File.chunks fails to read the entire file for some file-like objects

[Erin Kelly]

The File.chunks method gets the size of the file and then decrements a counter by the chunk size for each read until it reaches 0. The problem is that the read method is not guaranteed to return the full amount requested, and so the amount read may be less than the chunk size, resulting in the counter being decremented erroneously. I ran into this bug while using a file-like object that wraps a java.io.InputStream object, whose read method returns as soon as data is available.

This patch is one possible solution, although it still has the problem that it will result in a busy-loop if the read method is entirely non-blocking. It also potentially yields non-terminal chunks smaller than the chunk size, which may not be desirable.

See also #9632, which may affect the implementation here if it is accepted.

[Alex Gaynor]

This patch doesn't look right, it should be len(chunk , not chunk_size. Also I'm pretty sure this is a dupe.

[Erin Kelly]

Corrected patch

[Erin Kelly]

You're at least right about the first part. That's what I get for once again writing a small patch without testing it. I've uploaded a corrected patch.

[Karen Tracey]

#13756 is very similar (may be the dupe Alex was thinking of)...same issue but different bit of code.

[Alex Gaynor]

That is indeed what I was thinking of.

[Daniel F Moisset]

Patch is ok, but needs a regression test.

(The issue on non-blocking read happens in the existing codebase, and should be treated as another ticket if considered a problem... although using a File with a non blocking read source looks like a bad idea anyway)

[Aymeric Augustin]

Change UI/UX from NULL to False.

[Aymeric Augustin]

Change Easy pickings from NULL to False.

[Claude Paroz]

The implementation of the chunks() method has changed in r17871, to not depend on size anymore. Hence rendering this ticket obsolete.