2005-02-20

• Added documentation of patch pools, explication of commands in terms of file state transitions, uncommit, journalling through recording changesets. Feedback from Aaron on some of the design documentation.

• Much cleaner diff and status commands, based on a new Tree class that abstracts access to either a current working copy or a historical revision.

• remove command now both removes the working copy and updates the inventory, and has a new alias rm.

• Basic local versioning works well:

  % bzr init   
  % echo 'int main() {}' > hello.c
  % bzr add hello.c
  % bzr commit 'my test program'
  % bzr status
  U    hello.c
  % echo 'int main() { return 0; }' > hello.c 
  % bzr status                               
  M    hello.c
  % bzr diff        
  --- old/hello.c 
  +++ new/hello.c 
  @@ -1,1 +1,1 @@
  -int main() {}
  +int main() { return 0; }
  % bzr commit 'fix return code'
  % bzr log
  ----------------------------------------
  revno: 1
  committer: mbp@sourcefrog.net
  timestamp: Wed 2005-02-16 11:39:13.003095 EST +1100
  
  my test program
  ----------------------------------------
  revno: 2
  committer: mbp@sourcefrog.net
  timestamp: Wed 2005-02-16 11:43:13.010274 EST +1100
  
  fix return code
  % bzr status
  U    hello.c
  % bzr rm hello.c
  % ls
  % bzr status    
  D    hello.c
  % bzr commit 'this program sucks!'
  % bzr stauts
  bzr: error: unknown command 'stauts'
  exit 1
  % bzr status
  %
  

• Many internal/hacker commands: revno, get-inventory, get-file, ...

• Most of the implementation is split into reasonably clean objects: Branch, Revision, Tree, Inventory. I think these will give a good foundation for non-intrusive remote operations. They live in submodules of bzrlib.

596	bzr.py
608	bzrlib/branch.py
39	bzrlib/errors.py
25	bzrlib/__init__.py
320	bzrlib/inventory.py
126	bzrlib/osutils.py
66	bzrlib/revision.py
173	bzrlib/store.py
107	bzrlib/tests.py
62	bzrlib/trace.py
239	bzrlib/tree.py
41	bzrlib/xml.py
2402	total

• After feedback and reflection I have decided not to use SHA-1 hashes as the main identifier of texts, inventories, revisions, etc. It would probably be safe, but there is a certain amount of concern (or even FUD) about the security of these hashes. Since they cannot be trusted to be safe in the long term, it is better just to use something cheaper to compute, and equally unique in the absence of malice: UUIDs.

We now store the SHA-1 of files, but use it only as a check that the files were stored safely. At tridge's suggestion we also keep the size, which can catch some classes of bug or data corruption:

% bzr diff
bzr: error: wrong SHA-1 for file
'680757bf-2f6e-4ef3-9fb4-1b81ba04b73f' 
in ImmutableStore('/home/mbp/work/bazaar-ng/toy/.bzr/text-store')
  inventory expects d643a377c01d3e29f3e6e05b1618eb6833992dd0
  file is actually 352586b84597ea8915ef9b1fb5c9c6c5cdd26d7b
  store is probably damaged/corrupt

2005-02-27

Revisions are now named by something based on the username and date, plus some random bytes to make it unique. This may be a reasonable tradeoff against uniqueness and readability. This same id is used by default for revision inventories:

mbp@sourcefrog.net-20050220113441-34e486671a10f75297e03986

Keeping them separate is still a good thing, because the inventory may be large and we often want only the revision and not the inventory. It is possible for the revision to point to an older inventory if it has not changed (but this is not implemented yet.)

Tridge pointed out some possible performance problems with the assumption that we can simply compare all files to check for changes.

XML has newlines to make it a bit more readable.

Nested subdirectories are now supported down to an arbitrary depth.

bzr can successfully import and reproduce itself.

ongoing

• Systematic command-line handling and option parsing, etc.