Test case:
---------------------------
import std.c.stdio;
int a() {
return V.s(0);
}
struct V {
static pure auto s(int x) {
return 55;
}
}
void main() {
printf("%d\n", a());
}
---------------------------
Bus error
---------------------------
The bug appears only if all of the following conditions are met:
1. The function 's' must be 'pure'.
2. The function 's' must have auto-return.
3. The function 's' must have at least one named parameter. For example, if
the parameter is changed to 'pure auto s(...)' the bug disappears.
4. The function 's' must be placed after its first use. For example, moving
'struct V' above 'int a()' the bug disappears.
The program works correctly in 2.048. Therefore I have marked this as a
regression.
============================
If instead of calling the function, we just compute the pointer of 's', we'll
get a forward reference error instead:
---------------------------
void a() {
cast(void)(&V.s);
}
struct V {
static pure auto s(int x) {
return 0;
}
}
---------------------------
x.d(2): Error: forward reference to s
---------------------------
But this error also happens on 2.048.
Comment #1 by clugdbug — 2011-06-06T06:18:17Z
The segfault is a simple null pointer dereference which can easily be patched.
mtype.c, TypeFunction::parameterEscapes, line 5369
if (purity)
{ /* With pure functions, we need only be concerned if p escapes
* via any return statement.
*/
- Type* tret = nextOf()->toBasetype();
- if (!isref && !tret->hasPointers())
+ Type* tret = nextOf() ? nextOf()->toBasetype() : NULL;
+ if (!isref && tret && !tret->hasPointers())
{ /* The result has no references, so p could not be escaping
* that way.
*/
return FALSE;
}
Comment #2 by clugdbug — 2011-06-06T06:21:15Z
*** Issue 5844 has been marked as a duplicate of this issue. ***
Comment #3 by kennytm — 2011-06-30T11:27:04Z
(In reply to comment #1)
> The segfault is a simple null pointer dereference which can easily be patched.
>
> mtype.c, TypeFunction::parameterEscapes, line 5369
>
> if (purity)
> { /* With pure functions, we need only be concerned if p escapes
> * via any return statement.
> */
> - Type* tret = nextOf()->toBasetype();
> - if (!isref && !tret->hasPointers())
> + Type* tret = nextOf() ? nextOf()->toBasetype() : NULL;
> + if (!isref && tret && !tret->hasPointers())
> { /* The result has no references, so p could not be escaping
> * that way.
> */
> return FALSE;
> }
This patch only fixes 5844. In git master this becomes
Internal error: toir.c 190
with or without your patch.
Comment #4 by kennytm — 2011-07-27T11:48:10Z
(In reply to comment #3)
> (In reply to comment #1)
> This patch only fixes 5844. In git master this becomes
>
> Internal error: toir.c 190
>
> with or without your patch.
As of the latest git master the ICE no longer happens and the test cases now behaves normally.
Also, one more test case which involves CTFE, based on std.complex.complex:
---------------------------
auto bug5936c(R)(R i) @safe pure nothrow {
return true;
}
static assert( bug5936c(0) );
---------------------------