instance initializer

[Vicente Romero]

On 9/5/19 6:14 PM, forax at univ-mlv.fr wrote:
>
>
> ------------------------------------------------------------------------
>
>     *De: *"Brian Goetz" <brian.goetz at oracle.com>
>     *À: *"Tagir Valeev" <amaembo at gmail.com>, "Remi Forax"
>     <forax at univ-mlv.fr>
>     *Cc: *"amber-spec-experts" <amber-spec-experts at openjdk.java.net>
>     *Envoyé: *Jeudi 5 Septembre 2019 23:26:34
>     *Objet: *Re: Draft JLS spec for records
>
>
>
>         I don't think so. A compact constructor (or require
>         initializer, as you propose) could be not the only
>         constructor. An instance initializer is convenient because
>         it's added to every constructor, regardless of whether it's
>         compact or not. So the new thing doesn't supersede the
>         instance initializer and I see no good reason to explicitly
>         disable it.
>
>
>     Vicente offered another reason why we might want to prohibit the
>     instance initializer, if only out of expediency; it complicates
>     the analysis of which fields are DA on all paths through the
>     constructor (and therefore, do not need to be automatically
>     initialized.)  If you have a record:
>
>         record Foo(int i) {
>             { this.i = 0; }
>         }
>
>     then the canonical constructor has to see that `i` is always
>     initialized by the static init, and therefore should be not
>     initialized.  Worse, if we have:
>
>         record Foo(int i) {
>             { if (tuesday) this.i = 0; }
>         }
>
>     then we have to issue a compilation error, since we have fields
>     that are neither DA nor DU at the end of the initializer.
>
>     None of this is impossible to do, of course; it's just not clear
>     whether it's worth it, given the limited utility of instance
>     initializers in records (because we've already banned instance
>     fields.)
>
>
> If we want to fix that with an the analysis, it will have to work on 
> user-written code mixed with generated code,
> it doesn't seem to be a good idea.

right I was thinking about the case:

record R(int i, int j) {
     public R {  // compact constructor
         if (i < 0) {
             this.i = -i;
         } else {
             this.j = j;
         }
     }
}

which doesn't have instance initializer blocks. The thing is where to 
draw between cases where the compiler will just issue an error and cases 
in which it will generate the missing initializations. In this 
particular case the compiler will have to be very clever and generate:

public R {  // compact constructor
         if (i < 0) {
             this.i = -i;
             this.j = j;  // automatic code
         } else {
             this.i = i; // automatic code
             this.j = j;
         }
     }

we can do it, it will be complex though, but do we want to do it? Where 
to draw the line?

Thanks,
Vicente
>
> so it's Ok to not support instance initializers in record.
>
> Rémi
>
>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/amber-spec-experts/attachments/20190905/cc430b49/attachment.html>