• Thomas Kuehne (55/59) Nov 08 2006 -----BEGIN PGP SIGNED MESSAGE-----
• Don Clugston (51/119) Nov 08 2006 Here's another irreversible demangling.

Thomas Kuehne <thomas-dloop kuehne.cn> writes:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


D's current name mangling achieves the primary target:
 Make it possible for linkers to distinguish between different
 versions of overloaded functions.

However it fails to address the consistency issue:
 Make it possible for linkers to check that objects and
 functions are declared in exactly the same way in all modules.

The problem is that it isn't reversible,
causing all kinds of problems for runtime reflections.


issue 1: missing "static" information






_D8mangling1C3fooFZv
_D8mangling1C3barFZv


issue 2: missing "struct"/"union" information







_init_8mangling3Cat
_init_8mangling3Cow
_D8mangling3fooFZS8mangling3Cat
_D8mangling3barFZS8mangling3Cow


issue 3: missing "const" information




_D8mangling1Xi
_D8mangling1Yi


issue 4: missing protection attributes

While the ELF object format used by Linux supports differentiation

"package" and "protected" can't be differentiated.

issue 5: missing "final" information






_D8mangling1C3fooFZv
_D8mangling1C3barFZv


Especially the "static" and "struct"/"enum" issues pose a problem for
runtime reflection due to their effects on calling conventions.

Thomas


-----BEGIN PGP SIGNATURE-----

iD8DBQFFUndcLK5blCcjpWoRArDfAJ48mM7rnA5kMOTQ2hNa/fAcK2wrNwCbB4C8
8oy2C9UVZeb0Hl4vaeKoKdI=
=SqZb
-----END PGP SIGNATURE-----

Don Clugston <dac nospam.com.au> writes:

Thomas Kuehne wrote:

 D's current name mangling achieves the primary target:
 Make it possible for linkers to distinguish between different
 versions of overloaded functions.

 
 However it fails to address the consistency issue:
 Make it possible for linkers to check that objects and
 functions are declared in exactly the same way in all modules.

 
 The problem is that it isn't reversible,
 causing all kinds of problems for runtime reflections.
 
 
 issue 1: missing "static" information






 _D8mangling1C3fooFZv
 _D8mangling1C3barFZv
 
 
 issue 2: missing "struct"/"union" information






 
 _init_8mangling3Cat
 _init_8mangling3Cow
 _D8mangling3fooFZS8mangling3Cat
 _D8mangling3barFZS8mangling3Cow
 
 
 issue 3: missing "const" information




 _D8mangling1Xi
 _D8mangling1Yi
 
 
 issue 4: missing protection attributes
 
 While the ELF object format used by Linux supports differentiation

 "package" and "protected" can't be differentiated.
 
 issue 5: missing "final" information






 _D8mangling1C3fooFZv
 _D8mangling1C3barFZv
 
 
 Especially the "static" and "struct"/"enum" issues pose a problem for
 runtime reflection due to their effects on calling conventions.
 
 Thomas

Here's another irreversible demangling.

Normally, anything declared as "extern (Windows)"/extern(C)/... has no 
type information in its name mangling; only the undecorated name is 
used. Yet nested classes, types and functions inside such an extern(XXX) 
function have decorated names. This is a very obscure situation because 
normally, such entities won't have external linkage. It might be more 
important now that local variables are allowed as template alias parameters.

In example 1 below, the mangled name of fox is:
Cwolf3fox. Note that there's no length before 'wolf' (it's not 4wolf).
Example 2 shows a (highly contrived) situation where two distinct 
classes have the same
mangled name "Cwolf5dingo3fox".
If length information were included, the mangled names would be 
"C4wolf5dingo3fox"
and "C10wolf5dingo3fox".
In example 3 (which has no extern(Windows)), the mangled name is 
"C4wolf5dingo3fox".

---
Example 1.
extern(Windows) {

void wolf()
{
	class fox {}
	pragma(msg, fox.mangleof);
}

}
---
Example 2.

extern(Windows) {
void wolf()
{
	class dingo {
		class fox {}
		pragma(msg, fox.mangleof);
	}
}
void wolf5dingo()
{
	class fox {}
	pragma(msg, fox.mangleof);
}

}
---
Example 3: file called 'wolf.d'.

module wolf;

class dingo {
	class fox {}
	pragma(msg, fox.mangleof);
}

=================