Bugfixes: python3, s3 paths with subfolders

[rebkwok]

A few fixes for python 3 bugs and a couple of others I came across when upgrading to the latest master:

1. d98c9b0: YAS3FSPlugin. load_from_file tries to use e when raising the last exception, however it will never get to this point with e available, because it comes from a previously raised exception.
2. 95e9402: Wrap some dict keys in list() when iterating over them, otherwise it fails on python 3 because it's modifying the dict while iterating over it
3. bcbd913: Python 3 doesn't let you raise simple strings as exceptions any more
4. f4b8e39: The changes from commit 692e36c in the last PR merged introduce a bug if your s3 path has subfolders. (e.g. if my s3 path is s3://mybucket/foo/bar it will try to do tempfile.mkdtemp with the prefix 'mybucket-foo/bar-random' which fails because it's looking for a folder '/tmp/mybucket-foo/' which doesn't exists. This just replaces / with - in the s3_prefix so the temp folder created is 'mybucket-foo-bar-random'. Also updates the README for --cache-path to match the updated parameter help text
5. 9739764: make sure dirname in replace_empty_dirs is a str before doing replace(), otherwise raises an error on python 3
6. 40cb448: Fix inconsistent tabs and spaces introduced at some point which are making things break for me (yas3fs will not start up without this fix)

[rebkwok]

@danilop I'm leaving this PR open for review, but for our actual use we have reverted to commit c5cf4f8 (before PR 167 was merged), with the python 3 fixes from here, because we were running into various issues with the more recent changes on master. One fairly major concern was that the randomly named temp cache folders never get removed, and a new one is created each time yas3fs is restarted. Even if the cache is cleared, these (usually) empty folders apparently hang around forever.

[jazzl0ver]

Hello @rebkwok. Thanks for your commits!
I've already tried to fix this part and ended up with the following construction:

            data = self.cache.get(path, 'data')
            if data and not os.path.exists(self.cache.get_cache_filename(path)):
                logger.debug("Cache leak found for '%s', cleaning up..." % (path))
                self.cache.delete(path)
                with self.cache.lock and self.cache.new_locks[path]:
                    del self.cache.new_locks[path]
                del self.cache.unused_locks[path]

os.path.exists(path) didn't work for me well in some situations. See PR#168

[jazzl0ver]

@rebkwok , it would be great if you could fix the conflicts!

[rebkwok]

Conflicts fixed, but I'm still dubious about actually using this as per previous comment

[jazzl0ver]

Are you talking about the temp folders which aren't got removed? That might be similar issue to this one: #14

[jazzl0ver]

not sure this commit is required anymore - 399acdb

[jazzl0ver]

@rebkwok would you please create a new PR with your changes against the latest master?

[rebkwok]

@jazzl0ver I rebased this PR already when I fixed the conflicts, so it is against the latest master

[jazzl0ver]

@rebkwok, the point is that the items 5 and 6 from your PR seem to revert the changes I made before to fix some bugs. May I ask you to remove that items from this PR, so I could merge the ones that seem OK. For the items 5 and 6 - let's discuss them separately to make sure they won't break anything.

[rebkwok]

I have updated the PR description. The replace() fix (item 5) is still needed, but i have revised that commit. Item 6 fixes some inconsistent tabs/spaces - not sure when those were introduced, but they stop yas3fs from even starting up for me locally.