I’m moving this discussion to stan-dev so everyone can
see what’s going on—hadn’t noticed it’d slipped off.
Andrew and I are running into our standard impedance mismatch where
Andrew thinks he’s giving us a spec (see below) but Daniel and I can’t
figure out what he really wants. I know you’ve mentioned something
like arXiv, but then you’re calling it “facebook” as if you want it to
be more communication/community/friends oriented.
Actually, I think I know what you really want and it’s not
very well specified by what you sent. So here goes with some
initial pushback on clarifying what we need and what some of
the issues are going to be.
issue 1: is it OK to link to a discussion offline? Someone suggested
hacking this up with GitHub submissions linked (how, I don’t know) to discussions
on Discourse. That can handle your submission, but somehow I don’t
think it’s what you want. This is what I mean when I say what you’re
giving us isn’t a spec for a web app.
issue 2: logins. without logins, we can’t ever let
anyone change anything. GitHub solvers this for us, but…
issue 3: if we don’t use someone else’s logins, then we all of
a sudden have security and statefulness issues just like any other
web site.
issue 4: how do we create a web form that lets people submit
all these different bits?
issue 5: is this no holds barred R, python, sed scripts, makefiles, LaTeX,
knitr, Jupyter, whatever? We can’t just run foreign
R code on our servers without isolating them in some kind of sandbox.
If someone submits Jupyter or knitr, do they also submit renderings
or do we do that?
issue 6: do we support some kind of markdown (with MathJax?) for people
to write about things, or is it all embedded in their submitted files?
issue 7: how do we display these models to users? Let’s suppose we get
300 of them from 150 different users. Do we display anything other than a
file directory the way GitHub does? If so, what? Is there some kind
of tagging mechanism? This has come up before with the case studies and
Sophia and crew wanting an education model landing page.
issue 8: if we want to let people update what they submit, do we keep the
older submissions or just the new one? If the old ones, how does that
impact the interface? how do we count versions? again, if it’s on GitHub,
this is all solved for us in a way that’s easy for programmers but not the
rest of our users.
issue 9: database? if so, which one? do any of us know DB programming?
I know Mitzi does. She knows all this back-end web stuff. But she
doesn’t enjoy doing it. Nobody does as far as I know.
issue 10: hosting. we need somewhere to run this. Amazon? A server
in one of our offices doesn’t seem sufficient.
Let’s just say I could keep going. But let’s start with these
and see if we can scope out a minimally useful framework with some
room to grow if there are future directions.
The cost to change our mind is also going to be higher in a web
app than in a paper where someone’s just generating figures statically,
so it’d be nice to get a reasonably fleshed out design up front.
- Bob