Small amount of escape analysis to allow more @nogc functions

[dlangBugzillaToGithub]

bearophile_hugs reported this on 2014-05-26T21:53:34Z

Transferred from https://issues.dlang.org/show_bug.cgi?id=12808

CC List

• Stanislav Blinov

Description

DMD 2.066alpha accepts this code:

// Program#1
void main() @nogc {
    int[2] tmp = [1, 2];
    foreach (x; tmp) {}
}


While it refuses this code:

// Program#2
void main() @nogc {
    foreach (x; [1, 2]) {}
}


With:

test.d(3,17): Error: array literal in @nogc function main may cause GC allocation

I suggest to start introducing a small amount of Escape Analysis in D, to support the @nogc attribute for the Program#2.


Eventually even this program could be supported:

// Program#3
void main() @nogc {
    import std.algorithm: filter;
    foreach (x; [1, 2].filter!(x => true)) {}
}



Note that with the []s suffix syntax for fixed-size arrays there is no ambiguity:


// Program#4
void main() @nogc {
    foreach (x; [1, 2]s) {}
}



And this can generate a clean error message (escape of pointer to stack frame fixed-size array):


// Program#5
int[] foo() @nogc {
    return [1, 2]s;
}
void main() {}


See also Issue 10242

[dlangBugzillaToGithub]

stanislav.blinov commented on 2021-12-08T16:23:42Z

Rewritten relevant parts into contemprary D. With dmd 2.098:

void program2() @nogc {
    foreach (x; [1, 2]) {}
}

import std.array : staticArray;

void program3() @nogc {
    import std.algorithm: filter;
    foreach (x; [1, 2].filter!(x => true)) {} // fails to infer @nogc
    // workaround:
    foreach (x; [1, 2].staticArray[].filter!(x => true)) {}
}

int[] program5() @nogc {
    return [1, 2].staticArray; // detects the escape correctly
}
void main() {}

So I guess the only outstanding enhancement here is to make this infer @nogc:

foreach (x; [1, 2].filter!(x => true)) {}

...if at all possible. If not, I guess this could be closed?