910614 - (log-SnR) [meta] Improve the SnR of the log error summaries by reducing spurious errors in the log

Status: NEW

(Testing :: General, defect)

Description

[Ed Morley [:emorley]]

When a job runs, buildbot uses a number of criteria to determine whether it passed or failed (and if failed, with what state). TBPL imports this result & then parses the stdout using another set of regex [1] in order to generate a failure summary - which is what is displayed when clicking on the job in the main UI & is also used to generate the search terms for the bug suggestions for intermittent failures.

However, there are times when this summary's SnR is pretty poor:
1) Certain strings in the log are falsely matching against the TBPL regex, even though they weren't the real failure that caused the run to not succeed - distracting people from the real failure lines.
2) Some failure modes result in many garbage lines that don't add much to the understanding of the failure (eg return codes when we already have a stack, or logcat entries that are identical to the already output failure line).

This bug is to track reducing the cases of both #1 and #2 - and will primarily need changes to test harnesses/parts of automation as opposed to TBPL (since the regex is already about as good as we can get it).

[1] https://hg.mozilla.org/webtools/tbpl/file/tip/php/inc/GeneralErrorFilter.php