Skip to content

Latest commit

 

History

History
116 lines (83 loc) · 3.95 KB

File metadata and controls

116 lines (83 loc) · 3.95 KB

Arrays

Problem

Currently it is inefficient to detect the length of an array and it is impossible to know if a key (without subkeys) should be an array or not.

Constraints

  • None

Assumptions

Considered Alternatives

  • Metadata arrays simply work by convention as they are not serialized in special ways nor they get validated.
  • ###empty_array as in yajl, problem: does not allow efficient access of first element
  • store length (and not last element), problem: needs prepending of #_...
  • store element after last element (C++-Style), would not fit nicely with key-value
  • use value and not the metadata array, problem: is ambiguous
  • use metadata on all children

Decision

Store length in metadata array of key, or keep metadata array empty if empty array. Only children that have # syntax are allowed in a valid array. The index start with #0. Both keyAddName("#12") or keyAddBaseName("#_12") is allowed to add the 13th index.

For example (ni syntax, sections used for metadata):

myarray/#0 = value0
myarray/#1 = value1
myarray/#2 = value2
myarray/#3 = value3
myarray/#4 = value4
myarray/#5 = value5
[myarray]
  array = #5

It is not allowed to have anything else than #5 in the metadata array, e.g. even #4 would be a malformed array.

With the metadata array in spec:/ the spec plugin will add the array marker with the correct length. This needs to be happening early, so that plugins can rely on having correct arrays.

For example:

spec:/myarray      # <- has array marker

user:/myarray      # <- either has correct array marker or no array marker
user:/myarray/#0
user:/myarray/#1

Here, the spec plugin would add array=#1 to user:/myarray if it was not there before.

To look up an array, first do ksLookupByName (ks, "/myarray", 0) on the parent. With the last index you get from its metadata array, iterate over the children. A cascading lookup on every individual child is also needed to make sure that overrides on individual elements are respected.

For example:

spec:/myarray    # <- contains the specification for the array
spec:/myarray/#  # <- contains the specification for the array elements

dir:/myarray/#0  # <- not an array, just an override for user:/myarray/#

user:/myarray    # <- with metadata array=#0, this would be the array we get
user:/myarray/#0

system:/myarray  # <- not found in cascading lookup, as user:/myarray exists

The spec plugin should check if it is a valid array, i.e.:

  • that the parent key always contains the metadata array,
  • that the correct length is in array,
  • that the array only contains # children, and
  • that the children are numbered from #0 to #n, without holes.

Rationale

  • Is very similar to binary metadata but without its problems.
  • The key alone suffices to know if it is an array.
  • One can distinguish an array with keys that are called by chance e.g. #0.

Implications

  • The plugin yajl needs to be fixed
  • metadata library needs to be adapted
  • spec plugin needs to be fixed, a lot of work needed there. Most amount of work is to detect misstructured nested arrays (# intermixed with non # keys) which is a possibility also in all the alternatives of this decision.
  • A user:/ or dir:/ key can change the semantics of a system:/ array, if not avoided by spec.
  • user-facing documentation should contain a note like: "Mixing array and non-array keys in arrays is not supported. In many cases the trivial solution is to move the array part into a separate child-key."

Related Decisions

Notes

ElektraInitiative#182