Feature Request: Instrument option to generate a baseline coverage file

Problem

Currently when testing with files created by the standalone instrument command, if a file isn’t hit it will not produce a coverage output file. This can lead to untouched files being excluded from the coverage report. Istanbul handled this by being able to produce a baseline coverage file with content similar to that produced by running nyc --all=true /usr/bin/true. When the baseline coverage file is included with test generated coverage files and fed to the reporter, files that aren’t hit during tests are included in the report.

In hindsight I think this was the problem being faced in #1044.

Proposal

Add an option to the instrument command, ‘baseline’, that will produce a baseline coverage output json file similar to that produced by running nyc --all=true /usr/bin/true.

Workaround

You can generate a report that includes untouched files with:

1. run nyc --all <options> /usr/bin/true against the uninstrumented code
2. store the output coverage file
3. instrument the source
4. run tests
5. combine baseline and test coverage files
6. report on combined coverage files

[AndrewFinlay]

I’ll try and put together a demo project but it might take a little time as I’m seeing the issue on internal source that I can’t share without some redaction. I’ll drop it into a new issue as it’s a little off topic in here.

[AndrewFinlay]

I’ve been thinking a little bit about this feature and it seems to me that we really want is what nyc --all ... currently does. We should implement nyc instrument --all, then it’s quite reasonable to expect the two --all commands to behave in the same way. No need for a separate baseline directory, the filename can stay <uuid>.json and the processInfo files would still be there if needed.

This also simplifies the interface and the documentation.