Repository navigation
New metadata API: Validate dictionary keys as well as values #1356
Description
Description of issue or feature request:
By resolving issue #1139 we are going to add classes for all complex fields inside the top-level metadata classes e.g. meta/ targets (in Timestamp, Snapshot, Targets), delegations (in Targets), keys/roles (in not yet existent Delegation).
The problem is, that many of those complex fields are represented by a dictionary and we are adding classes for the values of
the dictionary in order to easily validate them later, but we should make sure to validate the dictionary keys as well.
Current behavior:
We are not validating dictionary keys yet.
Expected behavior:
We should make sure to validate:
- KEYID in Root and Delegations
- ROLE in Root | Metadata API validation: validate root roles dictionary content #1516
- METAPATH ? in Snapshot and Timestamp
- TARGETPATH ? in Targets
Activity
MVrachev commentedon Oct 19, 2021
It's validated that
keyidis a string inside theKeyclass during initialization.During serialization, it's not validated. Maybe there is no sense to add that as
Keyis an internal helper class and we don't expect the user to modify theKeyinstances.Maybe we can validate that the given
keyidactually points to thekeyvalgiven during the initialization of aKeyinstance?It's problematic to validate that
keyidpoints to exactly onekeyval.Validate that
roleis exactly one of the allowed in the specification: https://theupdateframework.github.io/specification/latest/#root-roleThis should happen during instance creation when calling
from_dictprobably as we are already traversing it there.In timestamp, we have generally fixed that problem by removing
metaand replacing it withsnapshot_meta.For Snapshot, I think we should at least check that
METAPATHis a non-empty string.I think we should at least validate that it's a non-empty string. as defined by the spec.
Maybe we do want to validate that
TARGETPATHis not one of the top-level metadata files or roles? That sounds like a reasonable decision to prevent accidental mistakes.@jku what's your opinion? I ask you as you reviewed other of my prs about validation.
jku commentedon Oct 26, 2021
With keyids I think there's very little we can or should do.
root rolenames are now validated in 1516. The additional things we maybe could still do are
rolesas a Mapping (to indicate that users should not add or remove values from the dictionary during the lifetime of the Root object).roles=Nonein the constructor (and create the default set in that case). This is not strictly validation but lowers the need for validation since then no-one ever needs to manually create therolesdict.I think these make sense, yes
that does not make sense for TARGETPATH but if you meant checking that METAPATH is not in ["root.json", "snapshot.json", "targets.json", "timestamp.json"]... that might make sense as a sanity check.