Re: [Monotone-devel] Thoughts about 'testresult'...

[Nathaniel Smith]

On Thu, Apr 21, 2005 at 10:03:16AM -0700, Emile Snyder wrote:
> Ahhh, yes, this clarifies things.  In my mind, revision selectors should
> always be trying to find the single revision that meets your selection
> criteria.  There might not be such a single match, and then they should
> complain, and give you a list.

Right.

> What we would like is two new things: a way to specify a testresult
> selector (or possibly just a general purpose cert selector a'la
> c:certname=certval) *and* a way to say "take the latest rev that
> satisfies this selector."  The latest rev part might still be ambiguous
> (multiple heads) but then it just give you the list of possibilities
> like usual.  This could be implemented as a flag, or as an extension of
> the selector syntax (capitalize the selector character?  add an 'l'
> after it?)
>
> So al:richard or A:richard or '--latest-selector a:richard' would be the
> latest rev authored by richard, and rl:address@hidden would be the latest
> revision with a true testresult cert signed by richard.

Random idea: adding an "h:" selector, meaning "head", is on the todo
list (I think someone was working on it?).  Could make the selector
syntax recursive, so h:a:richard means "head of the set of revisions
specified by a:richard", i.e., the latest.

This is probably a terrible idea, though, because then people will
want a way to distinguish between h:(a:richard/b:net.venge.monotone)
type parsing and (h:a:richard)/b:net.venge.monotone type parsing.

Designing minilanguages is hard :-/.  Especially the "figuring out what
'mini' should mean" part...

> Does this make sense?
> 
> But as a separate issue, is there anything that we would want the
> --cert=certname:certvalue flag that works like the --branch flag (in
> fact, --branch would just be a special case, --cert=branch:branchname)
> for?  Or am I just inventing useless features?

I don't see this as particularly useful.  --branch is not used to
restrict to a given branch, primarily; why it's special is that it's
used to determine, e.g., what sort of branch certs monotone _issues_,
what branch monotone should default to when it needs to set a branch
on a new working copy, that sort of thing.  There are some places like
"heads" and "update" that use it in a restrictiony kind of way, but
these seem to be special cases, not the main idea...

It seems like what you want would be better implemented as a
c:certname=certvalue selector?  That's been proposed on the list
before, even, just nothing ever happened.

-- Nathaniel

-- 
"...All of this suggests that if we wished to find a modern-day model
for British and American speech of the late eighteenth century, we could
probably do no better than Yosemite Sam."