[Haskell-cafe] Anyone up for Google SoC 2010?

[Niklas Broberg]

On Fri, Feb 5, 2010 at 8:55 PM, Edward Kmett <ekmett at gmail.com> wrote:
> You can add me to the list of voices that were unwilling to use it before
> the summer-of-code project due to the random incompatibilities caused by the
> huge supply of extensions it supported out of the box, but who were happy to
> switch to it after the changes were made to make them configurable.

This was indeed the main priority of the project, and the reason why
even I would not have recommended anyone to use haskell-src-exts in
production before the project.

> The rest is just gravy that happens to permit a number of applications such
> as refactoring browsers that were impossible with the previous
> implementation. And, as I recall, the fairly radical exploratory "pretty
> print . parse = id" goal was explicitly listed merely as a secondary goal on
> the original application.

Indeed it was, and I am not aware of any major applications that
actually use the exact-print functionality yet (please, tell me if you
have one!). I do know of several that make very good use of the new
Annotated syntax tree, though, which was introduced as a step towards
exact-printing. The benefits of that, together with the configurable
extensions, is more than enough to now make me happily recommend
haskell-src-exts to anyone working with Haskell source code in any
application. The rest is, as you accurately put it, just gravy.

I must admit I'm a bit sad to have the value of my project questioned
in this way, a project that I myself was more than pleased with, both
with the actual work achieved and the significant positive feedback I
have received after its conclusion. If haskell-src-exts was indeed
popular even before the project, that's all well and good to me. But
it doesn't mean that the library offered to the users then was
satisfactory, nor does it mean that the project failed to deliver
something that those same users needed and/or could make good use of.
Even if the number of direct users did not rise dramatically as a
consequence of the project, why would it not be valid use of a project
slot to greatly improve a library that was already popular? Browsing
the numbers [1] posted by Don Stewart in September last year (the
Haskell Symposium figures, which is the latest I could find) suggests
a substantial increase of downloads of the package both before, during
and after the project, but I can only speculate why. And since the
project concluded late August, figures for September and onwards would
have been more telling.

I'm at a loss as to what criteria is actually used to judge success
here. It seems to me a bit like the eternal discussion between "basic
research" and "applied research". Just because something
(research/library/project) doesn't have an immediate, palpable impact
and/or delivers a visible tool, that certainly doesn't imply that it
doesn't have merit or won't have as profound an impact on the domain,
if more diffuse than a tool (or other palpable deliverable) would.

/Niklas

[1] http://www.galois.com/~dons/hackage/september-2009/total-downloads.html