On 06/16/11 14:44, Brock Pytlik wrote:
On 06/16/11 11:23, Shawn Walker wrote:
On 06/10/11 00:13, Brock Pytlik wrote:
Webrev:
http://cr.opensolaris.org/~bpytlik/ips-18441-v1/
Bug:
18441 pkg needs a freeze mechanism
These changes provide the freeze mechanism I outlined previously. The
biggest changes I know of from the original use cases I sent out are
that:
1) If pkg unfreeze <blah> doesn't match any frozen packages, no error is
reported. I decided that seems reasonable since pkg unfreeze details
each package that was unfrozen, so no output means no packages were
unfrozen.
2) Publishers can be part of the patterns used for matching, but a
package is only frozen to a version, not to a publisher.
I think that covers the major changes.
src/client.py:
line 2159: docstring indicates this only sets freezes, but it also
displays them
I'll fix this once we decide what we're doing below.
lines 2199-2200: Having unfreeze display packages as well as freeze
is inconsistent with the paradigm we use for other commands. I'd
rather not have both of these commands be capable of displaying the
list of frozen packages. This is also not documented in the man
page.
It exactly mirrors the paradigm that avoid uses. Since that looked like
freeze's closest cousin, I used that as my reference.
A single subcommand does not consistency make :-)
With that said, I'll defer to Danek or Dan's opinion here.
...
You may also want to mention that publisher information in the FMRI
will be ignored. (That is, you can't place publisher-specific
freezes, although that might be a nice future RFE to allow
per-package stickiness.)
Well, it's not ignored, it's used during the matching, that's why I
didn't complain if the user put a publisher there.
What I meant was, as you've written it, I don't think specifying a
freeze of 'pkg://solaris/web/firefox' (assuming web/firefox is
installed) would prevent installation of web/firefox from a different
publisher. Or does it?
...
src/modules/client/imageplan.py:
lines 2323: The variable names are a bit hard to grok, can you rename:
'not_match_ok' -> 'raise_unmatched' and 'not_inst_ok' to
'raise_not_installed'? That would also have the benefit of getting
rid of the double negative expressions on lines 2454-2455 (e.g.
not not_match_ok).
gee, that was what I had originally, then changed to match the naming
scheme that Ed used in his recent putback.
Thanks for putting back what you had originally :-)
...
src/tests/cli/t_pkg_freeze.py:
line 38: shouldn't this be True? (image are destroyed and created
every time anyway I thought...)
Doesn't appear to be the case. Turning on persistent setup immediately
caused all but the first test to fail. Setting this to true then moving
the image creation into each test, instead of the setUp method saved 3
seconds.
Seems worth doing then. Thanks for that. Maybe we should have a
persistent_image property too that defaults based on persistent_setup
unless explicitly overridden?
-Shawn
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
