Q: Use rebase to get around corrupted knit file?

[Kevin Cole]

Hi,

After searching the web, and a brief chat in IRC, I'm still a bit lost, though 
I think I'm closing in on an answer and looking for clarification here.

A while back, on a dapper machine, I had bzr 0.8.2.  It appeared to be doing 
things reasonably okay, until I tried to move the whole repository to a newer 
system.  Then, it told me a knit file was corrupt. Specifically:

| KnitCorrupt: Knit <bzrlib.knit._KnitAccess object at 0x8ebde0c>
| corrupt: While reading {...20070302012644-b26b4baa985aa020} got 
| error(Error -3 while decompressing: invalid distance too far back)

Eventually I got more details via:

cd .bzr
find . -name "*.knit" -exec gunzip -t {} \;

which revealed:

| gunzip:
| ./.bzr/repository/knits/75/types.html-20070302012458-1d7f486c93a59c92.knit:
| invalid compressed data--crc error
| gunzip:
| ./.bzr/repository/knits/75/types.html-20070302012458-1d7f486c93a59c92.knit:
| invalid compressed data--length error

types.html-*.knit also show up in /knits/24/ and /knits/90/.

Further investigation shows this file (types.html) to be part of the Python 
Library Reference, deleted from the repository ago (revision 3, according to 
the bzr log).  The repository is now up to revision 90.

Experiments with rebase so far have failed me.  How do I eliminate the first 
two revisions so that it starts with revision 3, where the log comment is that 
I got rid of the python library reference files?

Thanks!