Empty file handing

[bblay]

A fix for #101 (replaces #165).

[rhattersley]

I'm not entirely convinced this qualifies as a fix to "Avoid obscure crash...". Invoking iris.load('obs.pp') on an empty file results in a large nine level traceback which finishes with:

  File ".../iris/io/format_picker.py", line 242, in _read_n
    raise iris.exceptions.TranslationError("Unexpected end of file: '{}'".format(filename))
iris.exceptions.TranslationError: Unexpected end of file: 'obs.pp'

Not exactly friendly. Someone unfamiliar with the code won't know the significance of the error:

• it's emitted from deep in the call stack;
• and the message doesn't clearly explain why the error occurred.

If we have to emit an error, could we emit it from iris.load instead?

Also, there's no test. ;-)

[bblay]

rebased

[bblay]

rebased onto latest master for more thorough testing

[esc24]

Please consider use the temp_filename() context manager for this. It avoids the need to remove the file.

[rhattersley]

If you're seeing a lot of weird commits in this PR it's because we removed a commit from master and some of the PRs have got themselves confused. A rebase will clear the problem.

NB. It doesn't need a rebase, you can create a new PR with exactly the same commit (i.e. pelson/iris@2bcf972) and it will be just fine. It might be possible to move the head of your PR somewhere else and then back again... but caveat emptor!

[bjlittle]

Extra blank line please, while you're there!

[bjlittle]

Could you please squash your commit history, thanks.

[bjlittle]

👍 Lovely work @bblay, it's good to go!

[pelson]

TranslationError seems a bit strange in this context. I would have said an IOError was better, but it doesn't madly matter now its been merged.