Comment by luckydude

10 years ago

Each history file contains it's internal name, much like a file system has an inode # that is the internal name for that file. We call the names "keys" and you can dig out the inode key like so (every delta has a key, the first delta is called the rootkey):

    $ bk log -nd:ROOTKEY: -r+ slib.c
    lm@bitmover.com|src/slib.c|19970518232928|52808|f3733b2c327712e5

The key is user@host|relative path|UTC|checksum|64 bits of /dev/random and they are guaranteed to be unique if your DNS config is correct (we don't create duplicate keys, look for the uniq db in the code).

The way we version the entire tree is simple, it's logically (implemented differently for performance):

    <rootkey> <delta key>
    <rootkey> <delta key>
    ...

where each rootkey uniquely identifies a file and each delta key identifies the tip as of that commit.

Not sure if that is clear enough or not, ask away if not.

5 comments

luckydude

Reply
danielbot  10 years ago

Not tracking namespace accurately is Git's biggest weakness and would most probably be BK's best argument for getting a tryout. With Git, renaming a file and editing it at the same time tends to make a mess of the history and cause much pain. You can say just don't do that, but I say it happens, and it should just work.

  • luckydude  10 years ago

    That and a UI that is uniform and sub-modules that work correctly and a system that doesn't let you do things that mess up your data. Oh and blame that is fast:

      $ time bk annotate slib.c | wc -l
  18508

  real    0m0.031s

4ad  10 years ago

What happens when you duplicate a file? Does it track history back to the original parent? Can it help with future merges?

  • luckydude  10 years ago

    You mean copy the file? There is a bk cp command that copies the file and gives it a new rootkey. But from that point on the history of the two files diverge.