Working towards dual python2/python3 support. #1972
Conversation
|
Ping. Any update on merging this? |
|
Hi @eli-schwartz . Thanks for the patch. I just made a note of it in issue #1176 . Have you been able to successfully use Zenmap on a python3-only system with this patch? I was under the impression that we would need a more fundamental overhaul because I thought some of our libraries (pygtk in particular, if I recall) wouldn't work with Python3. |
But it's a terrible way to discover if something exists, anyway (e.g. empty strings evaluate false), so let's switch to something simpler. NmapOptions already imported functools.reduce, but this is once again an overly complicated way to detect the existence of an option.
The code still runs under python2, but, where trivially possible,
python2-only code is converted to code that runs on both python2 and
python3.
Changes made with the command:
find . -name \*.py -exec python-modernize --no-six -wn {} +
It's available since python 2.1 and cmp is considered inferior to use in python2 and totally removed in python3. For Gtk ListStore sort functions that need a cmp interface, implement cmp manually as (a > b) - (a < b) Note: get_addrs must always return a list, since on python3 you cannot compare a Nonetype to a list. An empty list will always compare < a list with something in it, which was the python2 behavior when comparing to None.
8f24d0b to
89a3a1e
Compare
|
I haven't tried running it in python3 yet, but I'm a big believer in baby steps. :) First things first, let's get the codebase into a healthy place where hopefully the only thing that is left to port is pygtk usage itself. I've revised my patch series, by the way, adding some more future imports and fixing a division error that cropped up as a result. I can run or load scans on python2, so it should work as is. Also ff357f9 is just weird and broken right now. |
We explicitly rely on this not being a float. Use modern idiom which works on python2 and python3.
We want to print assuming "unknown". like we do in other parts of the file too. But for some reason, it was assuming \unknown\. Which is strange, and also fails to compile on python3, or python2 with unicode_literals, because \unknown isn't a valid unicode \uXXXX escape. This was broken in commit da0c947 which changed u"assuming \"unknown\"." to u'assuming \unknown\.' with single quotes, but dropped the quotes instead of the escapes.
|
Partially rolled back the unicode_literals change as I found more places where pygtk is oddly finicky about using the native |
absolute_import and print_function are kind of no-brainers division seems safe as there don't seem to be many violations of such, and if there are, it is broken and needs to be fixed. Some errors have been caught by enforcing this on python2 as well. unicode_literals is usually safe, but some things were excluded due to restrictions on needing str() itself. Specifically, pygtk is very picky. Rather than manually casting many strings back to str(), these files are simply excluded for now.
The io module was introduced in python 2.6 and in python 3.x fully replaces the cStringIO/StringIO modules from python 2.x, which are no longer available. It is mostly a drop-in replacement, with the caveat that io.StringIO and io.BytesIO are not interchangeable due to python3 separating the str and bytes types, but the nmap usage is restricted to manipulating files as text, so it should work as-is.
Add additional fixes which rely on the python-six module.
Changes made with the command:
find . -name \*.py -exec python-modernize -wn {} +
manual fixup of gettext.install, which takes a unicode= parameter, but
only on python2.
This patch series implements an incremental move to python3 in the nmap codebase. All changes should continue to work in python2, but improve the compatibility with python3.
Currently, I've mostly upgraded old-style python2 code that does not work on python3, to new-style python2 code that also works on python3; these changes are "the correct way" to do something on both python2 and python3, and thus have value even in an exclusively python2 codebase (but obviously python3 is important. ;))
The last patch adds additional compatibility shims only needed on python3 by using https://pypi.org/project/six/. Once python3 support is complete and python2 is dropped,
six.*imports can be converted to their canonical python3 representations instead.