Martin Kretzschmar wrote:
Kay Ramme - Sun Germany - Hamburg schrieb:
as the performance of rtl::OUStrings (the UNO C/C++ String) is coming
up for discussions once a while (because of showing up in profiles
e.g. when loading a large spreadsheet), I wrote a short paper
http://udk.openoffice.org/common/man/concept/string_invest.html
investigating especially into construction&destruction and suggesting
a slightly modified variant, with improved behavior. So, if you have
some time left, I would like to get your feedback ;-).
If I understand correctly, the current rtl::OUString has the interesting
feature of being reinterpretable as/binary-identical to an rtl_uString*
(the shared sequence on the heap) and vice versa. I have definitely seen
code that freely casts between the two.
The "Shared-Heap-String-with-a-Buffer" sounds like a rtl::OUString
(C++)-only plan. Wouldn't that introduce problems with code doing such
casts?
Kay's proposal implies that we change UNO in an incompatible way (which
would indeed affect code of the kind you mention, for example).
My criticism with the proposal is that it does not go far enough: If we
ever take the big big step of changing UNO incompatibly, then instead of
exchanging one concrete implementation of rtl::OUString with a slightly
different one, we should rather investigate in replacing the concrete
implementation with an abstract interface. Behind that interface, we
can then experiment with well-tuned implementations for specific usage
scenarios (see Eike's mail for an example), without the need to become
incompatible once again.
-Stephan
Regards,
Martin
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
