Fwd: Re: How hg signals error and aux messages

[Matt Mackall]

On Tue, 2013-01-29 at 15:25 +0000, Jan Vrany wrote:
> Yes, I've read this, but unfortunately I haven't found answers
> to following questions:
> 
> - Is an error alwyas indicated by a line to stderr (or error channel)
> starting with 'abort: '?

Errors are always on stderr. They may or may not be prefixed with abort.
Warnings also appear on stderr, usually prefixed by "warning:".

> - Is an error always indicated on stderr (or error channel) before
> command output?

Generally -after-.

> - Where are auxiliary messages such as 'invalidating branch cache (tip
> differs') written?

Generally stdout.

> Also, I haven't found command's output "grammar" - how the output
> (stdout or 'o'utput channel) of an hg merge command (for instance)
> look like.

The code and the test suite is the current spec. If you'd like to turn
that spec into a format you think is more convenient, be my guest. It
doesn't see to be a priority for anyone else.

> Actually, I do have a feeling and my brief experiments shows,
> that command server solves only the problem of startup (and
> maybe language neutrality) and does not specify the output of
> individual HG commands like merge,  push, update. Am I wrong?

Not precisely. It is our position that the command line is and has
always been the defacto public, stable API. It's thus not the mission of
the _command server_ to document the API, just to make it available more
conveniently.

Recommend you look at python-hglib for an example of how to translate hg
output into native data structures. We continually regression test
python-hglib against multiple versions of hg.

http://www.selenic.com/repo/python-hglib

-- 
Mathematics is the supreme nostalgia of our time.