Apache and standard C++ library

Paolo Carlini pcarlini@suse.de
Tue May 17 10:55:00 GMT 2005


Joe Buck wrote:

>But this is arguably a political problem, not an R&D problem; not really
>something we can do anything about on this list.
>
I'm not 100% sure. From the R&D point of view we are unavoidably
suffering restrictions dictated by ABI-stability. Indeed, during the
next months we will re-evaluate how much of the ongoing development work
in v7 could be actually merged without breaking the 3.4/4.0 ABI (we
would do that anyway, not because of the announcement ;), but maybe
having an additional competitor can lead us the re-calibrate and
re-discuss the unavoidable trade-offs between (ABI) stability and
innovative development.

To be concrete, the basic_string example comes immediately to mind: we
all agree, I think, that on modern, heavily MT-applications we could
have an edge over the reference-counted Apache implementation if we
could move away from our-own reference-counted implementation: I can
guarantee (you have to trust me, for now) that, if we want that from the
"political" (call that stabiliy, maybe) point of view we *can* do that
pretty quickly.

Luckily, the whole area of TR1, which sees libstdc++-v3 rather well
positioned, doesn't depend on ABI issues and we are definitely going to
further concentrate our efforts. But we need help.

Paolo.



More information about the Libstdc++ mailing list