[omniORB] Sequence memory management performance improvement

[Dirk O. Siebnich]

I have removed the magic number checks in sequence buffer alloc and
free, plus I have improved the efficiency of the copybuffer functions by
removing the double initializations. This should improve the performance
of sequence operations, at the expense of not logging an error when
buffers are mixed between different sequence types - which I don't think
is a real issue.
The patch is against the current CVS of omniORB 4.1, what do you all
think? I haven't given any thought to whether this breaks the ABI or not
- Duncan?

Regards,
Dirk
-------------- next part --------------
A non-text attachment was scrubbed...
Name: magicnumbers.patch
Type: text/x-patch
Size: 9118 bytes
Desc: not available
Url : http://www.omniorb-support.com/pipermail/omniorb-list/attachments/20080719/d0391244/magicnumbers.bin

[Duncan Grisby]

Post by Dirk O. Siebnich
I have removed the magic number checks in sequence buffer alloc and
free, plus I have improved the efficiency of the copybuffer functions by
removing the double initializations. This should improve the performance
of sequence operations, at the expense of not logging an error when
buffers are mixed between different sequence types - which I don't think
is a real issue.

You are right that there's no real need for those. The checks were added
when we did the CORBA compliance testing years ago, because the
compliance tests tested giving invalid pointers to the buffer management
functions. That kind of bug ought to be caught by memory checking tools
rather than by omniORB.

Having said that, have you actually seen that code as a bottleneck in
profiling? I've not seen it appear in the profiling I've done.

Post by Dirk O. Siebnich
The patch is against the current CVS of omniORB 4.1, what do you all
think? I haven't given any thought to whether this breaks the ABI or not

I think it breaks the ABI, so it can't go in 4.1.x. It'll have to wait
for 4.2.x.

Cheers,

Duncan.