Code review comment for lp://staging/~xaav/bzr/tarball-generator

[question] Given that wsgi does actually support having write() called repeatedly for streaming data, why not use it? I realize it's not the recommended interface, but it is the interface that does fit most naturally with the lower level tarball code we're using. It seems a bit like we're reimplementing something that could be better done in the wsgi framework -- specifically, turning push commands into pull. That would have the concrete benefit of allowing streaming just exactly as data is produced by the compressor, on finer granularity than exporting a single file.

It seemed like calling that method wouldn't support chunked encoding - it would instead store it in a buffer and calculate the content length. Also the documentation regarding chunked encoding isn't good, and this seemed the most appropriate way to do this.