bzr too slow

[Denys Duchier]

Robert Collins <robertc at robertcollins.net> writes:

> I think TransactionFS as an approach is fine for local disk tasks, but I
> dont get the impression that its suitable for use with our transports
> that deal with sftp etc - where the link may go away at *any point* -
> that is, we must maintain a valid uncorrupt repository/branch no matter
> what stage the tcp etc link disappears.

The part where you are right is that TransactionFS was not specificaly designed
for a remote FS.  It is of course possible that my design exhibits infelicities
with respect to remote operations, but I think you may be jumping to hasty
conclusions before giving it a fair shake.

The point of the TransactionFS is to be exactly like the actual FS except that
modifications are made "to the side" (so to speak).  This is not any more unsafe
than what happens currently (where there is no way to prohibit concurrent
modification of an entire FS tree).

Cheers,

--Denys