bzr branch is slow and memory-hungry (just a data point)

[Aaron Bentley]

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

Robert Collins wrote:
>>The branch process takes 4.5 minutes, during which the bzr process
>>dominates the CPU (not that there's any competition on my laptop) and
>>balloons to 425MB.
>
>
> That sounds like there is a leak of some sort there. By leak I mean
> references we dont need being kept around. It would be good to track
> that down and fix it.

That depends on what tree was.  If it was the linux kernel, it is
supposed to be ~ 206 MB, the aggregate size of the file contents.  This
is because merge changesets are represented in-memory.  That was
originally due to your suggestion.  I believe you said it was important
for embedded developers to have changesets represented in memory.

I think we can reduce the storage requirements considerably by replacing
FileCreate with TreeFileCreate, which would defer reading the file
contents from the tree until it was time to write the file.

It is important to implement equality comparisons, but this can be done
using the sha1 sum.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDH04V0F+nu1YWqI0RApzwAJ41SwdFvCdcuMsJfObvvnFhpwTB4gCeI/E2
DbKTYMsy4dFMietjsj9IMtI=
=ywNR
-----END PGP SIGNATURE-----