|
|
|
|
1. Manifests need type explorers to be run before
|
|
|
|
|
Only this way the manifest can use the values
|
|
|
|
|
|
|
|
|
|
2. Type explorers can depend on the (successful) execution of other objects
|
|
|
|
|
A type explorer may lookup values (ownership, size, processes, etc.)
|
|
|
|
|
that are depending on the code of other objects.
|
|
|
|
|
|
|
|
|
|
3. The 'require="a" b' dependency specifies only that execution of b needs to wait until a is finished
|
|
|
|
|
Other requirements given at another position may not be
|
|
|
|
|
executed before b, as cdist interprets require="a" as
|
|
|
|
|
"a is everything that is needed to realise b"
|
|
|
|
|
|
|
|
|
|
4. The 'require' state is thus *not* suitable to trigger reloads
|
|
|
|
|
A type b, which uses the require="" statement to wait for all
|
|
|
|
|
require="" dependencies, may be executed in an incomplete
|
|
|
|
|
status.
|
|
|
|
|
|
|
|
|
|
5. Depending on multiple dependencies defined at different locations requires all objects to be present
|
|
|
|
|
Because otherwise the dependent object cannot be sure that the
|
|
|
|
|
list of objects it waits for is complete.
|
|
|
|
|
|
|
|
|
|
6. If a statement requires all objects to be present, all manifests must have run already.
|
|
|
|
|
If all manifests have run already, also all type explorers have run already (#1).
|
|
|
|
|
If all type explorers have run already, some objects may have been finished completly already (#2).
|
|
|
|
|
|
|
|
|
|
7. Because of #4 and #6, we need to introduce another dependency that allows reloading after
|
|
|
|
|
|
|
|
|
|
8. Because the object could be finished already due to a 'require' dependency, we need another script
|
|
|
|
|
Maybe "gen-post-run-local", "gen-post-run-remote"
|
