Re: [Monotone-devel] RFE: Hard Barrier Between Branches in Netsync

[Joel Crisp]

Not really commenting on anyone else's comments, but hopefully some
benefit of experience:

Just for a (slightly extreme) real world example: The company for
which I work holds *all* its source code in one database. _One_ of our
products is over 30G of source history. We have over 5000 developers,
and a large portfolio of products. The IT support team wrote their own
in-house SCM system after Clearcase failed to cut the mustard. The CEO
has mandated a central source code control as it is one of the
company's major assets. The system supports many read-only replicas
for performance, but only one database is able to accept commits to
the mainline. This is partly for compliance and audit reasons.

One key aspect of this system is managing the myriad and complex
dependencies between products, we have a lot of infrastructure common
across many different products, such as a J2EE server and associated
heavyweight frameworks. Each of these is principally worked on by a
separate team, but many teams need access to import the source of
pre-production or early release versions (just binaries isn't good
enough for advanced debugging). Yet each team should only be allowed
to write to the code for which they are responsible.

In addition, and currently not well enforced, there are sections of
products which should be restricted further - a recent example is the
configuration file parser as all configuration options need to be
appropriately documented by the documentation team, including
translations etc. Enforcing this is tricky in many SCM systems, as you
really need an access control system which supports delegation and
escalation (e.g. in case someone critical is off sick). Triggers go
some way towards fixing this issue.

I just though this might feed into the current discussion and the
policy branch stuff a bit.

Joel