On Tue, Jul 20, 1999 at 10:55:20AM +0100, Ben Laurie wrote:
> Bodo Moeller:
>>>> So, is backwards compatibility an important issue here
>>>> and is it worth this kind of evil hack, or should we
>>>> just add the parameters?
>>> Backward compatibility is an important issue. Think "dynamic
>>> library".
>> The funny thing here is that if we just add the parameters, then on
>> systems with "standard" calling convention old programs should run
>> without any problems with newly compiled shared libraries: The new
>> library functions will expect one parameter that is not actually there
>> and will take whatever happens to be on the stack and pass it to the
>> callback function as a "pointer argument"; the callback function in
>> turn just ignores this extra (garbage) argument.
> Although you are likely to get away with this, it isn't strictly legal.
I know that. But how much effort would you think achieving real
(guaranteed) compatability is worth? Should the functions that call
those callbacks have an "if" so that old-style callbacks are called
with a different function call than new callbacks with the extra
argument? Or can't we just decide that future releases of the library
should have a clean interface always with the void * argument and
without such complications for backwards compatability? It violates
the abstraction provided by the C language, that's for sure; but
binary compatibility is really just important for those platforms on
which binaries do actually exist, and the dirty function calls should
soon be a thing of the past.
______________________________________________________________________
OpenSSL Project http://www.openssl.org
User Support Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
