[Monotone-devel] Sticking my nose into the hash controversy

[Joel Crisp]

Hi Guys

Time to unlurk on this one I think.

As a 4+ years experienced Clearcase admin, I don't actually think the 
hashes are that bad. The number of times I find myself using Clearcase 
UUIDs to specify things is quite large - many of the 'disaster recovery' 
options required UUIDs. In the end, it comes down to cut-and-paste, and 
having some simple way to find the UUID which is the relevant one - 
often from either an error message or one of the ubiquitous ls* 
(lsvtree, lsstream, lsbl, etc ) commands which show information about 
metadata entities in the UCM(*) project vob. The dot based graphing GUI 
(lovely!) seems to be the best option for comprehending archive 
structure and providing a source of selecting hashes. (BitKeeper also 
provides a similar GUI for changeset dependencies as well as file 
version tree). Most of the developers where I work just co/ci using the 
SSC integration with JDeveloper, so never care about version tags. For 
communication, we tend to use baseline labels, which need to be human 
readable and understandable, but can also be human assigned.

(*) UCM is the Clearcase Unified Change Management 
projects/streams/components management which adds activities aka 
changesets to Clearcase

On a slightly different tack:

We just had a real tarball problem with Clearcase. The problem was that 
we have a three stream hierarchy of a global integration stream, then a 
number of team integration streams, then a number of developer streams 
at either one per developer or one shared between a small group of 
developers. (Each stream is backed by a clearcase branch type, which is 
a branch specifier shared over multiple files). Developers make changes 
grouped into activities (changesets) then deliver them to the team 
stream, team leaders ensure stability in the team stream then deliver to 
the global stream. One team had a _very_ large number of changes, both 
file and lots of directory deltas. When we tried to roll their team 
stream up the global one Clearcase barfed. It appears it tried to merge 
just the tips of the entities in the team stream onto the global branch, 
and was getting confused as the checkedout directories on the 
integration stream had the same file linked in lots of places due to all 
of the directory structure changes! We ended up having to export, drop 
the stream and re-import to fix it - which is a bit of a nuclear option!
One option would be to provide an incremental deliver operation. Not 
sure how relevant this is to monotone, but its the first time I've seen 
a problem this nasty with Clearcase so I thought it would be worth 
highlighting.

To summarize a few points:
1) I don't think hashes and hash notation is a big problem. I have a 
slight preference for the separation of the hash into groups of 4 hex 
digits.
2) Providing a good source of cut-and-paste for hashes would go a long 
way to alleviate the issues around unfriendly identifiers.
3) Developers using an IDE probably don't care anyway.
4) Watch out for merging of very divergent trees
4) Publishing test code coverage figures a la svk would be a good thing ;-)
5) No-one is going to adopt a new SCM system without confidence that it 
does the right thing - which means either a large established user base 
or very comprehensive test suite and proof that the test suite covers 
all of the common cases and most of the uncommon ones.

Hope this is helpful

Cydergoth