On 05/ 7/10 03:48 PM, Danek Duvall wrote:
Brock Pytlik wrote:
http://cr.opensolaris.org/~bpytlik/ips-15843-v1/
As far as I can tell, you're resolving the dependencies implied by a
file only in the proto area that the file is found in, which isn't the
right thing to do. Dependencies can be resolved in any of the proto
areas, which makes me think that proto_dir isn't per-action, but still
needs to be passed around as its own object, only now it's a list.
This had me confused originally too. Remember there are 2 phases generate
and resolve. Generate actually looks at and examines the files, resolve
merely looks at manifests. So, when doing the generate phase, we need to
look @ the files, so we need the proto_dirs around to see where the files
might be found. Once we know which proto_dir a file is found in, we have
all the information generate needs.
Resolve, on the other hand and including the internal resolve deps bit
that generate does, always works on image-relative paths. The proto_dirs
are totally irrelevant at the resolve stage.
Okay, so why do we store the proto dir in the action at all, then?
Because things like run paths from elf dump are given in absolute paths.
So we need the proto dir the file was in to make those paths into image
relative paths. It's there for the same reason the localpath to the file
is in that the action itself is reasonable way to carry around the
action specific information needed to generate the files the action
depends on.
What's up with process_hardlink_deps() passing "" as proto_dir to
HardlinkDependency?
The proto_dir is used by make_relative to relativize the run_paths.
HardlinkDeps never call this function and thus have no use for the
proto_dir. Which is a good thing becuase with this wad, the hardlink
action doesn't have its proto_dir set, so it doesn't have a value to pass
in there in the first place. If you prefer, I can refactor the Dependency
class structure in base.py to create more levels so that it doesn't need
to pass in a proto_dir at all.
Could you just remove the proto_dir parameter from HardlinkDependency's
constructor, and pass None up to PublishingDependency, documenting that
special value there?
Otherwise, I'd pass None up from process_hardlink_deps(). Or would None
just not work at all (I can't tell if _check_paths() would ever be called)?
If that's the case, then I'd put a quick comment in explaining why we're
doing something special. "" just seems like we're using the cwd as the
proto area, not that we're ignoring it.
Sure
Brock
Danek
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
