Re: AA ignores disabling of postblit for key types

[Nordlöw via Digitalmars-d]

On Friday, 21 October 2016 at 18:35:38 UTC, Jonathan M Davis 
wrote:
On Friday, October 21, 2016 12:18:28 Nordlöw via Digitalmars-d 
wrote:
It seems AA's doesn't respect disabling of postblit for its 
Key-type even when it's sent as an r-value.


This is a serious bug.


Well, I wouldn't expect it to work to use a non-copyable type 
as a key, but that should just result in a compilation error, 
not the mess that you're seeing. It wouldn't surprise me at all 
if the problem relates to the fact that the AA implementation 
uses void* internally (at least, from what I recall, that's 
what it does, which historically has meant a number of fun 
type-related bugs).


- Jonathan M Davis


AFACT, it *should* be able to work with non-copyable types if 
argument is an r-value. If somebody points out where to start 
digging for this DMD I can give it at try.


1. issue a error when index is an l-value of noncopyable type
2. move the index if it's an r-value

Re: AA ignores disabling of postblit for key types

[Jonathan M Davis via Digitalmars-d]

On Friday, October 21, 2016 12:18:28 Nordlöw via Digitalmars-d wrote:
> It seems AA's doesn't respect disabling of postblit for its
> Key-type even when it's sent as an r-value.
>
> This is a serious bug.

Well, I wouldn't expect it to work to use a non-copyable type as a key, but
that should just result in a compilation error, not the mess that you're
seeing. It wouldn't surprise me at all if the problem relates to the fact
that the AA implementation uses void* internally (at least, from what I
recall, that's what it does, which historically has meant a number of fun
type-related bugs).

- Jonathan M Davis

Re: AA ignores disabling of postblit for key types

[Nordlöw via Digitalmars-d]

On Friday, 21 October 2016 at 12:18:28 UTC, Nordlöw wrote:

this(this) { assert(false); }


If this is changed to


  @disable this(this);


I instead get a segfault because of a double free:


before
freeing:591490
after
freeing:591490
*** Error in ...

Pretty serious bug.

AA ignores disabling of postblit for key types

[Nordlöw via Digitalmars-d]

The following code


import std.stdio;

struct S(E)
{
static typeof(this) withElement(E x)
{
typeof(return) that;
that._ptr = cast(E*)malloc(1*E.sizeof);
*(that._ptr) = x;
return that;
}

// @disable this(this);
this(this) { assert(false); }

~this()
{
writeln("freeing:", _ptr);
free(_ptr);
}

E* _ptr;
}

unittest
{
alias Key = S!int;
int[Key] x;
writeln("before");
x[Key.withElement(11)] = 42;
writeln("after");
}

extern (C):
void* malloc(size_t);
void free(void*);


assert as


before
freeing:706490
core.exception.AssertError@aaNoMoveCrash.d(18): Assertion failure

??:? [0x411a3f]
??:? [0x41028f]
aaNoMoveCrash.d:18 [0x402b73]
??:? [0x413be5]
??:? [0x4126a8]
aaNoMoveCrash.d:34 [0x402a46]
??:? [0x410230]
??:? [0x41c8a0]
??:? [0x4113cb]
??:? [0x4162e6]
??:? [0x416374]
??:? [0x416277]
??:? [0x4113a7]
??:? [0x41c792]
??:? [0x412dfa]
??:? [0x412d98]
??:? [0x412d12]
??:? [0x4103bf]
??:? __libc_start_main [0x8452a82f]
freeing:706490


It seems AA's doesn't respect disabling of postblit for its 
Key-type even when it's sent as an r-value.


This is a serious bug.

Is this know?

Can somebody give hints to where I can start digging in DMD for 
fixing this?