[llvm-dev] Suppressing lit's non-error logs when --quiet is specified

[Nate Chandler via llvm-dev]

Lit can accept a --quiet/-q flag which is documented to "suppress no
error output".  If I'm understanding correctly, the intent here is to
suppress both notes and warnings and only show errors.

Such a behavior would be handy.  Editors could pass the --quiet flag to
specify that they are only interested in getting logs for errors, and,
when encountering errors, use the location data in lit's output to allow
highlighting and jumping to the problematic lines.  For example, the
output could populate vim's quickfix window.  While this is possible
currently, there can be a number of locations which must first be
skipped past before getting to the locations of interest.

Currently, however, the flag does not necessarily have that behavior.
Maintainers of lit.cfg files must manually check the flag to determine
whether to call LitConfig.note and LitConfig.warning.  Here's an
example: https://github.com/apple/swift/blob/master/test/lit.cfg#L246 .
It is possible ( https://github.com/apple/swift/pull/27361/files ) to be
complete about this, but it would be nicer to just check the flag inside
the function being called.

Here ( https://reviews.llvm.org/D68044 ) is a patch which implements
this behavior, tweaking LitConfig's note and warning methods to check
the quiet flag before writing the message.


Would suppressing notes and warnings in lit itself break users of that
flag and/or is this in fact the intended behavior of the flag?


If --quiet is intended to have a different purpose, then it would be
nice to add a different flag ( "--print-errors-only" or similar) which
does suppress notes and warnings.


Nate