On 20 Jul 2009, at 15:13, Hartmut Reuter wrote:
I think the client should be compiled without ifdefs because it may
have
to deal with different cells among which may be cells deploying OSDs.
There should be no overhead for non-osd AFS traffic even if the client
contains the OSD extensions. And on the client use of the osd
extensions
can be controlled by the "fs protocol" command (e.g. at start time).
The server side, however, should depend on the configure.
Personally, I'm clean to avoid #ifdefs wherever possible. They
dramatically complicate the testing process, and mean that packagers
have to make decisions on behalf of end users. They're appropriate for
items in the development stream, where people may want to test one
piece of code, but not another, but I'm strongly opposed to large
numbers of --enable switches appearing in the production code.
Also, it's vital that it's impossible to shoot yourself in the foot by
accidentally installing a non-rxosd server on a rxosd machine, or vice
versa. This means that even if you do end up using configure switches,
both versions have to equally aware of each other's data to correctly
preserve it.
S.
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
