during/spawn used not to add linkage assertions to its
initial-assertion set. In addition, if a spawned actor died in its
initial boot procedure, its initial assertions would never be visible.
These two problems interlocked to cause a space leak in during/spawn,
where monitoring facets would never be cleaned up.
This change does two things:
- adds linkage assertions to the initial-assertion set in during/spawn
- properly briefly signals initial-assertions even when a new actor
immediately crashes.
Together, these repair the space leak in during/spawn with a crashy
child startup procedure.
Instead of having restriction-paths as an adjunct to a change, they're
more propertly a part of each assertion itself. The new `skeleton.rkt`
keeps an optional restriction-path with each assertion, treating it as
distinct from its underlying assertion. The idea of not signalling
changes in assertions that have a restriction-path mismatch stays.
This is used when relaying: because we don't have access to the full
term, but only the projection results, we are inserting various
`(discard)`s. This is the cause of the failure in
`test/core/nesting-confusion.rkt`.
By adding `restriction-path`, we allow the inner dataspace to avoid
showing a reconstructed term to endpoints that might be able to
observe the reconstructed parts.
An alternative implementation approach is to generalize patterns in
the inner relay actor, translating all `(discard)`s into captures,
which would give us all the relevant terms that we need. The way I've
chosen to go (or at least, to try out) allows us to potentially keep
the "efficient" idea of just transmitting pattern-bound values across
some network link connecting dataspaces. The alternative would require
transmission of the full assertions, eliding no irrelevant detail.
The `add-endpoint!` call is changed in two ways:
- the old `assertion-fn` has become `update-fn`, yielding both
an assertion *and* an optional handler, because if the handler
depends on a field which changes, previously the handler wasn't
being updated
- a new parameter, `dynamic?`, can be set to #f (it's usually #t)
to ensure that the assertion and skeleton-interest are calculated
only once ever, and are not connected to the dataflow machinery.
The first change makes it possible for the `(later-than (deadline))`
pattern, where `deadline` is a field, to work; the second change makes
`during` and `during/spawn` work correctly in the face of field
updates.
Tony Garnock-Jones
Tony Garnock-Jones