Branko Čibej <brane_at_wandisco.com> writes:
> Evgeny, please take a look at r1684940.
>
> I don't really like the fact that with this change and 'svnadmin
> --keep-going --quiet', the text "Error verifying revision N" gets printed
> to stderr; but I couldn't think of a better way to join the error to the
> revision it was emitted for. I'd love to find a better solution.
After giving this change a look, I cannot get rid of the feeling that what we
are doing here is just adding different workarounds for the API issues, e.g.,
booleans like b->silent_errors or b->silent_running. I should also say that
I find the b->feedback_stream manipulations questionable, because now, for
instance, we write warnings to either stdout or stderr depending on the --quiet
argument. I always thought that --quiet should only suppress a part of the
output, but shouldn't really retarget it to a different standard stream.
With that in mind, I am -0 on the corresponding backport proposal.
I sketched the other possible option with redesigning svn_repos_verify_fs3()
API to only report errors via the notify_func(), please see the attached patch.
Although I won't insist on going this way, I like it better, as it allows us
to get rid of things like b->silent_errors, b->silent_running, juggling with
standard streams and API that can yield the same error in two different ways.
What do you think?
Regards,
Evgeny Kotkov
Received on 2015-06-15 20:55:03 CEST