revisor

Revisor

Revisor allows you to define specifications for NewsDoc contents as a series of declarations and pattern matching extensions to existing declarations.

Local testing

For running the actual tests and benchmarks, see the section on Testing.

The easiest way to test specifications against documents is by running the "revisor" command like so:

$ revisor document ./testdata/article-borked.json

That will validate the document using only the specifications in "./constraints/core.json".

Try running the same validation against a document with organisation specific content:

$ revisor document ./testdata/example-article.json
meta block 2 (tt/slugline): undeclared block type or rel
attribute "type" of meta block 2 (tt/slugline): undeclared block attribute
attribute "value" of meta block 2 (tt/slugline): undeclared block attribute
content block 2 (tt/visual): undeclared block type or rel
attribute "type" of content block 2 (tt/visual): undeclared block attribute
data attribute "caption" of content block 2 (tt/visual): unknown attribute
link 1 self(tt/picture) of content block 2 (tt/visual): undeclared block type or rel
attribute "type" of link 1 self(tt/picture) of content block 2 (tt/visual): undeclared block attribute
attribute "uri" of link 1 self(tt/picture) of content block 2 (tt/visual): undeclared block attribute
attribute "url" of link 1 self(tt/picture) of content block 2 (tt/visual): undeclared block attribute
attribute "rel" of link 1 self(tt/picture) of content block 2 (tt/visual): undeclared block attribute
data attribute "credit" of link 1 self(tt/picture) of content block 2 (tt/visual): unknown attribute
data attribute "height" of link 1 self(tt/picture) of content block 2 (tt/visual): unknown attribute
data attribute "hiresScale" of link 1 self(tt/picture) of content block 2 (tt/visual): unknown attribute
data attribute "width" of link 1 self(tt/picture) of content block 2 (tt/visual): unknown attribute
content block 3 (tt/dateline): undeclared block type or rel
attribute "type" of content block 3 (tt/dateline): undeclared block attribute
data attribute "text" of content block 3 (tt/dateline): unknown attribute
documents had validation errors

Use the flag -spec ./constraints/tt.json to load the organisation specific constraints for TT.

Running a revisor server

It's also possible to run revisor as a service with the serve command, it takes the same --spec/--core-spec as the document command, and adds --addr to control the address to listen to.

Start the server in one shell:

$ revisor serve

...and post the example article to it in another using curl:

$ curl --data @testdata/example-article.json localhost:8000

You should get the same validation errors as in the previous example, but in JSON format. An empty array is returned for valid documents.

Writing specifications

The main entities points in a specification are documents, blocks and properties. Documents are declared by type, blocks by type, rel, and/or role, and properties by name. An entity is not valid if we don't have a matching declaration for it, regardless of whether somebody has pattern-matched against it.

Both pattern matching and a lot of the validation that's performed is done though key value pairs of a name and a string constraint. Say that we want to match all links that have a rel of "subject", "channel", or "section" and add the ability to have "broader" links added to them, the specification would then look like this:

{
  "name": "Associated with and broader links",
  "description": "Extends subject, channel, and section links with broader links",
  "match": {"rel": {
    "enum": ["subject", "channel", "section"]
  }},
  "links": [
    {
      "declares": {"rel":"broader"},
      "attributes": {
        "type": {},
        "title": {}
      }
    }
  ]
}

Here we declare that links with rel "broader" are valid for all blocks that matches our expression, see "Block attributes" for a list of attributes that can be used in pattern matching. We also define that the attributes type and title must be present. The {"enum":...} object and the empty objects ({}) for type and title are all examples of string constraints.

String constraints

The distinction between optional and allowEmpty is only relevant for data attributes. The document and block attributes defined in the NewsDoc schema always exist, so optional and allowEmpty will be treated as equivalent.

Formats

The following formats are available:

• RFC3339: an RFC3339 timestamp ("2022-05-11T14:10:32Z")
• int: an integer ("1234")
• float: a floating point number ("12.34")
• bool: a boolean ("true" or "false")
• html: validate the contents as HTML
• uuid: validate the string as a UUID
• wkt: validate the string as a WKT geometry.

When using the format "html" it's also possible to use htmlPolicy to use a specific HTML policy. See the section on HTML policies.

The document and block uuid attributes are always validated as UUIDs and need no additional "uuid" format specified.

Time formats

A Go time parsing layout (see the time package for documentation) that should be used to validate the timestamp.

Globs

Glob matching uses https://github.com/gobwas/glob for matching, and the glob patterns are compiled with "/" and "+" as separators.

WKT geometry

The geometry specification is a combination of the geometry type to expect, and optionally the types of coordinates it should contain, in the format {geometry-type}[-{coordinates}]. If no geometry is specified any of the supported types and coordinates are allowed. If no coordinates are specified the default X and Y coordinates are assumed.

Geometry types:

• point
• multipoint
• linestring
• multilinestring
• polygon
• multipolygon
• circularstring

Coordinates, X and Y is the default if nothing else is specified:

• z: X, Y, and Z coordinates
• m: X and Y coordinates and a measurement
• zm:X, Y and Z coordinates and a measurement

Labels and hints

Labels and hints do not play any role in the validation of documents. They are instead meant to describe the value for systems that use the information in the revisor schema to process the data correctly. It could f.ex. be used to tell a system that a specific WKT point is the position of the document itself, that a string value should be indexed as a keyword (non-tokenised), or provide other kinds of processing hints unrelated to the validation.

Example block declaration:

{
  "declares": {
    "type": "tt/slugline"
  },
  "maxCount": 1,
  "attributes": {
    "value": {
      "labels": ["keyword"],
      "hints": {"alias": ["slug"]}
    }
  }
}

Writing a document specification

A specification for a document contains:

• documentation attributes name and description
• a declaration (declares) or pattern matching rule (match)
• attribute constraints (attributes)
• meta, links, and content block specifications

{
  "name": "Planning item",
  "description": "Planned news coverage",
  "declares": "core/newscoverage",
  "meta": [
    {
      "name": "Main metadata block",
      "declares": {"type":"core/newscoverage"},
      "count": 1,
      "data": {
        "dateGranularity": {"enum":["date", "datetime"]},
        "description": {"allowEmpty":true},
        "start": {"format":"RFC3339"},
        "end": {"format":"RFC3339"},
        "priority": {},
        "publicDescription":{"allowEmpty":true},
        "slug": {"allowEmpty":true}
      }
    }
  ],
  "links": [
    {
      "declares": {"type": "x-im/assignment"},
      "links": [
        {
          "declares": {
            "rel":"assignment", "type": "x-im/assignment"
          },
          "attributes": {
            "uuid": {}
          }
        }
      ]
    }
  ]
}

Writing a block specification

A block specification can contain:

• documentation attributes name and description
• a declaration (declares) or pattern matching rule (match)
• attribute constraints (attributes)
• data constraints
• meta, links, and content block specifications
• count, minCount and maxCount to control how many times a block can occur in the list of blocks it's in
• blocksFrom directives that borrows the allowed blocks from a declared document type.

{
  "declares": {"type": "core/socialembed"},
  "links": [
    {
      "declares": {"rel":"self", "type":"core/tweet"},
      "maxCount": 1,
      "attributes": {
        "uri": {"glob":["core://tweet/*"]},
        "url": {"glob":["https://twitter.com/narendramodi/status/*"]}
      }
    },
    {
      "declares": {"rel":"alternate", "type":"text/html"},
      "maxCount": 1,
      "attributes": {
        "url": {"glob":["https://**"]},
        "title": {}
      },
      "data": {
        "context": {},
        "provider": {}
      }
    }
  ]
}

HTML policies

HTML policies are used to restrict what elements and attributes can be used in strings with the format "html". Attributes are defined as string constraints on elements. The default policy could look like this:

  "htmlPolicies": [
    {
      "name": "default",
      "elements": {
        "strong": {
          "attributes": {
            "id": {"optional":true}
          }
        },
        "a": {
          "attributes": {
            "id": {"optional":true},
            "href": {}
          }
        }
      }
    },
    {
      "name": "table",
      "uses": "default",
      "elements": {
        "tr": {
          "attributes": {
            "id": {"optional":true}
          }
        },
        "td": {
          "attributes": {
            "id": {"optional":true}
          }
        },
        "th": {
          "attributes": {
            "id": {"optional":true}
          }
        }
      }
    }
  ]

All "html" strings that use the default policy would then be able to use <strong> and <a>, and the "href" attribute would be requred for <a>. A "html" string that uses the "table" policy would be able to use everything from the default policy and <tr>, <td>, and <th>.

A customer can extend HTML policies using the "extend" attribute:

  "htmlPolicies": [
    {
      "extends": "default",
      "elements": {
        "personTag": {
          "attributes": {
            "id": {}
          }
        }
      }
    }
  ]

This would add support for "/" (HTML is case insensitive) to the default policy, and any policies that use it. Only one level of "extends" and "uses" is allowed, further chaining policies will result in an error.

Attribute reference

Document attributes

A list of available document attributes, and whether they can be used in pattern matching.

Block attributes

A list of available block attributes, and whether they can be used in pattern matching.

Testing

Revisor implements a file-driven test in TestValidateDocument that checks so that all the "testdata/results/*.json" files match the validation results for the corresponding document under "testdata/". Result files with the prefix "base-" will be validated against "constraints/naviga.json", for result files with the prefix "example-" the "constraints/example.json" constraints will be used as well.

If the constraints have been updated, or new example documents have been added, the result files can be regenerated using ./update-test-results.sh.

Benchmarks

The benchmark BenchmarkValidateDocument tests the performance of validating "testdata/example-article.json" against the naviga and example organisation contsraint sets.

To run the benchmark execute:

$ go test -bench . -benchmem -cpu 1

Add the flags -memprofile memprofile.out -cpuprofile profile.out to collect CPU and memory profiles. Run go tool pprof -web profile.out for the respective profile files to open a profile graph in your web browser.

Comparing benchmarks

Install benchstat: go install golang.org/x/perf/cmd/benchstat@latest.

Run the benchmark on the unchanged code (stash your changes or check out main):

$ go test -bench . -benchmem -count 5 -cpu 1 | tee old.txt

Then run the benchmarks on the new code:

$ go test -bench . -benchmem -count 5 -cpu 1 | tee new.txt

Finally, run benchstat to get a summary of the change:

$ benchstat old.txt new.txt
name              old time/op    new time/op    delta
ValidateDocument     203µs ± 7%      99µs ± 3%  -51.03%  (p=0.008 n=5+5)

name              old alloc/op   new alloc/op   delta
ValidateDocument     134kB ± 0%      35kB ± 0%  -73.74%  (p=0.008 n=5+5)

name              old allocs/op  new allocs/op  delta
ValidateDocument     1.05k ± 0%     0.59k ± 0%  -43.48%  (p=0.008 n=5+5)

Fuzz tests

There are two fuzz targets in the project: FuzzValidationWide that allows fuzzing of the document and two constraint sets. It will load the core constraints, the example organisation constraints, and all documents in "./testdata/" and add them as fuzzing seeds. FuzzValidationConstraints adds all constraint sets from the "./constraints/" and adds them as fuzzing seeds. The fuzzing operation is then done against all documents in "./testdata/".