Re: Ugly warnings under 64 bit compilation

Donal K. Fellows <donal.k.fellows@manchester.ac.uk> wrote:
> Andreas Leitgeb wrote:
>> Wouldn't it be just best to define a dummy-procedure whose
>> signature matches Tcl_FreeProc, which does either nothing,
>> or abort with trace, and use that dummy function's address
>> as a second "special" value (which appears to me to be the
>> whole point of TCL_VOLATILE)
>
> Longer term, yes. Not in the 8.* series though due to backward
> compatability (it's really bad form to change the value of constants
> unless you also allow the old constants to be used, which takes you
> almost back to square one anyway).

not all the way "back to square one":
 if TCL_VOLATILE was newly defined to the address of a real
function, only the public functions that take a TCL_VOLATILE
need to know about the old value (and will throw warnings
at their own compilation time.)

Any user code that compiles&links to the tcl-libraries would be
clean after that change, and code that has been compiled (not
necessarily linked) in the past, will continue to work.

What's left is code that might have extracted the numerical
value from TCL_VOLATILE, and this is broken anyway.
Received on Sun Apr 30 02:29:12 2006