Add an inspect command that gathers metrics about a Jelly file

[Karolina-Bogacka]

Issue: #39

• By default aggregates the metrics based on the whole file, but can also return them --per-frame
• Returns a valid Yaml as an output
• Can return the metrics to a file with --to
• Accepts an input stream as well as a file

[Karolina-Bogacka]

Context - Test coverage OK, AOT tested

[Ostrzyciel]

• There is no newline at the end of the inspect output.
• Please add an empty line between the block about stream options and frames, for readability.

[Ostrzyciel]

• Logical and physical types should be written not only by ID, but also by name, for readability – e.g., GRAPHS (3)

[Ostrzyciel]

• Boolean options should be printed as true/false, not 0/1.
• The order of outputted keys seems random. Can you use an ordered structure so that it makes more sense?

[Karolina-Bogacka]

Format after revisions for --per-frame

stream_options: 
  stream_name: ""
  physical_type: PHYSICAL_STREAM_TYPE_TRIPLES (1)
  generalized_statements: true
  rdf_star: true
  max_name_table_size: 128
  max_prefix_table_size: 16
  max_datatype_table_size: 16
  logical_type: LOGICAL_STREAM_TYPE_FLAT_TRIPLES (1)
  version: 1

frames: 
  - frame_index: 0
    option_count: 1
    triple_count: 3
    quad_count: 0
    graph_start_count: 0
    graph_end_count: 0
    namespace_count: 0
    name_count: 3
    prefix_count: 3
    datatype_count: 0
    
  - frame_index: 1
    option_count: 1
    triple_count: 3
    quad_count: 0
    graph_start_count: 0
    graph_end_count: 0
    namespace_count: 0
    name_count: 3
    prefix_count: 3
    datatype_count: 0

[Karolina-Bogacka]

And default aggregation

stream_options: 
  stream_name: ""
  physical_type: PHYSICAL_STREAM_TYPE_TRIPLES (1)
  generalized_statements: true
  rdf_star: true
  max_name_table_size: 128
  max_prefix_table_size: 16
  max_datatype_table_size: 16
  logical_type: LOGICAL_STREAM_TYPE_FLAT_TRIPLES (1)
  version: 1

frames: 
  frame_count: 1
  option_count: 1
  triple_count: 3
  quad_count: 0
  graph_start_count: 0
  graph_end_count: 0
  namespace_count: 0
  name_count: 3
  prefix_count: 3
  datatype_count: 0

[Ostrzyciel]

This requires you to allocate a data structure per frame and keep it in memory... making all of this a non-streaming algorithm. If you feed in a very long file, you're going to have OOMs.

Can you rewrite it so that it operates on iterators?

[Karolina-Bogacka]

I rewrote the whole thing to work on iterators - only in the case of the last step of the --per-frame aggregation, I write each frame stat to the output inside of a foreach statement, which in my understanding should be fine because I'm immediately discarding any materialized objects, but please let me know if it is not so.

[Ostrzyciel]

Wooo, it works amazing :) I tested it on a stream with 14M frames (https://w3id.org/riverbench/datasets/officegraph/dev), and it didn't even break a sweat.