On Fri, 2018-06-01 at 16:19 -0400, Steven Schveighoffer via Digitalmars-d-learn wrote:
> On 6/1/18 1:41 PM, Russel Winder wrote:
> > struct Datum {
> > 	public const int a;
> > 	public const int b;
> > }
> > 
> > struct Message {
> > 	Datum datum;
> > }
> 
> I found the bug. Basically, the Variant static if is failing because
> the
> assignment it is checking (assigning a Message to a Message) would
> overwrite const data. But it's not really overwriting existing data,
> it's emplacing into new data (always). So this check is not correct.

I think then my bug report 18934 is slightly wrong and it isn't the struct within struct that is the problem, it is the const fields in a struct that is?

> Much simpler test case:
> 
> struct S
> {
>     const int x;
> }
> 
> void main()
> {
>     import std.variant;
>     import std.stdio;
>     Variant v = S(1);
>     writeln(v.get!S); // same failure
> }
> 

Aha, much nicer since it is far more localised to the issue. I was still focused on the receive and the struct within struct, which seems to be the wrong issue: it is the const fields that are the problem. Feel free to ignore my example and stack trace on

https://issues.dlang.org/show_bug.cgi?id=18934

and replace it with the above!

> If I replace the check in std.variant:
> 
>                  static if (is(typeof(*cast(T*) target = *src)) ||
> 
> with:
> 
>                  static if (is(typeof(delegate T() {return *src;}))
> ||
> 
> Then it works (both your code, and the simple example).
> 
> Note that in all cases, variant is returning a COPY of the data, not
> a
> reference, so it shouldn't be possible to violate const.

Excellently done. Thanks for attacking this problem and finding a solution. I will not now even contemplate hacking up my code. Actually I would have put const in and still got a problem it seems, as I had the wrong reason for the problem.

> Please, file a bug. I will see if I can submit a PR to fix.
> 

Bug report is 18934.

If there can be a fast release of this via 2.080.1 with subsequence fast releases via d-apt and Debian and Fedora packaging, I would be a very happy bunny.

In the meantime onward with ACCU 2019 organisation.

ACCU needs more D content. Yes it is 50% C++, but that is exactly why it needs D content. It also needs Go content. in 2018 we had Rust content and it went down well.

-- 
Russel.
===========================================
Dr Russel Winder      t: +44 20 7585 2200
41 Buckmaster Road    m: +44 7770 465 077
London SW11 1EN, UK   w: www.russel.org.uk

On Saturday, 2 June 2018 at 09:52:59 UTC, Russel Winder wrote:
> I get ldc2 from the Debian/Fedora repositories and dmd from d-apt (no Fedora equivalent) I really want to avoid fiddling with files that are installed via packaging. Though I guess any changes can be fixed by a reinstallation of the package.

Editing in place would be the easiest and shouldn't be a problem, but at least ldc is relocatable: import folder is specified in config, so you can have it anywhere.

For anyone still interested in this problem: Steve's pull request fixing the problem is included in D 2.081.0.

I can experiment as soon as DMD 2.081.0 is released but will have to wait till the fixes get into LDC to create production code.

Moral of the story: silent, except during gdb debugging, assertion fails in
Phobos are a pain in the arse.

-- 
Russel.
===========================================
Dr Russel Winder      t: +44 20 7585 2200
41 Buckmaster Road    m: +44 7770 465 077
London SW11 1EN, UK   w: www.russel.org.uk