If someone is having problems coding an algorithm, and they provide the code, should we redirect them to the stackoverflow website or some other language specific website?
I agree with Dan, and I'd add that conceptual debugging questions are more likely to be on topic than posting code. Even if the problem isn't with a specific library, if you say that you're using a particular numerical method and describe your algorithm, along with the anomalous results, someone may be able to point you in the right direction by suggesting different methods.
StackOverflow gets a lot of "My code doesn't work!" questions from real beginners to programming, so when you ask about debugging questions my mind latches onto those.
I'd like to suggest that "help me learn to program" questions may not be a good fit for this site even when the ultimate goal is something that does fit. That is, while I think we have agreed to accept raw beginners' questions on scientific computation we should not accept question that expose a current inability to program at all.
Having disposed of the "I don't even know how to program" types of debugging requests this issue becomes clearer: if they're looking for help debugging a scientific computation then it is on topic.
Posting a long section of code and saying "debug this" should probably be off-topic and should either be migrated or closed. However, questions that ask about a specific problem with a numerical library (e.g. "Why is numerical_library_x spitting out error_y in this situation?") feel like they should be on-topic.
I ran into this weird bug with the R standalone math libraries recently. See r-Debian Bug report 657573: mathlib: The following code (using rgamma) hangs. The Debian R maintainer is not interested, as can be seen by the bug report. I also sent this bug report to r-devel, but unsurprisingly, nobody responded. If you want to look at it, read the r-devel posting, it's a better summary.
The issue seems a little specialist for StackOverflow. This is really a strange bug, and is about random number generation so I thought it might be suitable for scicomp. On the other hand it is something that affects a library (the standalone R math library) that is not easily available on most OSs. Aside for Debian/Ubuntu there are few others packaging it. FreeBSD packages it (I'm not sure if the packaging is current), but I don't know of any others. Personally I think bug reports that fall within the perview of scicomp are Ok here.Thoughts?