How should we set up version control (specifically with Subversion)?
use the guidance provided in the documentation for Siebel Tools. But please note that Siebel does not build from the files in SVN so it will only be useful as an archival tool; you cannot manage your code or build from SVN.
What kind of structure should our repository have? How should we handle branches and tags?
Siebel development code is not built or managed in SVN so this is a pretty useless thing to do. Just note the date that you built your SRF and exported your Repo and match with a tag or branch in SVN.
How can we do code reviews? How can we peer review configuration changes made through Siebel Tools that don't necessarily have any "code"? We want to review these changes for quality assurance and knowledge transfer, as well as compliance with change management policies.
Use Siebel Tools to do this. It has a built in 'checking' tool for obvious errors (all devs should be using this before they check in) and a diff tool (you will need to check against an older version of the same object - which you could drag out of SVN if you want). I normally automate the checking tool once a day and review the output logs, and automate build from the Siebel server 5 times a day and look for errors during the compile. Diffs via SVN and a standard diff tool might be possible, but the Siebel objects are stored as XML-like files in SVN and so are hard to read sometimes.
What sort of change management works well with Siebel? How do we verify that only things listed in our change log are actually changed when we do a new deploy?
?
How can we automate testing of our application? Is unit testing even possible with Siebel? I saw another question suggesting QTP for web testing, but are there other options that work?
QTP is the standard way to go - check on the Oracle web site for other vendors that they may recommend. You could also try Sikuli.
Are there other things we can do to implement Continuous Integration practices with our Siebel development efforts?
Not really.
What recommendations do you have for naming conventions and other things that would traditionally fall under "coding style" guidelines?
Checkout the appropriate section of Siebel Bookshelf for current naming guidelines and use these always.
How should we separate development roles from Siebel Administrator roles?
Not sure what you mean.
What should our build/test/deploy cycle look like?
Build a new SRF and export a new Repo from Dev once a night. Once all the dev work has been checked-in and unit tests are done take the next SRF and Repo and push into the test environment. At this point in normal software development you'd branch your SVN and continue to develop on the trunk but Siebel is different because you cannot build from SVN and you cannot easily restore a whole lot of files from SVN into your build environment, so you're best to make hot fixes for test either in dev (and pause mainline dev development until that is done) or in the test environment, and do ugly backports to the development environment (that's what most people do in fact). Build a new SRF and export a new Repo from Test once a night and once that's good, snap a copy for your Production release.
Try to stick to cycles of no more than 4 weeks (1 week for desing/prototyping. 1 week for dev, 1 week for test, 1 week for bug fixes and deployment) - any longer than that and the overhead of planning will become too great.
Hints for an easier life: Avoid eScript except in Business Services (otherwise it becomes unmanageable); use all the Siebel built-in tools instead of rolling-your own; try to avoid any roll-up functionality (it always seems like a good idea but it always destroys performance); keep the number of screens and views to an absolute minimum; do not build views when you should be building reports instead; always make sure that EIM tables match and schema extensions that you make - even if you don't use EIM right now; try to build Integration Objects to match your logical schema - they are always useful (for web services, XML publishing) and a hell of a job to build after the fact; prefer Workflow Policies over run-time Events; don't add new sort or search specifications without indexes - ever ever ever; don't make by-reference links to the LOV table; always patch; if the vendor doesn't say that you can do something, never do it.