--only-findings should not omit scan_errors from JSON output

[sschuberth]

When running ScanCode 2.2.1 with --only-findings and a scan error like exceeding a given --timeout occurs, that error is not written to the JSON output. Instead, the file is treated as if no findings occurred and is completely omitted from the output due to --only-findings.

I believe scan errors should be included in the JSON output despite --only-findings being specified because if there are errors we in fact do not know for sure if there were findings or not, hence it's unsafe to exclude those files when --only-findings is given.

[pombredanne]

@sschuberth excellent point.... Now one thing to consider too would be to reform how errors are reported. One idea would be to have a single "header-level" data structure that would contain errors (and could be keyed by file path) and therefore NOT report errors with the file data at all, but only once in a central place? What do you think?

[sschuberth]

I believe having both would be beneficial: We need the per-file scan_errors so we know for which files errors occurred, but also having a top-level error_summary (?) which basically just is a de-duplicated set of all scan_errors for all files would be nice to have to get a quick overview of what type of errors occurred. For example, if that error_summary only contains "ERROR: Processing interrupted: timeout after X seconds. we might chose to ignore these errors since we're in control of X and have chosen its value as a compromise between scanning speed and scanning completeness.

[pombredanne]

Excellent point again 👍

[sschuberth]

The main issue has been fixed via #848, thanks @haikoschol! The additional idea that came up of having an error_summary (?) should be dealt with as part of #211. So I'm closing this.