[rfc] [patch] pycurl transport

[Johan Rydberg]

Martin Pool <mbp at sourcefrog.net> writes:

> Right, so we could turn a 
>
>    seek()
>    read()
>
> into a byte-range request.  But perhaps it would be better for Transport
> callers to use something like a pread interface, where they give the
> offset and length all in one go.

In my versionedfile, IIRC, the contract for Transport.get was that it
should return a file-like object which at least has "read" and "seek"
methods.

Maybe we could just extend that contract and force the transport to
return a file-like object that has a "read_block" method, that takes
a list of byte ranges and yields the data (in the same order as the
ranges.)   Sounds simple, and should work for most protocols.

~j