Hi all,
I have a couple of extra toys for injection points in my bucket that
I'd like to propose for integration in v18, based on some feedback I
have received:
1) Preload an injection point into the backend-level cache without
running it. This has come up because an injection point run for the
first time needs to be loaded with load_external_function that
internally does some allocations, and this would not work if the
injection point is in a critical section. Being able to preload an
injection point requires an extra macro, called
INJECTION_POINT_PRELOAD. Perhaps "load" makes more sense as keyword,
here.
2) Grab values at runtime from the code path where an injection point
is run and give them to the callback. The case here is to be able to
do some dynamic manipulation or a stack, reads of some runtime data or
even decide of a different set of actions in a callback based on what
the input has provided. One case that I've been playing with here is
the dynamic manipulation of pages in specific code paths to stress
internal checks, as one example. This introduces a 1-argument
version, as multiple args could always be passed down to the callback
within a structure.
The in-core module injection_points is extended to provide a SQL
interface to be able to do the preloading or define a callback with
arguments. The two changes are split into patches of their own.
These new facilities could be backpatched if there is a need for them
in the future in stable branches, as these are aimed for tests and the
changes do not introduce any ABI breakages with the existing APIs or
the in-core module.
Thoughts and comments are welcome.
--
Michael