More detail from test_runner coverage report

[philnash]

What is the problem this feature will solve?

First, I understand that test coverage is still experimental. I'm hoping to move it in the direction of less experimental! I've been working on reporters for both test executions and then looking into test coverage reporters.

I've been trying to implement both the SonarQube generic test coverage report and then also an lcov-text report. With the current output of the test:coverage event for a reporter, there isn't enough detail to produce either of these reports.

What is the feature you are proposing to solve the problem?

I propose that the test:coverage event produces more data. Namely:

• Coverage count for each line of code
• Detail on functions, their name, first line number, and coverage count
• Detail on branches, their first line number, and coverage count

As part of this, I would also propose that the array of uncoveredLineNumbers is removed, as it is serviced by coverage counts per line.

I would like to see each file object in the summary look like the following, with keys for functions, branches and lines:

{
  "path": "/path/to/index.js",
  "totalLineCount": 9,
  "totalBranchCount": 3,
  "totalFunctionCount": 1,
  "coveredLineCount": 7,
  "coveredBranchCount": 2,
  "coveredFunctionCount": 1,
  "coveredLinePercent": 77.77777777777779,
  "coveredBranchPercent": 66.66666666666666,
  "coveredFunctionPercent": 100,
  "functions": [
    {
      "name": "conditionalAdd",
      "count": 1,
      "line": 3
    }
  ],
  "branches": [
    {
      "line": 1,
      "count": 1
    },
    {
      "line": 3,
      "count": 1
    },
    {
      "line": 6,
      "count": 0
    }
  ],
  "lines": [
    {
      "line": 1,
      "count": 1
    },
    {
      "line": 2,
      "count": 1
    },
    {
      "line": 3,
      "count": 1
    },
    {
      "line": 4,
      "count": 1
    },
    {
      "line": 5,
      "count": 1
    },
    {
      "line": 6,
      "count": 1
    },
    {
      "line": 7,
      "count": 0
    },
    {
      "line": 8,
      "count": 0
    },
    {
      "line": 9,
      "count": 1
    }
  ]
}

I have a basic implementation of this, and would be happy to tidy it up for a pull request as well as update the existing reporters to handle the new format. I also wondered whether an lcov-text coverage report would be worth providing out of the box or whether that is better in a userland module?

What alternatives have you considered?

I considered keeping the uncoveredLineNumbers array, but that puts more work on report implementors to merge an array of numbers and an array of objects if they want to report in order.

There is certainly a consideration that this will be a large object. I considered whether sending the data as tuple style arrays (e.g. for a function [name, count, line], for a line [line,count] would save memory. It likely would, but I don't know how important that is, so I erred towards developer experience and using objects for the data.

[cjihrig]

I have a basic implementation of this, and would be happy to tidy it up for a pull request as well as update the existing reporters to handle the new format.

As long as it works as well or better than what is currently there, I say go for it.