https://issues.dlang.org/show_bug.cgi?id=20674

Walter Bright <bugzilla@digitalmars.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bugzilla@digitalmars.com
           Severity|normal                      |enhancement

--- Comment #1 from Walter Bright <bugzilla@digitalmars.com> ---
To do what this example asks for, data flow analysis would be required. This can be done, but is a large increase in implementation complexity and time to compile. It is not a bug that DFA is not currently done.

Marked as an enhancement.

--

https://issues.dlang.org/show_bug.cgi?id=20674

ag0aep6g <ag0aep6g@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://issues.dlang.org/sh
                   |                            |ow_bug.cgi?id=21209

--

https://issues.dlang.org/show_bug.cgi?id=20674

Paul Backus <snarwin+bugzilla@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |snarwin+bugzilla@gmail.com

--- Comment #2 from Paul Backus <snarwin+bugzilla@gmail.com> ---
Note that this can be worked around by explicitly declaring the local variable in the second example `scope`:

---
int* f()(int* p)
{
    static int g;
    g = 0;
    scope p2 = p; /* the only difference */
    return new int;
}

int* g() @safe
{
    int x;
    return f(&x); /* ok */
}
---

--

https://issues.dlang.org/show_bug.cgi?id=20674

Dennis <dkorpel@live.nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dkorpel@live.nl

--- Comment #3 from Dennis <dkorpel@live.nl> ---
(In reply to Walter Bright from comment #1)
> To do what this example asks for, data flow analysis would be required. This can be done, but is a large increase in implementation complexity and time to compile.

This puzzles me. The examples don't contain a single branch or loop. Surely you can infer scope here without creating a control-flow graph, right? Also, while I don't know exactly how the current inference works (it's not documented anywhere), it looks like it has a pretty complex loop already:

https://github.com/dlang/dmd/blob/8884ad0d232d300b424c025cf65ecd2623ec8df5/src/dmd/escape.d#L2176

--

https://issues.dlang.org/show_bug.cgi?id=20674

--- Comment #4 from Paul Backus <snarwin+bugzilla@gmail.com> ---
(In reply to Dennis from comment #3)
> (In reply to Walter Bright from comment #1)
> > To do what this example asks for, data flow analysis would be required. This can be done, but is a large increase in implementation complexity and time to compile.
> 
> This puzzles me. The examples don't contain a single branch or loop. Surely you can infer scope here without creating a control-flow graph, right?

Currently, in order to infer scope for `p`, you only need to check each usage of `p` for compliance with scope.

To make the example (and others like it) work, you would need to recursively check each usage of any variable that `p` might be assigned to (like `p2`), and each usage of any variable that any of *those* variables might be assigned to, and so on, until you cannot find any new variables to check.

What I've just described is a graph search problem. Whether the compiler actually builds an explicit control-flow graph or not, it amounts to the same thing.

--

https://issues.dlang.org/show_bug.cgi?id=20674

--- Comment #5 from Walter Bright <bugzilla@digitalmars.com> ---
(In reply to Dennis from comment #3)
> Surely you can infer scope here without creating a control-flow graph, right?

The trouble is, once you start doing a lame DFA, then people want a real DFA. Doing DFA in the front end will significantly slow it down.

--

https://issues.dlang.org/show_bug.cgi?id=20674

Ate Eskola <Ajieskola@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Ajieskola@gmail.com

--

https://issues.dlang.org/show_bug.cgi?id=20674

timon.gehr@gmx.ch changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |timon.gehr@gmx.ch

--- Comment #6 from timon.gehr@gmx.ch ---
AFAIU DIP1000 just copies the lifetime attribute from the initializer to the variable, so I don't think you need anything more fancy than computing connected components.

--

https://issues.dlang.org/show_bug.cgi?id=20674

Iain Buclaw <ibuclaw@gdcproject.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P4

--

https://issues.dlang.org/show_bug.cgi?id=20674

Dlang Bot <dlang-bot@dlang.rocks> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |pull

--- Comment #7 from Dlang Bot <dlang-bot@dlang.rocks> ---
@dkorpel updated dlang/dmd pull request #14492 "Fix 20674, 23208, 23300 - improve `scope` inference" fixing this issue:

- Fix 20674, 23208, 23300, 23294 - improve `scope` inference

https://github.com/dlang/dmd/pull/14492

--