This question is in the same spirit as the meta question "How appropriate are software package specific questions?" and has come up as a result of the question "How do you install an R executable package?".

Are questions on the installation of a scientific software package or component within the scope of this site? I'm asking to get an idea of the community's opinion; I'll post my own opinion below as an answer.

  • 5
    $\begingroup$ A general comment about scope: as the number of questions/day has trailed off, I've become increasingly of the opinion that we should have a "big tent". We're not in danger of being overwhelmed by volume of questions any time soon. Why not allow anything that may be useful to a significant number of computational scientists? $\endgroup$ – David Ketcheson Jan 17 '12 at 19:22
  • $\begingroup$ I agree with the big tent philosophy, especially as we're getting started. I based my answer off of precedent. I like @MarkBooth's answer a lot; I do think there's a niche for software package-specific installation questions, and even software package-specific usage questions. "How do I include a header" is not appropriate, but I think for certain packages, "What headers should I include?" or "What libraries should I link to?" are reasonable questions. I'm glad that people support those questions. I think they're useful; my initial impression was that people wouldn't support it. $\endgroup$ – Geoff Oxberry Jan 18 '12 at 11:38
  • $\begingroup$ Also, based on the discussion, and the comment I made to Dan's answer, I'm re-opening the question on R packages that I linked to. $\endgroup$ – Geoff Oxberry Jan 18 '12 at 11:39

The criteria on issues like this should always be: Are answers to this question likely to help other people in the future?

If an answer is so localised that it will never be of use to anyone else, then it should be closed as such.

If it is so general that it fits within the remit of Superuser or one of the operating system specific sites (as, I believe, with the R package installation question), then it should be migrated there. If we can't migrate the question for some reason, it should be closed with a suggestion that the other place is more appropriate.

It is the grey area in-between that we should consider appropriate for scicomp.

That might include questions where the documentation for a package used by a Computational Scientist is insufficient to complete installation, or makes assumptions about the knowledge of the installer that is inappropriate.

For instance, I recently went through the effort of installing wxMaxima from source, and the documentation doesn't actually tell you (when) to install Maxima, it just assumed that it is already installed and working in the default location. Similarly when I came to install Maxima from source, I discovered several essential links were broken, which meant that I had to make several leaps of faith to get it to work.

Other questions which might still be appropriate include those to do with configuring the software optimally for use cases outside of the main purpose of the software.

Obviously there are caveats. The Questioner should show what they have done to try to get the software installed. They should explain how that failed and what else they tried beyond the documentation.

Ultimately though, many of the packages we use are complex and often highly configurable. Setting up those packages in an optimal way is often beyond the scope of standard documentation. These sorts of questions thus have the potential to be very useful to the scicomp community.

  • $\begingroup$ If it is appropriate for Superuser, perhaps we should just migrate it there? $\endgroup$ – David Ketcheson Jan 23 '12 at 10:56
  • $\begingroup$ That sounds good to me. $\endgroup$ – David Ketcheson Jan 23 '12 at 19:21

I think that installation and configuration questions should be on-topic; configuration and installation of software is, at least to me, a fundamental part of using it.

I still think the linked question should be closed as off-topic, but not because of language choice. Installing a standalone executable is, in my opinion, so basic and so general that it doesn't belong here.

A PETSc example along the same lines would be "how do I include PETSc headers": it's not really a scientific computing question, it's a question about C syntax. Similarly, this isn't a question about installing R specifically, it's about installing executables in general (at least as far as I can tell, they don't go into much detail).

  • $\begingroup$ So I actually looked up how to install R packages. There's no such thing as an "executable R package", but there is such a thing as a binary R package, and for Windows, it appears that the process is nontrivial. $\endgroup$ – Geoff Oxberry Jan 17 '12 at 10:48
  • $\begingroup$ In that case, I'd say the question is probably on topic. $\endgroup$ – Dan Jan 17 '12 at 19:41

Based on the answer to "How appropriate are software package-specific questions?", I think installation questions are not within the scope of the site, because installation is related to configuration. Configuration was considered off-topic (though still up for debate) in "How appropriate are software package-specific questions?", and there was some grumbling when the question "COMSOL on Ubuntu 10.04 LTS and 11.04" came up.

I'd also be interested to hear what people at Cross Validated have to say, because I'd think that similar questions related to statistical software come up there. (Also, R is statistical software, so their perspective would be particularly relevant.)

  • $\begingroup$ @Downvoter: Care to explain? $\endgroup$ – Geoff Oxberry Jan 17 '12 at 2:08
  • 1
    $\begingroup$ I suspect it is simply someone who believes installation and/or configuration questions should be on-topic. Votes on meta sites often seem to indicate approval/disapproval rather than the requested useful/not useful distinction. But I agree that it would be better if down voters explained their reasoning. $\endgroup$ – Mark Booth Jan 17 '12 at 14:35
  • $\begingroup$ As David Zaslavsky said on that thread: appropriateness of configuration and installation questions is its own question. So let's hash it out here. $\endgroup$ – David Ketcheson Jan 17 '12 at 19:20

In response to a comment over on another board, I think that questions should be focusing on the computational science aspect. It's fine to talk about specific examples that focus on solving a scientific question, but I don't think this site should replace package-specific boards. An example that says

What options do I use to couple code X to code Y.

doesn't add anything to computational science. Making reference to specific examples and problems ("How do I best use X and Y together to do Z?") gives the problem enough of a context that it actually becomes computational science.

  • 1
    $\begingroup$ I think the discussion above is about enabling computational science. There must be some role for such questions here; otherwise, all of the PETSc questions that are "How do I do ______ in PETSc?" would be off-topic, but they were the types of questions used to seed the forum. I don't think this forum is trying to replace package-specific boards. I do think it supplements them, and the broader audience facilitates responses like "You know, it might be a better idea to use code Z instead, because I've used code Y with code X and it doesn't work very well...". Tools are important for us. $\endgroup$ – Geoff Oxberry Jan 23 '12 at 12:32
  • $\begingroup$ I agree with that view. However, how you ask the question is important. You could ask the question above as: "Is there a better way to do Z than using X with Y?" $\endgroup$ – aeismail Jan 24 '12 at 7:52

You know what, I think you're right... but what if the question regarded something like Trillinos or PETSc? Would it then fall under our scope because it is a software specifically embraced by the computational science community? If we embrace one, shouldn't we embrace them all?

  • 2
    $\begingroup$ I think that whatever gets decided has to apply to ALL scientific software packages. I don't see why Trilinos or PETSc should get special treatment when R is also used in the computational science community. $\endgroup$ – Geoff Oxberry Jan 17 '12 at 1:51
  • $\begingroup$ I was responding to the latest post (by GeoffOxberry)... No worries:) Next time, I'll put @GeoffOxberry to be more specific. $\endgroup$ – Paul Jan 17 '12 at 19:45
  • $\begingroup$ @Paul I understood; the trouble is that the order will change as things get voted up/down. $\endgroup$ – David Ketcheson Jan 21 '12 at 11:55
  • $\begingroup$ @DavidKetcheson: Interesting... I didn't realize that. I'll keep that in mind in my future posts. Thanks for the heads up :) $\endgroup$ – Paul Jan 21 '12 at 14:04

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .