On 09/13/10 04:38 PM, Shawn Walker wrote:
On 09/13/10 04:18 PM, Keith Mitchell wrote:
...
From what I recollect of the logs, the issue is before the rsync. It
seems to be at or just before merge.py is run (i.e., the issue *seems*
to be with either how I call merge.py, or how I set-up the repo).
The symptom I'm seeing from the output is that merge.py is finding the
x86 packages, but not the sparc ones, when attempting to merge. The x86
packages/repo were built on a system with 146 pkg bits; the sparc stuff
was made on a system running stock 147. (When I look at the x86 repo, I
see what appears to my untrained eye to be a version 3 repo; when I look
at the sparc repo I see a publisher/install-nightly directory which I'm
guessing is version 4?).
From the sound of it, the issue is likely trying to do the merge and
publish step on a system using a version of pkg(5) older than the one
used to create the repositories you're merging.
Indeed, that appears to be the cause. Once the x86 machine (which was
used to merge the repos) was successfully updated to 147, the merging
completed successfully and normally.
Thanks!
- Keith
Based on your earlier comments, the logic for this was lifted from the
ON build stuff. In which case, it likely creates the repository's
directory during the build process before calling 'pkgsend
create-repository'. If you change it to not do that, you should then
get a repository in version 3 format and everything should work again.
This subtle difference in behaviour is because of some compatibility
logic for pkgsend that handled the case where you executed
create-repository and the repository directory didn't exist yet, but
not the case where you were using it on an empty directory.
If you look at changeset 13235 in the ON gate, you'll see the change I
made there to workaround your issue.
-Shawn
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
