Python style standards in developer docs
As we start to get python code in Gromacs, it might be good to document style standards. I would propose a modified version of the Google python style guides (I may be biased because I used them at Google for a couple years and continue to use them in my own group, but I think they give good readable code). There's also a pylint template we can use.
Erik suggested that we keep naming conventions consistent with the C/C++ code. I'm fine with that (we might want to revisit if we get a substantial amount of python code, but especially when it's relatively small in quantity it makes sense to prioritize codebase consistency over language idiom).
If this sounds reasonable, I'm happy to start working on a patch. If people want to discuss more here beforehand, that's good too.
#1 Updated by Teemu Murtola over 4 years ago
I won't mind if we have such standards, and the developer guide is the best place for this. Should this then apply also to all the small (and not-so-small) Python scripts we have (e.g., in releng)?
The most important thing to standardize, though, would be the expectations on what Python versions we want to support. If both 2 and 3, how do we deal with that, and how if someone wants to use features from newer Python for some limited-audience script?
#2 Updated by Peter Kasson over 4 years ago
I think it should apply to all Python scripts going forward, and as with all legacy code we try to do clean-up as feasible.
2.7 versus 3.4 is an interesting question. I'd personally favor 2.7, but there's certainly an argument for 3.4 as the more forward-looking choice. There's also the reverse argument that well-written 2.7 code that uses modern Python idioms will have a trivial migration path to 3.4.
As with all our code, commits to master shouldn't introduce new external dependencies (such as newer features) unless specifically permitted in our dev standards, right?
#3 Updated by Teemu Murtola over 4 years ago
Perhaps trivial, but not a no-op migration, since well-written code specifically targeting 2.7 likely uses methods such as
itervalues() that no longer exist in 3.x. So unless explicitly managed, we are most likely stuck with just either 2.7 or 3.x until we completely drop the support for 2.7.
Also, there can be, but not necessarily should be, a difference between user-facing code (possibly even distinguishing an API and other scripts) and code only (or mainly) used by developers.
All current scripts have been developed using 2.7, but a gmxtest.pl replacement that I have toyed around a bit with would benefit from subtest support in the Python
unittest library, which is in 3.4 only. If we standardize on 2.7, is the only solution for such cases then to reimplement that standard library functionality?
#4 Updated by Peter Kasson over 4 years ago
Hmm--so a few possibilities:
1. We choose python2.7 for everything and use a replacement for unittest subtests
2. We choose python3.4 for everything
3. All API's and user-facing code should be python 2.7-compatible (we could specify cross-compatible or just leave it as 2.7), and we allow engineering code to be 2.7 or 3.4.
4. All code could be either 2.7 or 3.4
If we go with #'s 3 or 4, we should specify version. A lot of code starts to deprecate the shebang line, but we could A) require /usr/bin/python2 vs. /usr/bin/python3 in the shebang line or B) require use of the six library (which provides versioning support) either in all code or in a specified subset (code that uses python3-specific; code that uses python2-specific, ...)