Python package naming guidelines

[Timothy Selivanow]

I feel this is more nit-picking, but I'm currently working on packaging
the following items (python/zope names) and need some direction:

zc.buildout
zope.proxy
ZConfig
zdaemon
zope.testing
ZODB3

I haven't exactly liked the current naming conventions, as they leave
way too much variation while still being "acceptable".  E.g., a package
I have already done, pysvn, could have been named "python-pysvn" or
"pysvn".  If it had been used as "import svn", I would have no
problems
calling it "python-svn", as that would denote that it is a python
package having something to do with SVN (since SVN is already a
well-known entity), but upstream calls itself "pysvn" while also using
that as the python package name, so calling it "python-pysvn" seemed
redundant.  However, with the above, there seems to be more ambiguity.

Adding to my naming troubles, there is already a package for
zope.interface, which is a dependency of ZODB3, and it's name is
python-zope-interface.  The problem I am having is that it seems to me
that if an entity is fairly standalone, i.e. there is no ambiguity
whether or not it is a python (or any other language)
library/application, that the "python" prefix would be dropped.  We
don't currently call TurboGears, "python-TurboGears" (yet there are
other turbo-related libs that follow the python prefix scheme, e.g.
python-turbokid and python-turbocheeta).

Really, I'm asking the Python consuming portion of the Fedora Community
what they would rather see.  I can also take this up on the packaging
list, but thought I'd ask here first because potentially it's the people
on this list that would be consuming these packages.

Would it be:

python-zc-buildout
python-zope-proxy
python-ZConfig
python-zdaemon
python-zope-testing
python-ZODB3
python-zope (thrown in to add contrast, and I'm still working on
packaging it...)

Or the above without the prefix?


It seems to me that there should be a more firm guideline that takes
into account python applications, python only frameworks/libs, and
python lib ports (these seem to be the bulk of how a name would be
determined).


--Tim
 _______________________________________________________________________ 
/ Much of the excitement we get out of our work is that we don't really \
| know what we are doing.                                               |
\                 -- E. Dijkstra                                        /
 ----------------------------------------------------------------------- 
  \
   \   \
        \ /\
        ( )
      .( o ).

[Toshio Kuratomi]

Timothy Selivanow wrote:
>  I feel this is more nit-picking, but I'm currently working on
> packaging
>  the following items (python/zope names) and need some direction:
>  
>  zc.buildout
>  zope.proxy
>  ZConfig
>  zdaemon
>  zope.testing
>  ZODB3
>  
>  I haven't exactly liked the current naming conventions, as they leave
>  way too much variation while still being "acceptable".  E.g., a package
>  I have already done, pysvn, could have been named "python-pysvn" or
>  "pysvn".  If it had been used as "import svn", I would have no
> problems
>  calling it "python-svn", as that would denote that it is a python
>  package having something to do with SVN (since SVN is already a
>  well-known entity), but upstream calls itself "pysvn" while also using
>  that as the python package name, so calling it "python-pysvn" seemed
>  redundant.  However, with the above, there seems to be more ambiguity.
>  
>  Adding to my naming troubles, there is already a package for
>  zope.interface, which is a dependency of ZODB3, and it's name is
>  python-zope-interface.  The problem I am having is that it seems to me
>  that if an entity is fairly standalone, i.e. there is no ambiguity
>  whether or not it is a python (or any other language)
>  library/application, that the "python" prefix would be dropped.  We
>  don't currently call TurboGears, "python-TurboGears" (yet there are
>  other turbo-related libs that follow the python prefix scheme, e.g.
>  python-turbokid and python-turbocheeta).
>  
TurboGears was built before the Guideline.

(Django, OTOH, was built in violation of the guidelines and I didn't get 
to the review bug in time to get it changed)

>  Really, I'm asking the Python consuming portion of the Fedora
> Community
>  what they would rather see.  I can also take this up on the packaging
>  list, but thought I'd ask here first because potentially it's the people
>  on this list that would be consuming these packages.
>  
>  Would it be:
>  
>  python-zc-buildout
>  python-zope-proxy
>  python-ZConfig
>  python-zdaemon
>  python-zope-testing
>  python-ZODB3
>  python-zope (thrown in to add contrast, and I'm still working on
>  packaging it...)
>  
>  Or the above without the prefix?
>  
>  
>  It seems to me that there should be a more firm guideline that takes
>  into account python applications, python only frameworks/libs, and
>  python lib ports (these seem to be the bulk of how a name would be
>  determined).
>   
An application doesn't need to be prefixed with python-.

The question is, of course, if something like gourmet which has a 
module... but that module really just exists to be imported by the 
program should be gourmet or python-gourmet.  I wouldn't stop someone 
from going either way with this.

Other than that, bit... python libs and python lib ports should both be 
prefixed with python-.  A python framework should be prefixed with 
python- as well but the definition of framework is a bit ambiguous: zope 
can run standalone and therefore could be considered an application 
while TurboGears cannot run until someone creates an application that 
imports it, therefore it should definitely have been named 
python-turbogears.

So in your list, you need to decide if it's an application, an 
application which happens to have a module to support its functionality, 
or if it's a module that people are expected to use in their programs. 
If the latter, you are most likely going to want a python- prefix.... 
although you might want to have a subpackage for the application in 
corner cases where it's equally likely that someone would import a 
certain module and that someone else would run the program.

-Toshio

[Timothy Selivanow]

On Thu, 2008-04-24 at 16:21 -0700, Toshio Kuratomi wrote:

>  Other than that, bit... python libs and python lib ports should both
> be 
>  prefixed with python-.  A python framework should be prefixed with 
>  python- as well but the definition of framework is a bit ambiguous: zope 
>  can run standalone and therefore could be considered an application 
>  while TurboGears cannot run until someone creates an application that 
>  imports it, therefore it should definitely have been named 
>  python-turbogears. 
So, under that logic, I should have named pysvn python-pysvn?  That
seems a bit redundant to me, but I'm willing to play along if that is
the consensus.  From now on, I'll make an effort to name things that
way, unless I hear a better reason not to (name space pollution,
lib-{package} naming in other distros, other historical reason that I'm
not aware of or privy to...).

If this is The Way, it should be updated on the wiki so other (future)
maintainers can avoid the ambiguity.  Yes, I've volunteered to make the
change (or propose it to Packaging Committee) as long as there is
sufficient consensus, or however the Python SIG is making decisions.
That is unless someone with more tenure would be more applicable in this
case.


--Tim
 _____________________________________________________________________________ 
< "But what we need to know is, do people want nasally-insertable computers?"
>
 ----------------------------------------------------------------------------- 
  \
   \   \
        \ /\
        ( )
      .( o ).

[Toshio Kuratomi]

Timothy Selivanow wrote:
>  On Thu, 2008-04-24 at 16:21 -0700, Toshio Kuratomi wrote:
>  
> > Other than that, bit... python libs and python lib ports should both be 
> > prefixed with python-.  A python framework should be prefixed with 
> > python- as well but the definition of framework is a bit ambiguous: zope 
> > can run standalone and therefore could be considered an application 
> > while TurboGears cannot run until someone creates an application that 
> > imports it, therefore it should definitely have been named 
> > python-turbogears.
>  
>  So, under that logic, I should have named pysvn python-pysvn?  That
>  seems a bit redundant to me, but I'm willing to play along if that is
>  the consensus. 
Yeah... but there is a specific exception in the guidelines for packages 
which start with 'py'.  scop talked about having that exception removed 
(old packages grandfathered) but it never reached the proposal stage.

>  If this is The Way, it should be updated on the wiki so other
> (future)
>  maintainers can avoid the ambiguity.  Yes, I've volunteered to make the
>  change (or propose it to Packaging Committee) as long as there is
>  sufficient consensus, or however the Python SIG is making decisions.
>  That is unless someone with more tenure would be more applicable in this
>  case.
>   
If the wiki isn't clear, definitely draft better wording and propose it 
to the Packaging Committee.

-Toshio

[Kyle VanderBeek]

On Fri, Apr 25, 2008 at 09:02:38AM -0700, Timothy Selivanow wrote:
>  So, under that logic, I should have named pysvn python-pysvn?  That
>  seems a bit redundant to me, but I'm willing to play along if that is
>  the consensus.  From now on, I'll make an effort to name things that
>  way, unless I hear a better reason not to (name space pollution,
>  lib-{package} naming in other distros, other historical reason that I'm
>  not aware of or privy to...). 
I agree that python-pysvn seems redundant, but that is the fault of the
project creators, and is probably not up to packagers to correct. 
Naming a python package with "py" in the name is redundant itself.  But,
since you're going to have to "import pysvn", you might was well be
installing python-pysvn.

-- 
kylev(a)kylev.com
  Some people have a way with words, while others... erm... thingy.

[Timothy Selivanow]

On Fri, 2008-04-25 at 10:25 -0700, Kyle VanderBeek wrote:
>  On Fri, Apr 25, 2008 at 09:02:38AM -0700, Timothy Selivanow wrote:
>  > So, under that logic, I should have named pysvn python-pysvn?  That
>  > seems a bit redundant to me, but I'm willing to play along if that is
>  > the consensus.  From now on, I'll make an effort to name things that
>  > way, unless I hear a better reason not to (name space pollution,
>  > lib-{package} naming in other distros, other historical reason that I'm
>  > not aware of or privy to...).
>  
>  I agree that python-pysvn seems redundant, but that is the fault of the
>  project creators, and is probably not up to packagers to correct. 
>  Naming a python package with "py" in the name is redundant itself.  But,
>  since you're going to have to "import pysvn", you might was well be
>  installing python-pysvn. 
Since I didn't make the deadline to have it included in F9, how
difficult/frowned upon is it to change the package name.  However, it is
in F8 now...


--Tim
 _____________________________________________________________________________ 
< "See - the thing is - I'm an absolutist.  I mean, kind of ... in a way
..." >
 ----------------------------------------------------------------------------- 
  \
   \   \
        \ /\
        ( )
      .( o ).