Invalid method called - reduced test case

[John Reimer]

Thomas Kuehne wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Tom S schrieb am 2006-03-03:
>> DMD seems to be picky about the order of files passed to it. The 
>> attached program is a reduced version of what Ant encountered while 
>> working on Duit. I believe this is the same bug that Tyler reported 
>> regarding DWT.
>> The .zip archive contains 3 D modules: a.d, b.d and main.d
>>
>> When compiled using:
>> dmd b.d a.d main.d
>>
>> ...the resulting executable will be buggy. More specifically, an 
>> assertion (that I put in a function which should never be called) will 
>> fail :)
>>
>> The behavior is the same if one builds a .lib or .a file using
>> dmd -c b.d a.d
>>
>> ...and then links to main.d
>>
>> Tested on Windows and Linux using the #D compiler farm ;)
> 
> Your test case seems to be broken:
> 
> === a.d ===
> private import b;
> 
> class Foo : Bar {
> 	void foo() {
> 		assert(false);	// this should never be called !
> 	}
> }
> 
> === b.d ===
> class Bar {
> 	private import a;
> 
> 	void bar() {
> 		printf("Bar.bar()\n");
> 	}
> }
> 
> === main.d ===
> private import a;
> 
> void main() {
> 	(new Foo()).bar();
> }
> 
> 
> You call Foo.bar in main - that's exactly the function you didn't expect
> to execute.
> 
> The real bug is: the private of "private import a;" isn't enforced and thus
> interpreted as "import a;" - the protection attributes are known not to
> work properly in many situations.
> 
> Thomas
> 
> 
> -----BEGIN PGP SIGNATURE-----
> 
> iD8DBQFEDASk3w+/yD4P9tIRAljXAKDCRA+aN1NNiC4cgMyX8WNyXjYjkQCgwViR
> rw5WgcU4IBE4e64MtO1zzUA=
> =NlAj
> -----END PGP SIGNATURE-----


How is it broken?

I fail to understand, then, why Foo.foo() is called when a call to 
Foo.bar() is made?  Their function signatures are quite different, no? 
Are the vtables getting reassigned from the compile/link order?

Change the compile order with this:

dmd main.d a.d b.d

and Foo.bar() is called as expected.  This seems to be a compile 
order-related issue.  Why does this work just because the objects are 
built in a different order?

In what situation would we see main.d /not/ put first in the build list?

This does appear to be a dangerous problem, nonetheless...

-JJR