532 – Wrong name mangling for template alias params of local vars

In this example, in moo!(A), the value of Mang!(A).mangledname depends on whether it is inside a 'static assert' or not. This seems to be a symptom of a more general problem: I've seen many similar effects, but this is the simplest so far.
---------
template Mang(alias F)
{
 class G { }
 alias void function (G ) H;
 const char [] mangledname = typeof(H).mangleof;
}
template moo(alias A)
{
 const char [] a = Mang!(A).mangledname;
 static assert(Mang!(A).mangledname == a); // FAILS !!!
}
void main()
{
 int q;
 char [] b = moo!(q).a;
}

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
d-bugmail@puremagic.com schrieb am 2006年11月16日:
> http://d.puremagic.com/issues/show_bug.cgi?id=532 
> In this example, in moo!(A), the value of Mang!(A).mangledname depends on
> whether it is inside a 'static assert' or not. This seems to be a symptom of a
> more general problem: I've seen many similar effects, but this is the simplest
> so far.
>
> ---------
> template Mang(alias F)
> {
> class G { }
> alias void function (G ) H;
> const char [] mangledname = typeof(H).mangleof;
> }
>
> template moo(alias A)
> {
> const char [] a = Mang!(A).mangledname;
> static assert(Mang!(A).mangledname == a); // FAILS !!!
> }
>
> void main()
> {
> int q;
> char [] b = moo!(q).a;
> }
Added to DStress as
http://dstress.kuehne.cn/compile/m/mangleof_19_A.d
http://dstress.kuehne.cn/compile/m/mangleof_19_B.d
http://dstress.kuehne.cn/compile/m/mangleof_19_C.d
http://dstress.kuehne.cn/compile/m/mangleof_19_D.d
http://dstress.kuehne.cn/compile/m/mangleof_19_E.d
http://dstress.kuehne.cn/compile/m/mangleof_19_F.d
http://dstress.kuehne.cn/compile/m/mangleof_19_G.d
Thomas
-----BEGIN PGP SIGNATURE-----
iD8DBQFFaEUsLK5blCcjpWoRAlGiAKCaYCpRKsMkuhre2cU48TbQL/cH5QCfUGKS
2I+momNVe3BO1MKnC22+u+c=
=pQ4D
-----END PGP SIGNATURE-----

(In reply to comment #0)
> template Mang(alias F)
> {
> class G { }
> alias void function (G ) H;
> const char [] mangledname = typeof(H).mangleof;
> }
> 
> template moo(alias A)
> {
> const char [] a = Mang!(A).mangledname;
> static assert(Mang!(A).mangledname == a); // FAILS !!!
> }
> 
> void main()
> {
> int q;
> char [] b = moo!(q).a;
> }
Some more information....
Inside the static assert, the type is:
void function ( Mang!(bug.main().q).G );
PFC3bug29__T4MangS17_D3bug4mainFZv1qiZ1GZv
but outside the static assert, it is:
void function ( main.moo(bug.main().q)).Mang!(bug.main().q).G );
PFCmain28__T3mooS17_D3bug4mainFZv1qiZ29__T4MangS17_D3bug4mainFZv1qiZ1GZv
The problem is that in this second case, the mangled name of the calling function gets appended to the start of the fully qualified name. I can't see why this would be necessary ('q' is already fully qualified in the alias parameter, so it cannot conflict with anything else).
A secondary problem is that in the second case, extern (C),(Windows), (Pascal) symbols such 'main' are not mangled correctly with the new scheme from D0.176: it doesn't have the length information included.
main.moo!(...) is Cmain28__T3moo...
but it should be C4main28_T3moo...
This applies to any extern(C), (Windows), or (Pascal) function, not just "main",
but doesn't apply to normal D functions.
It is still a problem with D1.0.
All other name mangling bugs I've reported since D0.176 seem to be consequences of this one. The mangling used inside the static assert seems natural, and I think it should be used in the other case as well, unless there's a known problem with it.

Changing status to critical because I've seen bad code generation in library code as a result of this bug. There seems to be a fundamental instability in the compiler at this point, almost as though there's an uninitialised variable.
This issue is a blocker for my meta.NameOf code.

i change d's version of this bug. and priority to 1 , hope walter could fix this bug sooner.

(In reply to comment #4)
> i change d's version of this bug. and priority to 1 , hope walter could fix
> this bug sooner.
> 
Please do not change the version number for bug reports. See D.bugs/10071 (http://www.digitalmars.com/webnews/newsgroups.php?search_txt=&group=digitalmars.D.bugs&article_id=10071)

Fixed DMD 1.012.