We update the sole protocol to more cleanly support multiple sessions.
Primarily, the "sole id" is updated to be a [@p @ta] instead of a @ta,
and it is now generated based off the connected dill session, rather
than statically.
This change ripples out to applications that support the sole protocol:
the subscription path becomes /sole/[ship]/[session] (as opposed to
/sole/[per-ship-constant]), and %sole-action pokes include the new id as
well.
For shoe agents, this means (at the very least) updating the function
signatures of the shoe arms.
/lib/sole has been updated to include helper functions for parsing a
sole-id from a subscription path, and turning a sole-id into its
corresponding path. It also has a function to aid in migrating old
sole-ids.
Existing sole agents are made to kick any known open sessions, forcing a
resubscribe by drum, so that they may use exclusively the new format
going forward. Third-party agents are recommended to do the same.
Note that some functionality, such as |link, still operates exclusively
on the default session. Improvements in this area to follow soon.
Resolves a good number of conflicts. Most notably, re-propagates removal
of gall's %onto, confirms new /app/herm behavior, coerces hood/drum
state adapters back into place, and updates webterm to use the latest
api.
No mark files exist for any of the drum marks, so trying to poke remote drums
would fail anyway, but relying on the mark system in that way seems a bit
fragile, so we add an explicit permission check.
No mark files exist for any of the helm marks (except `%helm-hi`), so trying to
poke remote helms would fail anyway, but relying on the mark system in that way
seems a bit fragile, so we add an explicit permission check.
When we receive a commit that we can apply from the source, we throw
away any skipped pending commits. This lets us receive updates at the
current kelvin even if we heard pending commits for later kelvins.
If a desk is suspended/paused/local, then we should exclude it when
bumping in the +on-load, because a suspended/paused/local desk cannot
have a pending commit that would allow it to bump successfully.
Previously this wasn't accounting for the "couldn't read kelvin from
local desk" case properly, resulting in crashes during updates to base,
after which other desks wouldn't be started back up.
We do not care about what's on foreign, the local desk is the thing
we're taking action on. This is more robust in the face of absent
information about foreign desks etc.
Also stops pretending to take-commet during take-merge-main. We'll get a
separate commit event.
Remote install send a %commit because they receive new files. Local
installs never receive files, so weren't notifying kiln subscribers.
Here we "manually" send a %commit fact when installing a local desk.
