www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - invariant constancy: revisited

reply "Jakob Ovrum" <jakobovrum gmail.com> writes:
Here we go again.

Back in 2012, invariants were misguidedly made const by default 
without going through any review whatsoever, breaking a ton of 
code in the process[1]. That means bitwise constancy is enforced 
for the this-reference in the body of the invariant. There is no 
way to opt out of this.

The well-intentioned argument is that invariants should not have 
side effects, which I wholeheartedly agree with. The problem is 
measuring side effects in terms of bitwise constancy and not 
logical constancy. I was using logically constant invariants at 
the time, but had to abandon them because of this change (again, 
no way to opt out). Also, debug code inside invariants is also 
hampered by bitwise const.

Pull requests were since made to revert the change, while 
allowing invariants to be type-qualified explicitly to enforce 
bitwise-constancy[1][2]. They eventually rot away, and now two 
years later, invariants are still artificially limited by bitwise 
const; as I had stopped using invariants, I never noticed.

I think we're running out of time to fix this, if we haven't 
already. Specifically, the reversion was never commented on by 
Walter, though Andrei did seem supportive. Please approve the 
change. Hopefully this is the time to reclaim invariants ;)

[1] 
https://github.com/D-Programming-Language/dmd/commit/3c138718365541cde67abbdcfc031d755c3a7969
[2] https://github.com/D-Programming-Language/dmd/pull/1073
[3] https://github.com/D-Programming-Language/dmd/pull/1560
Mar 14 2014
parent reply Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
On 3/14/14, Jakob Ovrum <jakobovrum gmail.com> wrote:

 Back in 2012, invariants were misguidedly made const by default
 without going through any review whatsoever, breaking a ton of
 code in the process[1].


How about a compromise based on Andrei's new final(bool) suggestion?
We could implement const(false) to allow overriding the default
behavior of invariants. E.g.:

class C
{
    invariant()  // const-by-default, as usual
    {
    }

    invariant() const(false) // new feature
    {
    }
}

That way existing code stays safe.
Mar 15 2014
parent reply "Daniel Murphy" <yebbliesnospam gmail.com> writes:
"Andrej Mitrovic"  wrote in message 
news:mailman.134.1394883267.23258.digitalmars-d puremagic.com...


 How about a compromise based on Andrei's new final(bool) suggestion?
 We could implement const(false) to allow overriding the default
 behavior of invariants. E.g.:

 class C
 {
     invariant()  // const-by-default, as usual
     {
     }

     invariant() const(false) // new feature
     {
     }
 }

 That way existing code stays safe.


Yuck.  We can always warn for a while on invariants not explicitly marked as 
const, if we're worried about breakage.

You should be able to use most of the language without using const, making 
invariants implicitly const was a bad idea.
Mar 15 2014
next sibling parent "Jakob Ovrum" <jakobovrum gmail.com> writes:
On Saturday, 15 March 2014 at 14:46:13 UTC, Daniel Murphy wrote:

 "Andrej Mitrovic"  wrote in message 
 news:mailman.134.1394883267.23258.digitalmars-d puremagic.com...


 How about a compromise based on Andrei's new final(bool) 
 suggestion?
 We could implement const(false) to allow overriding the default
 behavior of invariants. E.g.:

 class C
 {
    invariant()  // const-by-default, as usual
    {
    }

    invariant() const(false) // new feature
    {
    }
 }

 That way existing code stays safe.


 Yuck.  We can always warn for a while on invariants not 
 explicitly marked as const, if we're worried about breakage.

 You should be able to use most of the language without using 
 const, making invariants implicitly const was a bad idea.


I agree. Implicit const was only in play for a few releases; a 
couple of releases of deprecation time should be plenty, 
especially considering the number of users who have been unable 
to use invariants in those releases.
Mar 15 2014
prev sibling parent reply Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
On 3/15/14, Daniel Murphy <yebbliesnospam gmail.com> wrote:

 Yuck.  We can always warn for a while on invariants not explicitly marked as

 const, if we're worried about breakage.

 You should be able to use most of the language without using const, making
 invariants implicitly const was a bad idea.


Well someone has to convince Dear Leader(s) of this. The first pull[1]
was made 2 years ago (!).

[1] : https://github.com/D-Programming-Language/dmd/pull/1073
Mar 15 2014
parent reply "Daniel Murphy" <yebbliesnospam gmail.com> writes:
"Andrej Mitrovic"  wrote in message 
news:mailman.136.1394896688.23258.digitalmars-d puremagic.com...


 Well someone has to convince Dear Leader(s) of this. The first pull[1]
 was made 2 years ago (!).

 [1] : https://github.com/D-Programming-Language/dmd/pull/1073


Andrei's comment there sounds like he's on board.
Mar 15 2014
parent Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
On 3/15/14, Daniel Murphy <yebbliesnospam gmail.com> wrote:

 "Andrej Mitrovic"  wrote in message
 news:mailman.136.1394896688.23258.digitalmars-d puremagic.com...


 Well someone has to convince Dear Leader(s) of this. The first pull[1]
 was made 2 years ago (!).

 [1] : https://github.com/D-Programming-Language/dmd/pull/1073


 Andrei's comment there sounds like he's on board.


Heheh, and Walter was on board with final-by-default!
Mar 15 2014