Michael Van Canneyt wrote:

On Tue, 21 Sep 2010, Adriaan van Os wrote:

Michael Van Canneyt wrote:

On Tue, 21 Sep 2010, Florian Klaempfl wrote:

Am 21.09.2010 13:42, schrieb Adriaan van Os:
Michael Van Canneyt wrote:


On Tue, 21 Sep 2010, Adriaan van Os wrote:

I am looking for a way to obtain the class hierarchy of a program in
MacPas mode (where there are no published properties). TObject has
ClassParent, so a list of Class types would be sufficient, I think. I
can't find however a routine in the RTL to obtain such a list.

There is no such routine in the RTL, to my knowledge.

Can I add it ? I guess a low-level pointer to a list of VMT's would be
sufficient to implement it ?

Problem is: it breaks smartlinking.

Yes, I understand that. I looked at the assembly generated for a simple test program and there are separate al_globals entries for each vmt and separate al_rtti entries for the vmt rtti. I asssume the very reason for the separate entries is smart linking.

Hm, I guess I can look at the data section of the executable but it would be highly dependant of the executable binary format.

Is it possible to produce a separate data section/entry that refers to all al_globals and/or all al_rtti entries without breaking smart-linking ?

Maybe when doing whole-program optimization as implemented by Jonas.
I guess the WPO could be extended to generate the tree you want.

But under 'normal' compilation, I don't think it is possible to do as you
ask without breaking the smartlinking stuff; the tree would of necessity
contain all declared classes of all loaded units.

I implemented it now by looking at the Mach-O symbol table. It works like a charm (only classes that are actually used are listed and it doesn't break smart-linking).

If someone feels like implementing this for other binary formats (with the purpose of adding it to the rtl), I can donate the Mach-O source code under the usual license.

Regards,

Adriaan van Os

_______________________________________________
fpc-pascal maillist  -  fpc-pascal@lists.freepascal.org
http://lists.freepascal.org/mailman/listinfo/fpc-pascal

Reply via email to