Re: [Monotone-devel] restriction fixes

[Stephen Leake]

Derek Scherger <address@hidden> writes:

> Stephen Leake wrote:
>> Derek Scherger <address@hidden> writes:
>> 
>>> BTW, has anyone looked at this branch? I think it's an improvement
>>> and fixes a couple of invariant failures. Please see my earlier
>>> emails for details.
>> 
>> I just looked thru the log and the code. I gather the main point is
>
> Thanks!
>
>> the function make_restricted_roster. There are no comments saying what
>> it does or why it is there (vis-a-vis Richard's recent complaint :).
>
> My attempt to describe it is here
>    http://www.mail-archive.com/address@hidden/msg09026.html

Ok, that's clear.

> Yes, the point is to fix a specific problem with renamed directories.
> The *result* is the change from make_restricted_csets to
> make_restricted_roster.

Ok.

> I don't recall anything specific from Richard on this. Are you
> talking about the thread on commit/ChangeLog messages?

Yes. 

This seems like a good case for design comments in the code. Just
copying the text of the email would be pretty good.

>> It may be that automate inventory could use make_restricted_roster
>> rather than doing the same work itself. It depends on whether the
>> make_restriced_roster handles the renames (and other off-nominal)
>> cases the way inventory needs to. That definitely should be stated in
>
> It handles them *correctly* unlike make restricted cset in some cases!
> I'm not sure what would be special about what inventory needs?

I'm not sure either. I am sure that it is entirely possible that what
is "correct" for one use case is not "correct" for another, so I'd
have to test.

We should probably add the use cases in tests/restricted_diff_bug and
tests/fetching_nonexistent_entry_from_children_bug to the
automate inventory tests. I'll put that on my list.

-- 
-- Stephe