storage servers should announce that they support over-read

[zooko]

Since 6c4019ec33e7a253, storage servers have responded to an over-read (a read past the end of the data) by returning the valid span (up to the end of the data). Before that they raised an exception on over-read. Clients could use over-read for faster and nicer immutable file download, and the code to do so is in place in download.py and was tested, but is currently disabled. This ticket is to make storage servers announce that they support over-read by adding a field to their version structure -- src/allmydata/storage.py@3405#L957. This would enable clients to use over-reads on servers that had declared themselves as supporting it.

[zooko]

I'm planning to do this for tahoe-1.3.0 if Brian (or someone) doesn't beat me to it.

[warner]

I'll grab this one. Patch is in 7cfc74bcc99d04be.

For future reference, 0eb6b324a4fcda2f (on 21-Nov-2008) added version-announcement to the storage server, and 53b28c1650c74624 (on 03-Jan-2009) added the over-read tolerance.

The boolean flag I'm adding is named "tolerates-immutable-read-overrun", so client code should do the following:

v1 = rref.version["http://allmydata.org/tahoe/protocols/storage/v1"]
if v1.get("tolerates-immutable-read-overrun", False):
    use_overrunning_reads()
else:
    avoid_overrunning_reads()

I've updated VERSION_DEFAULTS, so modern clients will see the following values for get("tolerates-immutable-read-overrun"):

• before 0eb6b324a4fcda2f: False
• at or after 0eb6b324a4fcda2f, before 53b28c1650c74624: None
• at or after 53b28c1650c74624: True