Producing release notes at the end of the release is a becoming more of a problem as the project grows. I've been doing this by going through all of the LLVM commits since the last release and pulling out things that I find interesting and potentially impactful to people. However, as the project grows and there are more commits, this has been becoming less and less practical.
To solve this, I'd like us to move to a more incremental release notes model. Going forward, if you make a change that should be described in the release notes, please add a short blurb about it to llvm/docs/ReleaseNotes.html. By the end of the release, we'll have lots of content in the file, and I'll edit it and pull it together into something coherent. If there isn't a good place for some note that you'd like to add, just add it anywhere and I'll deal with the editing at the end of the release.
I think that this will be pretty straightforward, but there are a few logistical issues that I'd like to point out: first, this is a new process, so it's likely that people will forget to update the release notes sometimes. I figure that if we can remind each other to add testcases and update LangRef that we can remind each other to update the release notes too :).
Second is stuff that has already happened in 3.1 but that is not in the release notes. For this, I plan to go back through all the commits between r142037 (3.0 branch) and r146493 (now) the old way to add them to the release notes. If you know of anything specific that happened (e.g., a few targets got zapped!) feel free to add some notes to the release notes now.
Finally, when 3.1 branches, I plan to immediately clean out the contents of the release notes on mainline - The 3.1 release notes can be produced and edited on its branch, instead of doing it in mainline. That way, the inevitable deluge of stuff that happens right after the branch can go right into the release notes for LLVM 3.2.
Does this seem like a sensible process?