Just for the record, non-null static and runtime analysis is implemented in a number of Java projects.

http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html Anotation based Static analysis for eclipse

* javax.validation.constraints.NotNull
Created for runtime validation, not static analysis.

* edu.umd.cs.findbugs.annotations.NonNull
Static analysis

* com.intellij.annotations.NotNull
Used by IntelliJ IDEA IDE for static analysis.

* lombok.NonNull
Annotation used to control code generation in Project Lombok.

Le 11/12/2011 11:29, Somedude a écrit :
> Just for the record, non-null static and runtime analysis is implemented
> in a number of Java projects.
>
> http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html
> Anotation based Static analysis for eclipse
>
> * javax.validation.constraints.NotNull
> Created for runtime validation, not static analysis.
>
> * edu.umd.cs.findbugs.annotations.NonNull
> Static analysis
>
> * com.intellij.annotations.NotNull
> Used by IntelliJ IDEA IDE for static analysis.
>
> * lombok.NonNull
> Annotation used to control code generation in Project Lombok.

We definitively should have a non nullable type in D. Null check are:
 - Easy to forget, leading to catastrophic results when forgotten.
 - Crippling the codebase at an alarming speed if you want reliable code.

Plus, when an error occurs with null, it is usually hard to track and solve, because you have to understand how that b****y variable ends up being null.

Arguably, things should be non nullable by default.

Tony Hoare (inventor of null ?) recently gave this conference : http://www.infoq.com/presentations/Null-References-The-Billion-Dollar-Mistake-Tony-Hoare

On 11-12-2011 14:18, deadalnix wrote:
> Le 11/12/2011 11:29, Somedude a écrit :
>> Just for the record, non-null static and runtime analysis is implemented
>> in a number of Java projects.
>>
>> http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html
>>
>> Anotation based Static analysis for eclipse
>>
>> * javax.validation.constraints.NotNull
>> Created for runtime validation, not static analysis.
>>
>> * edu.umd.cs.findbugs.annotations.NonNull
>> Static analysis
>>
>> * com.intellij.annotations.NotNull
>> Used by IntelliJ IDEA IDE for static analysis.
>>
>> * lombok.NonNull
>> Annotation used to control code generation in Project Lombok.
>
> We definitively should have a non nullable type in D. Null check are:
> - Easy to forget, leading to catastrophic results when forgotten.
> - Crippling the codebase at an alarming speed if you want reliable code.
>
> Plus, when an error occurs with null, it is usually hard to track and
> solve, because you have to understand how that b****y variable ends up
> being null.
>
> Arguably, things should be non nullable by default.
>
> Tony Hoare (inventor of null ?) recently gave this conference :
> http://www.infoq.com/presentations/Null-References-The-Billion-Dollar-Mistake-Tony-Hoare
>

I think the idea with D's @disable attribute on default struct contructors is to allow implementing a NoNull type. Still, I'd prefer having it in the language.

- Alex

Le 11/12/2011 14:19, Alex Rønne Petersen a écrit :
> On 11-12-2011 14:18, deadalnix wrote:
>> Le 11/12/2011 11:29, Somedude a écrit :
>>> Just for the record, non-null static and runtime analysis is implemented
>>> in a number of Java projects.
>>>
>>> http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html
>>>
>>>
>>> Anotation based Static analysis for eclipse
>>>
>>> * javax.validation.constraints.NotNull
>>> Created for runtime validation, not static analysis.
>>>
>>> * edu.umd.cs.findbugs.annotations.NonNull
>>> Static analysis
>>>
>>> * com.intellij.annotations.NotNull
>>> Used by IntelliJ IDEA IDE for static analysis.
>>>
>>> * lombok.NonNull
>>> Annotation used to control code generation in Project Lombok.
>>
>> We definitively should have a non nullable type in D. Null check are:
>> - Easy to forget, leading to catastrophic results when forgotten.
>> - Crippling the codebase at an alarming speed if you want reliable code.
>>
>> Plus, when an error occurs with null, it is usually hard to track and
>> solve, because you have to understand how that b****y variable ends up
>> being null.
>>
>> Arguably, things should be non nullable by default.
>>
>> Tony Hoare (inventor of null ?) recently gave this conference :
>> http://www.infoq.com/presentations/Null-References-The-Billion-Dollar-Mistake-Tony-Hoare
>>
>>
>
> I think the idea with D's @disable attribute on default struct
> contructors is to allow implementing a NoNull type. Still, I'd prefer
> having it in the language.
>
> - Alex

Definitively, it should be done the other way around : Non null by default, and Nullable!(T) in the standard lib.

Most of the time, you don't want null, and have to check them.

On 12/11/2011 02:19 PM, Alex Rønne Petersen wrote:
> On 11-12-2011 14:18, deadalnix wrote:
>> Le 11/12/2011 11:29, Somedude a écrit :
>>> Just for the record, non-null static and runtime analysis is implemented
>>> in a number of Java projects.
>>>
>>> http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html
>>>
>>>
>>> Anotation based Static analysis for eclipse
>>>
>>> * javax.validation.constraints.NotNull
>>> Created for runtime validation, not static analysis.
>>>
>>> * edu.umd.cs.findbugs.annotations.NonNull
>>> Static analysis
>>>
>>> * com.intellij.annotations.NotNull
>>> Used by IntelliJ IDEA IDE for static analysis.
>>>
>>> * lombok.NonNull
>>> Annotation used to control code generation in Project Lombok.
>>
>> We definitively should have a non nullable type in D. Null check are:
>> - Easy to forget, leading to catastrophic results when forgotten.
>> - Crippling the codebase at an alarming speed if you want reliable code.
>>
>> Plus, when an error occurs with null, it is usually hard to track and
>> solve, because you have to understand how that b****y variable ends up
>> being null.
>>
>> Arguably, things should be non nullable by default.
>>
>> Tony Hoare (inventor of null ?) recently gave this conference :
>> http://www.infoq.com/presentations/Null-References-The-Billion-Dollar-Mistake-Tony-Hoare
>>
>>
>
> I think the idea with D's @disable attribute on default struct
> contructors is to allow implementing a NoNull type. Still, I'd prefer
> having it in the language.
>
> - Alex

D cannot support library NonNull types because there is no way to get initialization right.

Somedude:

> Just for the record, non-null static and runtime analysis is implemented in a number of Java projects.
> 
> http://eclipseandjazz.blogspot.com/2011/12/inter-procedural-null-analysis-using.html Anotation based Static analysis for eclipse

It looks very nice.

Instead of that Java syntax:
@NonNull Object foo(@NonNull Object x) {...

For D2 I have suggested a more handy and short syntax, a trailing @ means non-null: object@ foo(object@ x) {...

A leading (usually optional, because it's the default, but in some situations you need to use it) ? means nullable:
object? foo(object? x) {...

The question mark is useful in some situations:
Foo@ foo = new Foo(); // foo is non-null
auto f = cast(?)foo; // cast to nullable

Some notes here: http://d.puremagic.com/issues/show_bug.cgi?id=4571

Bye,
bearophile