On Fri, Sep 28, 2012 at 01:01:18AM +0200, Paul Boddie wrote: > On Friday 28 September 2012 00:23:10 Yaroslav Halchenko wrote: > > Thank you Paul ;-) > > > > Good comments -- once again, arguments seems to be oriented mostly > > toward developers... I guess I should explicitly guide the > > abstract more toward 'user-' and "sysadmin-" use cases: people > > in need to have easy and uniform paths for software installation and > > maintenance of the heterogeneous system. In scientific users domain it > > becomes even more fun with heavier reliance on computational and I/O > > libraries (blas/atlas, hdf5, etc) building and maintaining which might > > be quite a bit of a hassle. > > Yes, I had cause to build NumPy from scratch recently, and it was quite > intimidating. It did happen to involve a fairly low-performance device with > fairly severe memory constraints, and the experience really pushed me towards > looking harder at Debian and the packaging work that I knew would have been > done to potentially solve some of the issues I was experiencing, one of which > being modularisation of the library, although I'm not sure how much this is > actually done with NumPy in Debian. > > > Few inline comments. > > > > > I was going to give some feedback more as the kind of person who has gone > > > to Python conferences, and certainly, if you want a native speaker to > > > give feedback on the phrasing of your proposal, I'd be happy to make some > > > suggestions. > > > > I would appreciate "native speaker" feedback! since "accepting all > > types of proposals through September 28, 2012", I guess I have the whole > > tomorrow to revise and submit. I hope to find some time later today to > > revise my abstract and will post it again for further phrasing > > suggestions > > I'll try and follow the list and get back to you. > > > That is true... Somewhat offtopic -- that is why with neuro.debian.net > > we pretty much serve an unofficial backports repository for a good > > portion of Python modules we maintain. Besides immediate benefit for > > users, benefit from backporting for developers has been build-time > > testing across various releases of Debians and Ubuntus, picking out > > problems with specific versions of the core libraries... So, may be I > > should add an accent that availability in Debian doesn't only guarantee > > ease of installation (for users) but provides a good test bed for the > > developers to preclude problems with future deployments on Debian-based > > platforms... ? > > Having gone through the packaging process, I know that a lot of the hurdles > actually do help packagers improve the reliability of the eventual package, > even though the process can be drawn out and frustrating. Even non-technical > stuff like auditing the licences is a valuable activity, however, that gives > users confidence that the end product is of high quality and not just zipped > up and uploaded to some site or other. > > The Python conference scene seems to love testing, so if you can make a case > for Debian and quality assurance, and Debian has done things popular with > this crowd for years like automated builds and the use of very strict package > building tools that won't let you build without a precise specification of > the build dependencies, then that may appeal to some people. ^^ this is a great idea. It'd be nice if we could prototype a flake8 / pyflakes run against the archive, and filter for serious errors > > > > Python packaging has become somewhat insular over the years with > > > Python-centric solutions that work across different systems rather than > > > solutions that work well with the rest of the software on particular > > > systems. However, people appear to like things like virtualenv, > > > especially the Web crowd that makes up a lot of the audience at events > > > like this, because it lets them set up relatively cheap configurations > > > for separate Web applications or for experimenting. > > > > virtualenv is indeed great for the reasons you guys point out AND > > indeed, it is very Python-centric and maintenance of a configured > > virtualenv might become cumbersome for projects with lots of 3rd party > > dependencies and for regular users who would not want to care to switch > > among different virtualenvs etc. > > It's a Python tool for Python developers, primarily. If you're running a Web > operation with lots of Python developers and a commitment to Python then it > may make sense, but that sort of builds a wall that deters people from > adopting Python, too. I've (personally) seen a use-case for it, for applications. (btw, virtualenvwrapper is much nicer then doing all that voodoo by hand) This *is* a bit crazy, but this is the sort of thing virtualenv is good for -- We use the requirements.txt to work out a python application's dependencies. Let's say app1 needs pyfoo1, and app2 needs pyfoo2 (they broke api in pyfoo2) When we install a lib, it gets shuffled into something out of the way, and off the python path. /usr/share/debpy/libfoo/1/*, /usr/share/debpy/libfoo/2/*. When we install an app, it creates a virtualenv in /var/lib somewhere, and symlinks packages /usr/share into the env. This way, you can run app1 and app2 on the same system without conflicts globally (ok, so we're back to what SONAME / symbol versioning can do) Clearly, this idea is *insane*, but it's the sort of solution most hardcore pythonistas I know would go for, which would make us happy(ish), and them happy(ish). btw -- let's not do this. it's insane. > > > I guess I should revise abstract to aim a listener wondering "why should > > I care about Debian if there is virtualenv" WITHOUT explicitly pointing > > to its pros to not cause any flames. And not sure I would be able to > > convince hard-core Python-ians, so I might not even try and orient > > it more toward users/admins. > > I don't think you should worry too much about flames. My impression is that > the packaging people are trying to scale back their ambitions and just get > everyone to do the basic things like write decent metadata, mostly because I > think the process of delivering a comprehensive solution is deadlocked once > again, and I think that people do see the need to hear from distributions and > to try and get as much input from them as possible. I just don't want a good discussion to get deraled by a few people saying globally installing modules results in a bad dependency hell, and we're tainting their whatever. > > > > I have advocated solutions based on fakechrooted debootstrapped > > > installations > > > > btw -- how is it working out for you? i.e. are you still pushing it > > forward? > > Not really, mostly because I was frequently using them for accessing newer > distributions, and then fakechroot wasn't quite capable enough to avoid > low-level library conflicts. So I now routinely use these installations as > genuine chroots instead, and I can convert them to run as User Mode Linux > installations, but I suppose the principle still stands. :-) > > (For my packaging activities, I actually used squeeze under UML to launch a > sid chroot, since my host kernel is not modern enough for sid. I could > probably just launch sid under UML from an installer image, but that's > something for another time.) > > > > if only because you can manage the libraries below the Python modules and > > > extensions as well as the stuff that supports things like distutils and > > > setuptools. However, the people who can change this situation don't see > > > the need or the point: it's either "but I have root!" or "they can always > > > build > > > > many (users on managed boxes) -- don't, so I would have pushed these > > approaches for them as well ;) > > You can certainly make the case to anyone working in a large organisation. > > [...] > > > > The one case that many language-focused groups ignore, and where > > > distributions do well, is the case where a range of different > > > technologies needs to be managed and where administrators just wouldn't > > > be able to keep up with Python eggs, Ruby gems, CPAN, and the > > > language-specific technology of the week. Persuading the Python community > > > to feed packages into Debian so that they become a safer choice for > > > people who routinely use or know other technologies is definitely a > > > worthwhile cause. > > > > indeed safer and more accessible choice. > > I actually have to do this with RHEL at work, but the point stands: if you > can't rely on a stream of packages maintained by someone else, your ability > to deploy and manage a hetergeneous suite of applications is impaired. And > the result of that may involve you deploying a bunch of not so great > applications instead. That latter part is arguably a difficult point to make > to an audience who may be convinced that all the best tools are written in > Python and work perfectly with virtualenv, pip, easy_install or whatever, but > it matters to potential users of Python software, and it may end up mattering > to them if their bosses actually prefer Perl, Ruby or PHP. > > So you have something to really scare people with if you wanted to. ;-) > > Paul > > > -- > To UNSUBSCRIBE, email to debian-python-REQUEST@lists.debian.org > with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org > Archive: [🔎] 201209280101.18621.paul@boddie.org.uk">http://lists.debian.org/[🔎] 201209280101.18621.paul@boddie.org.uk > - Other Paul -- .''`. Paul Tagliamonte <paultag@debian.org> : :' : Proud Debian Developer `. `'` 4096R / 8F04 9AD8 2C92 066C 7352 D28A 7B58 5B30 807C 2A87 `- http://people.debian.org/~paultag
Attachment:
signature.asc
Description: Digital signature