Code review comment for lp:~maria-captains/bzr/bugtracker

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 3/21/2011 2:15 PM, Jelmer Vernooij wrote:
> On Mon, 2011-03-21 at 12:32 +0000, Sergei wrote:
>> On Mar 21, Jelmer Vernooij wrote:
>>> bzr uncommit only really changes the branch tip, it doesn't change the
>>> working tree or the repository. Either way, an exception being raised
>>> at a certain point should never lead to data loss or corruption,
>>> especially for things that are just used for display. If you can still
>>> reproduce the corruption issue, please file a bug.
>> this needs uncommit to be aborted in a specific place, which cannot be
>> done without modifying the code (bzr or bzr-gtk, or a special plugin).
>> I don't know whether it'll qualify as a valid bug report.
> Even if uncommit crashes it should not ever leave a broken tree around,
> as it has no business touching the working tree.

It actually does have to touch the WT state, because dirstate stores the
basis tree for faster status comparison.

For that, though, you'd have to die while updating the state, which
shouldn't be easy (and it should be atomic, either you update
successfully, or we don't change the file on disk.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk2HV50ACgkQJdeBCYSNAANsagCggIr37DxDfjl/hi1jtYEz8oc3
Cs0AoKD49Dkw9KJOb06C+frQwXQ1yKqB
=TMUw
-----END PGP SIGNATURE-----