like using __traits(compiles, ...) in metaprogramming, your going to get frustrated quickly. A large portion of the time, it's exactly the solution to the problem.
Sep 13 2016
On Tuesday, 13 September 2016 at 15:32:57 UTC, Jonathan M Davis wrote:On Tuesday, September 13, 2016 08:28:10 Jonathan M Davis via Digitalmars-d- learn wrote:OK. Seems that there is nothing that I could do more about my example code.. So the best way to be sure if something is assignable property is to try assign to it and test whether it compiles. The question was because utill this moment I somehow was living without __traits(compiles..). Seems that my use cases just was not enough complicated... Thanks for the answers. It could be good idea to have __traits( isDisable ... ) or something for it. I admit that not only ' disabled this();' regular methods could me marked disable too..On Tuesday, September 13, 2016 04:58:38 Uranuz via Digitalmars-d-learnwrote:What would make sense would be creating a trait to test for the disabled functionality in queston - e.g. there could be an eponymous template named something like hasDefaultInitializer (though that name is a bit long) which indicated whether a type had disabled this(); or not. Then you can use that trait in your code rather than using __traits(compiles, ...) all over the place. - Jonthan M DavisIn my code I iterate in CT over class methods marked as property and I have a probleme that one of methods is disable. So I just want to skip disable members. I found possible solution, but it's interesting to we if we have more clear and obvious way to test for disable without using __traits( compile ) for it? disable "looks" like attribute but seems that I cant't get it through __traits( getAttributes ) or __traits( getFunctionAttributes ). Maybe we could add something to test for disable if it's not already exists?I really don't think that it's going to scale properly to check whether something is marked with disable. The problem is that it propagates. For instance, if a struct has a member variable that has default initialization disabled via disable this(); then that struct effectively has disable this(); too even though it doesn't have it explicitly. So, ultimately what needs to be tested for is the behavior and not the presence of disable, and that means testing with __traits(compiles, ...). And I would point out that most traits test via __traits(compiles, ...) or is(typeof(...)) rather than checking for something like an attribute. So, if don't like using __traits(compiles, ...) in metaprogramming, your going to get frustrated quickly. A large portion of the time, it's exactly the solution to the problem.
Sep 13 2016
On Tuesday, September 13, 2016 17:29:26 Uranuz via Digitalmars-d-learn wrote:OK. Seems that there is nothing that I could do more about my example code.. So the best way to be sure if something is assignable property is to try assign to it and test whether it compiles. The question was because utill this moment I somehow was living without __traits(compiles..). Seems that my use cases just was not enough complicated... Thanks for the answers. It could be good idea to have __traits( isDisable ... ) or something for it. I admit that not only ' disabled this();' regular methods could me marked disable too..The main places that I can think of at the moment where disable makes sense is for disabling default initialization - disable this(); - and disabling copying - disable this(this);. It's really intended for disabling features that would normally be there. I don't know why it would ever make sense to disable a normal function. Why would it even exist if it were disabled? So, for the compiler to allow disable on normal functions sounds like a bug to me - or at least an oversight in the design and implementation of disable - but maybe there's a legitimate reason that I'm not thinking of at the moment. Regardless, testing for it is as simple as testing whether it can be called or not, and you have to worry about that in a number of cases anyway, because the access level of the function may be such that you can't call it (e.g. it's private, and the code in question is not in the module trying to call it). So, I don't really see what testing for disable specifically would buy you. - Jonathan M Davis
Sep 13 2016
On Tuesday, 13 September 2016 at 17:52:48 UTC, Jonathan M Davis wrote:It's really intended for disabling features that would normally be there. I don't know why it would ever make sense to disable a normal function.Consider the case of `alias this` or a mixin template. You might make a wrapper type that disables a particular operation by writing ` disable void opBinary(op)` so it won't forward to the underlying thing.
Sep 13 2016
On Tuesday, September 13, 2016 17:59:09 Adam D. Ruppe via Digitalmars-d-learn wrote:On Tuesday, 13 September 2016 at 17:52:48 UTC, Jonathan M Davis wrote:Ah. That makes sense. Thanks for pointing out that use case. And actually, I think that that use case further supports the idea that what code should be testing for is whether an operation works and not whether it's disabled. In the general case, you don't even have any guarantee that the type being aliased has an operation that would need to be disabled. And from the caller's perspective, it shouldn't matter whether the + operator doesn't work because it wasn't declared or because it was disabled. - Jonathan M DavisIt's really intended for disabling features that would normally be there. I don't know why it would ever make sense to disable a normal function.Consider the case of `alias this` or a mixin template. You might make a wrapper type that disables a particular operation by writing ` disable void opBinary(op)` so it won't forward to the underlying thing.
Sep 13 2016