[fpc-devel] properties with getter evaluation in gdb

Jonas Maebe jonas.maebe at elis.ugent.be
Fri Mar 12 17:55:38 CET 2010


On 12 Mar 2010, at 17:51, Paul Ishenin wrote:

> 12.03.2010 21:55, Jonas Maebe wrote:
>> 
>> PS: one reason that this is not advancing is because FPC does not support it correctly either, and I don't want to add support for a broken implementation to GDB since then it has to be changed again later anyway. I also don't know of any description of the Borland fastcall calling convention, so it is currently not possible to properly fix this (not without guessing and probably breaking it again in yet some other case). See http://bugs.freepascal.org/view.php?id=15691
> What exactly needs to be known?

The exact rules. Partial descriptions are not useful, because we already have those and that's how the current implementation was done. For example, rules that explain where the hidden function result parameter goes (like in the bug report above), what the relative order of self parameter, vmt parameter and hidden function result are, ...

But as far as I know, there is no real description anywhere. Furthermore, unless someone creates an extensive test full of assembler routines that check whether the correct parameters are in the correct registers or stack locations, this functionality cannot be tested and every change can easily break something else without anyone noticing immediately (since the compiler itself will get the values from whichever register it passes them in).


Jonas


More information about the fpc-devel mailing list