on BlobFile.

- Implement __del__ on BlobFile, which attempts to close the filehandle.

- Use a weak reference to the blobfile in the BlobDataManager so as to
  not keep the blobfile alive for longer than necessary (mainly to support
  the idiom of opening a blobfile without assigning it to a name).
  It is no longer necessary to explicitly close a blobfile.

- Raise IOError if an invalid mode is passed in to Blob.open.

- Add some comments about why things are the way they are.

 

Collector #1734.  Critical bug in BTree conflict resolution.

Stop silent data loss in some BTree cases where a transaction adds
a new key to a bucket while a concurrent transaction deletes all
keys from the same bucket.

Still needs porting to 3.2 line.

 - I made the IStreamIterator not gently close itself, but rely on e.g. Medusa
   closing it.
 - The tests now document the behaviour that a user should close the file handles.

Refactor some behavior of blob and blobfile (transparent), account for append mode not failing if a file doesn't exist (it acts like write mode in this situation).

Clean up formatting to 80 cols and add a restriction noting that BlobFiles can't live beyond the transaction in which their "open" call was invoked.

Stop silent data loss in some BTree cases where a transaction adds
a new key to a bucket while a concurrent transaction deletes all
keys from the same bucket.

Still needs porting to ZODB trunk and to 3.2 line.

 - Explicitly allow the different modes to be selected when opening a blob file.

 - added clean(er?) handling on cleaning up blobs on transaction boundaries and
   beeing more careful not to cause harmful states when leaking BlobFiles out
   of threads/transactions (e.g. to Medusa)

 - Added Import/Export support. Actually this is even backwards compatible.

Various Zopes probably need changes too (to avoid the new deprecation
warnings).

 - merged from head
 - added configuration methods to configure a blobfilestorage
 - made stuff work ;)

Jim (Fulton) points out that it's not entirely clear that
the new format can't be read by older ZODBs, and Tim doesn't
think it's worth the bother to find out for sure.

Added news blurbs about fsIndex improvements, and about that
.index files written by 3.4 won't be readable by earlier ZODBs.

test.py ran these tests ran after "setup build_ext -i", but
not after "setup build".  Repaired by teaching setup.py to
copy more stuff out of the relevant directories.

There's a mystery:  the tests for it appear to run after
"build_ext -i", but not after "build".  But that appears
to be true for the long-stitched-in zope.interface too.

trunk, so remove the branch (it's an attractive
(OK, maybe not all _that_ attractive <wink) nuisance
now).

This introduces a "multi-database" concept (a simplification
of Jim's Wiki proposal), and adds many interface definitions.

Work done during the PyCon 2005 ZODB sprint, by Christian
Theune, Jim Fulton and Tim Peters.