Danek Duvall wrote:
Shawn Walker wrote:
http://cr.opensolaris.org/~swalker/pkg-11324/
catalog.py:
- line 121: if you're not retrieving the data, why pop instead of del?
Was to avoid having to trap a key error or checking for the existence of
the key. Would a del wrapped with try/except be preferable?
- line 171: False, maybe?
- Is signing really that much of a problem, performance-wise? Is it just
the setting of sort_keys in the JSON dumper? We're eventually going to
do be doing it on all of our own catalogs, so optimizing for when it's
not there doesn't seem especially useful in the long run.
There seemed little point to signing the image catalogs or transformed
v0 catalogs.
image.py:
- A downside of having the explicit save_pkg_state() is that previously,
you were able to avoid saving one or the other if you could determine
it wasn't necessary. None of the save() routines check to see if they
actually need to update the bits on disk, so you're now going without
this savings. Not to say that it isn't better than it was, despite
this, but ...
About 15 minutes better on my system :/
Another downside is that image state isn't consistent until the very
end of the image operation. While there were still windows of
inconsistency before, it's much bigger now. You might consider having
a simple, append-only log file that you write in set_pkg_state(), and
coalesce into the main package during save_pkg_state(), or at startup,
if we'd crashed previously. Not this wad, though.
Yeah; the same had occurred to me, but it seemed not a terribly big deal
as we do several other important things after installing packages that
leave the image in a not quite right state. Like updating the image
configuration variant information, executing actuators, etc.
Is there a particular failure case or consequence of this change you had
in mind?
publisher.py:
- line 1056: you're not comparing the package count, though. At least,
not here.
Well, I'm checking that it's non-zero, but the comment needs to be
clearer :P
Cheers,
--
Shawn Walker
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss