[h323plus] Introducing PBoolean - API change alert!!!

Robert Jongbloed robert.jongbloed at bigpond.com
Sun Nov 18 20:31:02 EST 2007


If  you have written any programs based on PTLib/OpenH323/H323plus/OPAL,
please read this, IT'S IMPORTANT!!

 

 

Adam Butcher from Selex Communications submitted patches a while ago to
change BOOL to PBoolean. The reason was so that PBoolean could be defined as
the ANSI standard bool type and we get a whole pile of beneficial compile
time checks. The process found a couple of places where a function had a
BOOL return type but integer (tri-state) values were being returned!

 

Well, Craig and I had thought of this for a while as being a "good thing" to
do, but lack of time has always been an issue. Now, as we have been
presented with all the hack work, it seems that this is it's time.

 

 

Of course things are NEVER easy. The change has a nasty API issue that can
silently stop user programs from working. It revolves around my #1 pet hate
in C++, if you change the signature of a virtual function, overrides
silently no longer override the virtual. For example:

 

   //------ in the library

   class OpalXXX

   {

        virtual void SomeFunction( PBoolean );

   };

 

   //------- in some unchanged dependent code in another application

   class MyXXX : public OpalXXX

   {

         virtual void SomeFunction( BOOL );

   };

 

 

The function MyXXX::SomeFunction() will not be called as it should be. And
the stupid compiler (NONE of the compilers) give any warnings or indication
there is an issue. Way back in the preliminary versions of C++ they did have
a keyword "override" which would do exactly what we want, error if the base
class does not have the function being overridden as virtual. The original
designers (Stroustrup et al) took out, for presumably good reasons, though I
cannot for the life of me think what! There is even a number proposals, e.g.
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2108.html, to put a
variants of it back and Microsoft have even done so,
http://msdn2.microsoft.com/en-us/library/41w3sh1c.aspx, on their latest
compiler.

 

Anyway we have to deal with existing compilers and this problem is really
insidious, but we think the gains are worth the pain. And we have provided a
work around for people.

 

 

So, what we need from you guys is please check out the PTLib and OPAL in the
branches:

 

 
https://opalvoip.svn.sourceforge.net/svnroot/opalvoip/ptlib/branches/Boolean
Change

 
https://opalvoip.svn.sourceforge.net/svnroot/opalvoip/opal/branches/BooleanC
hange

 

and compile your programs against it. Now by default it will use the ANSI
bool form of PBoolean, so any overrides of PTLib or OPAL functions must be
changed to use PBoolean instead of BOOL. For those not using MFC, this is
really easy, just do a complete search and replace everywhere of BOOL to
PBoolean (or better yet, just bool) and that should do the job. Oh, you
might like to search and replace TRUE to true and FALSE to false while you
are at it, though you don't HAVE to do that.

 

 

Now, for those that simply do not wish to make these changes, we have a
backward compatibility option: use the

 

  --disable-ansi-bool

 

option with configure (both Unix and Win32) and PBoolean is defined as BOOL
instead of bool. This should (in theory) be 100% compatible with the
previous signatures. So no changes to applications based on PTLib need be
made.

 

 

But we need you guys to CHECK THAT THIS IS OK!

 

We are currently proposing to merge it into the trunk on Monday 26th of
November, we are happy to extend that if asked, but will not wait forever,
the back merging is really painful!

 

 

P.S. A random note to developers, the PBoolean is only there for backward
compatibility. It is recommended that any NEW functions all just use the
bool keyword. Use lowercase true and false too.

 

 

P.P.S. Sorry about all the shouting ... J

 

 

Robert Jongbloed

OPAL/OpenH323 Architect and Co-founder.

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.packetizer.com/pipermail/h323plus/attachments/20071119/361c95dd/attachment-0003.html>


More information about the h323plus mailing list