Yay, set up a Trac instance for the Linux kernel tracepoint stuff.
I'm working on adding support for GDB tracepoints to KGDB, a GDB stub for the Linux kernel. Essentially, a tracepoint is like a breakpoint except that, when you hit a tracepoint, the stub records some data that you specify (using arbitrary C expressions; you can deference pointers, index arrays, refer to structure members, etc.) in a log, and then immediately continues the program, without communicating with GDB. Then you can examine log hits with GDB; GDB acts as if the selected hit's recorded data were "live". If you log a hundred bytes or so off the top of the stack, you can even get backtraces.
I'd like to get a Trac site set up for it, so people can see what I'm doing.
I wonder --- would it be possible to provide Dwarf CFI data for the IRQ handlers, so you could unwind through IRQ's?
This week goes mostly to Red Hat --- I need to get a GDB release together for a customer.
However, I think I've figured out how to present a pretty nice interface in the Subversion filesystem library for building transactions. A transaction will behave just like a revision: it's a directory tree, which you can browse using the normal filesystem API. However, unlike a revision's tree, which is permanent and unchanging, a transaction's tree is mutable --- you can use additional functions to create, delete and modify nodes as you please. When you've munged the tree to your satisfaction, you can commit the transaction; if there are no conflicts, the transaction's tree becomes a new revision of the filesystem.
This will require some fancy logic, mostly to conceal the sharing of nodes between a transaction and extant, committed revisions, but it should make the interface consistent and easy to learn. And hopefully make it simpler to implement WebDAV's `activities' on top of the Subversion filesystem.
Keep up with the latest Advogato features by reading the Advogato status blog.