Problem with real parameters in overloaded procs: When defining an operator P with a CURRENCY parameter, and overloading that operator with a DOUBLE parameter, the compiler cannot decide which one to take when it is called with an EXTENDED actual parameter. The same seems to be true for overloaded procs/funcs.
I suggest the following change: the sort order for searching the appropriate routine should be done as follows: 1. exact match (as it is done now); 2. extending the precision upward; 3. shrinking the precision downward. This search for a "similar" real parameter should only be done for SINGLE/DOUBLE/EXTENDED; COMP and CURRENCY are too special. (I know that formally COMP and CURRENCY are reals; but I consider them too special for automatic conversions). The neccessary code can be found in the appendices. I'm sorry, I have my problems with SVN, I tried to download with "svn -co", but in the middle somewhere (in the directory tests) it stopped (it didn't really stop, the internet connection was busy, but no more files were transmitted). After that "svn -cleanup" didn't even work, the only thing that did work was to throw away the subdirectory and retry. But the retry came to the same result, and after the 3rd time I gave up. So the code comes in pieces in the appendix. "" is the is_real function in defutils.pas "" is the new function in htypechk.pas "" is the changed function in htypechk.pas; the old function shows the parameters in the debug info in reverse order Would be nice if this could be incorporated in the compiler. Gerhard
tcallcandidates.get_information_in_htypechk.pas
Description: Binary data
is_real_in_defutil.pas
Description: Binary data
paratreestr_in_htypechk.pas
Description: Binary data
_______________________________________________ fpc-devel maillist - fpc-devel@lists.freepascal.org http://lists.freepascal.org/mailman/listinfo/fpc-devel
