fix Issue 20691 - Converting scope static array to scope dynamic arra…

[WalterBright]

…y should be error

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	20691	normal	Converting scope static array to scope dynamic array should be error

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#10951"

[Geod24]

Added a link in 20691 to 20505 (why the duplication ??).
So do I understand correctly that the reasoning is that there can only be one level of scope-ness ?

[WalterBright]

why the duplication ??

I originally thought it was a different problem.

the reasoning is that there can only be one level of scope-ness ?

Yes. Scope is not transitive.

[WalterBright]

Unfortunately, several projects in buildkite are relying on this compiler bug.

[Geod24]

Yes. Scope is not transitive.

Then why do struct allow one to store a scope slice in them ?

Also, can you renamed the commit title to Fix issue 20505 - [...], since 20691 was a duplicate ?

[Geod24]

Unfortunately, several projects in buildkite are relying on this compiler bug.

We discussed this a million time, but I'm going to repeat it anyway: anyone trying to do something useful with scope (that is, past identity functions) expect it to be transitive.
If scope is not transitive, then how do we deal with struct storing slices ?

I tested this PR with some sample code, and this still compiles:

import std.stdio;

struct Bar
{
    const(char)[] v;
}

struct Store
{
    Bar[] v;
}

void main () @safe
{
    auto v = func1();
    writeln(v);
}

const(char)[] func1 () @safe
{
    char[10] str = 'a';
    return func2(str);
}

const(char)[] func2 (scope const(char)[] b) @safe
{
    Bar[10] array = Bar(b);
    return func3(array, b);
}

const(char)[] func3 (scope Bar[] array, scope const(char)[] b) @safe
{
    Store st = Store(array);
    return st.v[0].v;
}

[Geod24]

Any update on this ?

[WalterBright]

@Geod24 this still compiles

please add bug report.

[Geod24]

please add bug report.

I did, it's the one you are trying to fix (https://issues.dlang.org/show_bug.cgi?id=20505).

[WalterBright]

@Geod24 as for transitive scope, while a good idea, that would be a fairly major project, way beyond the scope of this PR. It is likely to also break much existing code.

[Geod24]

@WalterBright :

1. I agree with you, having transitive scope would be a fairly major undertaking, and TBH I'm not sure it would allow us to have @safe RC either;

2. By definition, code that is triggered by a -preview is experimental. So at the moment, whatever change we decide to go with, as long as it only triggers when -preview=dip1000 is used, it's not a breaking change;

I'll merge the druntime PR in the meantime.

[dkorpel]

Now that the blocker from Phobos has been fixed, can this be merged? I assume the failing test runs are old and should go away after a rebase.

[dkorpel]

@WalterBright @RazvanN7 Can you rerun the tests?

[dkorpel]

What's this for?

[MoonlightSentinel]

Maybe a remnant of the old bug that caused enclosesLifetimeOf to return wrong results for VarDeclarations created during semantic?

[dkorpel]

Ping @WalterBright @RazvanN7

[WalterBright]

Looks like the problem is in phobos:

std/process.d(1960): Error: cannot take address of `scope` local `args` in `@safe` function `spawnShell`
std/process.d(1960): Error: cannot cast expression `args` of type `const(char)[][3]` to `const(char[])[]`

[dkorpel]

Looks like the problem is in phobos:

So the error is still there. I thought it would be fixed by dlang/phobos#8119

[thewilsonator]

Removing auto merge this has failures all across build kite

[dkorpel]

The buildkite failures were in projects not using dip1000, so I added an if (global.params.useDIP1000 == FeatureState.enabled) and it's green now.