Fixing Windows issues

[instantname]

Here are a few fixes that make python-gnupg work on Windows.

The main change is about paths: the strategy in this PR is to escape the paths as late as possible in the code.

According to my tests, this solves #199 and #204 on Python 2 & 3.

Below is a summary of the tests results. On Windows, since the GPG class could not be instantiated before, the tests are necessarily better... On Debian, some tests are already failing on the master branch and many even make the process hang (example with "test_gnupg.py crypt" in note [11] of attached file). When removing the ones that make the tests hang, I saw no notable difference between the tests on the master branch and this pull request.

Tests summary (see attached log for the numbered notes):

Python 2.7.13, GnuPG 1.4.22, this pull request (ea96545):

• parsers: OK
• encodings: OK
• basic: OK
• genkey: OK
• sign: 2 fails out of 11 [1]
• crypt: 7 fails, 1 error out of 18 [2]
• listkeys: OK
• keyrings: OK
• recvkeys: OK
• revokekey: 2 errors out of 2 [3]
• expiration: hangs [4]
• signing: OK

Python 3.6, GnuPG 1.4.22, this pull request (ea96545):

• parsers: 1 fail out of 6 [5] (related to issue PY3: 2.01 - gnupg stream decode error when decrypting without asciiarmor; gpg --decrypt works #102, I think)
• encodings: OK
• basic: OK
• genkey: 1 error out of 11 [6]
• sign: 2 fails, 1 error out of 11 (similar to [1], plus [7])
• crypt: 7 fails, 1 error out of 18 (similar to [2])
• listkeys: OK
• keyrings: OK
• recvkeys: OK
• revokekey: 2 errors out of 2 (similar to [3])
• expiration: 3 errors out of 4 [8]
• signing: 4 errors out of 4 [9]

Debian Jessie, Python 2.7, master (705d7f4):

• parsers: 1 fail out of 6 (similar to [5])
• encodings: OK
• basic: OK
• genkey: 7 hangs out of 10
• sign: 7 hangs out of 11
• crypt: 7 hangs, 7 fail out of 18 ([10], pretty close to [2])
• listkeys: 1 hangs out of 1
• keyrings: 1 hangs out of 7
• recvkeys: 1 hangs out of 1
• revokekey: 2 errors out of 2 (similar to [3])
• expiration: 4 hangs out of 4
• signing: 4 hangs out of 4

Debian Jessie, Python 2.7, this pull request (ea96545), excluding the tests that hanged above:

• parsers: OK
• encodings: OK
• basic: OK
• genkey: OK (excludes 7 tests)
• sign: OK (excludes 7 tests)
• crypt: 7 fail out of 11 (excludes 7 tests, similar to [10])
• listkeys: OK (all tests excluded)
• keyrings: OK (1 test excluded)
• recvkeys: OK (all tests excluded)
• revokekey: 2 errors out of 2 (similar to [3])
• expiration: OK (all tests excluded)
• signing: OK (all tests excluded)

pr1_log.txt

[isislovecruft]

Hi @instantname! Thanks for the patches, and my apologies for taking so long to get to this.

[isislovecruft]

I would prefer it if we could instead find a way to use a shell input sanitisation method which doesn't break things on Windows. Maybe checkout what the shlex module does on python3 a Windows machine? (Not sure if that'll tell you anything useful, it's just an idea.)

[isislovecruft]

I'd strongly prefer to fix unsafe strings as early as possible to prevent potential problems while working with them later. (For example, this is part of what makes this library not vulnerable to SigSpoof.)