Arvados

• As discussed briefly in IRC, consider having slightly different behavior for StreamFileReader() based on how it is accessed. If accessed using existing Arvados methods, preserve existing behavior; if accessed using the new open() method, tweak the behavior to be maximally compatible as a Python file-like object.
  • For example, the behavior of seek()
• Going down the documented API for file-like objects, there's a few small things that could be added:
  • A no-op flush() method
  • readline() takes an optional size parameter (maximum line length)
  • readlines() takes an optional size hint (approximate amount of total data to be read)
• Regarding isatty():
  "If a file-like object is not associated with a real file, this method should not be implemented."

Peter Amstutz wrote:

• As discussed briefly in IRC, consider having slightly different behavior for StreamFileReader() based on how it is accessed.

We reviewed existing Crunch scripts, and found none that are using seek, or obviously relying on the decompression behavior in readlines. We'll take this opportunity to make a clean break and have our API be compatible with Python's as much as possible. All kinds of stuff is using the name() method, so we need to preserve backwards compatibility there.

• Going down the documented API for file-like objects, there's a few small things that could be added:

Implemented all of these. Even made flush() meaningful.

• Regarding isatty():
  "If a file-like object is not associated with a real file, this method should not be implemented."

Thanks for catching that. Removed. Ready for another look at 409a1a9.