[Python-Dev] Official citation for Python

[Stephen J. Turnbull]

Chris Barker via Python-Dev writes:

 > But "I wrote some code in Python to produce these statistics" --
 > does that need a citation?

That depends on what you mean by "statistics" and whether (as one
should) one makes the code available.  If the code is published or
"available on request", definitely, Python should be cited.  If not,
and by "statistics" you mean the kind of things provided by Steven
d'Aprano's excellent statistics module (mean, median, standard
deviation, etc), maybe no citation is needed.  But anything more
esoteric than that (even linear regression), yeah, I would say you
should cite both Python and any reference you used to learn the
algorithm or formulas, in the context of mentioning that your
statistics are home-brew, not produced by one of the recognized
applications for doing so.

 > If so, maybe that would take a different form.

Yes, it would.  But not so different: eg, version is analogous to
edition when citing a book.

 > Anyway, hard to make this decision without some idea how the
 > citation is intended to be used.

Same as any other citation, (1) to give credit to those responsible
for providing a resource (this is why publishers and their metadata of
city are still conventionally included), and (2) to show where that
resource can be obtained.  AFAICS, both motivations are universally
applicable in polite society.  NB: Replication is an important reason
for wanting to acquire the resource, but it's not the only one.

I think underlying your comment is the question of *what* resource is
being cited.  I can think of three offhand that might be characterized
as "Python".  First, the PSF, as a provider of funding.  There is a
conventional form for this: a footnote on the title or author's name
saying "The author acknowledges [a] <purpose of grant such as travel>
grant [grant identifier if available] from the Python Software
Foundation."  I usually orally mention them in presentations, too.
That one's easy; *everybody* should *always* do that.

The rest of these, sort of an ideal to strive for.  If you keep a
bibliographic database, and there are now quite a few efforts to crowd
source them, it's easier to go the whole 9 yards than to skimp.  But
except in cases where we don't need to even mention the code, probably
we should be citing, for reasons of courtesy to readers as well as
authors, editors, and publishers (as disgusting as many publishers are
as members of society, they do play a role in providing many resources
---we should find ways to compete them into good behavior, not
ostracize them).

The second is the Python *language and standard library*.  Then the
Language Reference and/or the Library Reference should be cited
briefly when Python is first mentioned, and in the text introducing a
program or program fragment, with a full citation in the bibliography.
I tentatively suggest that the metadata for the Language Reference
would be

    Author: principal author(s) (Guido?) et al. OR python.org OR
        Python Contributors
    Title: The Python Language Reference
    Version: to match Python version used (if relevant, different
        versions each get full citations), probably should not be
        "current"
    Publisher: Python Software Foundation
    Date: of the relevant version
    Location: City of legal address of PSF
    URL: to version used (probably should not be the default)
    Date accessed: if "current" was used

The Library reference would be the same except for Title.

The third is a *particular implementation*.  In that case the metadata
would be

    Author: principal author(s) (Guido) et al. OR python.org OR
        Python Contributors
    Title: The cPython Python distribution
    Python Version: as appropriate (if relevant, different versions each
        get full citations), never "current"
    Distributor Version: if different from Python version (eg, additional
        Debian cruft)
    Publisher: Distributor (eg, PSF, Debian Project, Anaconda Inc.)
    Date: of the relevant version
    Location: City of legal address of distributor

If downloaded:

    URL: to version used (including git commit SHA1 if available)
    Date accessed: download from distributor, not installation date

If received on physical medium: use the "usual" form of citation for a
collection of individual works (even if Python was the only thing on
it).  Probably the only additional information needed would be the
distributor as editor of the collection and the name of the
collection.

In most cases I can think of, if the implementation is cited, the
Language and Library References should be cited, too.

Finally, if Python or components were modified for the project, the
modified version should be preserved in a repository and a VCS
identifier provided.  This does not imply the repository need be
publicly accessible, of course, although it might be for other reasons
(eg, in a GSoC project,wherever or if hosted for free on GitHub).

I doubt that "URNs" like DOI and ISBN are applicable, but if available
they should be included in all cases as well.

Steve