[Haskell-cafe] Re: [Haskell] Google Summer of Code 2009
gwern0 at gmail.com
Tue Feb 10 23:35:46 EST 2009
(The following is a quasi essay/list of past Summer of Code projects;
my hope is to guide thinking about what Summer of Code projects would
be good to pick, and more specifically what should be avoided.
If you're in a hurry, my conclusions are at the bottom.
The whole thing is written in Markdown; for best results pass it
through Pandoc or view it via your friendly local Gitit wiki.)
# Summer of Code
As part of Google's [Summer of
program, Google sponsors 5-10 [projects for
The Haskell Summer of Codes have often produced excellent results, but
how excellent is excellent? Are there any features or commonalities
between successful projects or unsuccessful ones?
(This questions are particularly important as SoC 2009 isn't too far
away; yet we don't have even a general sense of where we are.)
## Example retrospective: Debian
An energetic blogger & Debian developer has produced
series on the many Debian-related Summer of Code projects.
The results are interesting: some projects were a failure and the
relevant student drifted away and had little to do with Debian again;
and some were great successes. I don't discern any particular lessons
there, except perhaps one against hubris or filling unclear needs.
Let's see whether that holds true of Haskell.
### Haskell retrospective
Haskell wasn't part of the first Summer of Code in 2005, but it was
accepted for 2006. We start there.
The 2006 [homepage](http://hackage.haskell.org/trac/summer-of-code/wiki/SoC2006)
lists the following projects:
- "Fast Mutable Collection Types for Haskell"
Caio Marcelo de Oliveira Filho, mentored by Audrey Tang
This ultimately resulted in the
library ('fast mutable collection' here meaning 'array'; see the
[application](http://darcs.haskell.org/judy/SOC_APP.txt)). HsJudy was
apparently used in Pugs at one time, but no more. Thus, I judge this
to have been **unsuccessful**.
- "Port Haddock to use GHC"
David Waern, mentored by Simon Marlow
**Successful**. Haddock has used the GHC API ever since.[^complaints]
- "A model for client-side scripts with HSP"
Joel Björnson, mentored by Niklas Broberg
Was initially unsuccessful, but seems to've been picked up again.
So I give this a tentative **successful**
- "GHCi based debugger for Haskell"
José Iborra López, mentored by David Himmelstrup
**Successful**. The GHCi debugger was accepted into GHC HEAD, and
by now is in production use,
Jun Mukai, mentored by Shae Erisson
**Unsuccessful**. HaskellNet is dead, and nothing of it has
propagated elsewhere. (I'm not entirely sure what happened with the
HaskellNet code - I know of
but that's about it.) Shae tells me that this poor uptake is probably
due to a lack of advertising, and not any actual defect in the
- "Language.C - a C parser written in Haskell"
Marc van Woerkom, mentored by Manuel Chakravarty
**Failure**. According to [Don Stewart's
of the 2006 SoC, this project was not completed.
- "Implement a better type checker for Yhc"
Leon P Smith, mentored by Malcolm Wallace
**Failure**. See Language.C
- "Thin out cabal-get and integrate in GHC"
Paolo Martini, mentored by Isaac Jones
**Successful**. Code is in Cabal, which we all know and love.
- "Unicode ByteString, Data.Rope, Parsec for generic strings"
Spencer Janssen, mentored by Don Stewart
**Successful**. (Again, per Don.)
4 successful; 2 unsuccessful; and 2 failures.
The [2007 homepage](http://hackage.haskell.org/trac/summer-of-code/wiki/SoC2007)
- "Darcs conflict handling"
Jason Dagit, mentored by David Roundy
**Successful**. The work was successful in almost completely
getting rid of the exponential conflict bug, and has been in the
regular releases of Darcs 2.x for some time now.
- "Automated building of packages and generation of Haddock documentation"
Sascha Böhme, mentored by Ross Paterson
**Successful**. The auto build and doc generation are
long-standing and very useful parts of Hackage.
- "Rewrite the typechecker for YHC and nhc98"
Mathieu Boespflug, mentored by Malcolm Wallace
- "Cabal Configurations"
Thomas Schilling, mentored by Michael Isaac Jones
**Successful**. Cabal configurations have been around for a while
and are very useful for enabling/disabling things
- Update the Hat tracer
Kenn Knowles, mentored by Malcolm Wallace
**Unsuccessful**. The update apparently happened, since the [Hat
homepage](http://www.haskell.org/hat/) says "Version 2.06 released 2nd
Oct 2008", but it is
[described](http://www.haskell.org/hat/download.html) as unmaintained,
and I can't seem to find any examples of people actually using Hat.
- Generalizing Parsec to ParsecT and arbitrary input (ByteStrings)
Paolo Martini, mentored by Philippa Jane Cowderoy
**Success**. But a mixed one – I understand the performance is
terrible so few people use it.
- Shared Libraries for GHC
Clemens Fruhwirth, mentored by Simon Marlow
**Unsuccessful**. The situation is unclear to me, but I know that
for some period dynamic linking worked for some platforms. However,
it's 2009 and I still have static linking; so I'm going to chalk this
one up as existing but not in use.
Mieczysław Bąk, mentored by Bryan O'Sullivan
**Success**. Libcurl does exist and has been used, and is
maintained. I understand it has been used, even if I personally don't
know of any examples.
- "Extending GuiHaskell: An IDE for Haskell Hackers"
Asumu Takikawa, mentored by Neil David Mitchell
**Failure**. GuiHaskell does not exist in any usable form. (The
homepage summarizes the situation thusly: ["**Warning**: This project
is fragile, unfinished, and I do not recommend that anyone tries using
4 successes; 2 unsuccessful; 1 failure, and 1 unknown.
##### See also
- [The Monad.Reader's](http://haskell.org/haskellwiki/The_Monad_Reader)
covers SoC projects
The [2008 homepage](http://hackage.haskell.org/trac/summer-of-code/wiki/SoC2008)
isn't so kind enough as to list all the projects as before, but it
does tell us that only 7 projects were accepted by Google.
So we can work from the
page which lists 6:
- "C99 Parser/Pretty-Printer"
by Benedikt Huber, mentored by Iavor Diatchki
**Success**. The first try failed, but the second won through, and
now people are doing things like [parsing the Linux
- "GMap - Fast composable maps"
by Jamie Brandon
**Successful**. GMap is on
and I believe I've seen it used.
- "Haskell API Search"
**Successful**. The improved performance and search capability
have made it into Hoogle releases.
- "Cabal 'make-like' dependency framework"
**Successful**? (I don't actually know, but this probably made it
- "GHC plugins"
Maximilian Conroy Bolingbroke
- "Data parallel physics engine"
**Unsuccessful**? It seems to be finished but no use made of the
actual engine (that I can see mentioned on the [engine's
- "GHC API".
According to the [Haskell Weekly
> "Thomas Schilling (nominolo) is working on improvements to the GHC API. Officials at HWN headquarters have released a statement reversing their previous position regarding the existence of Thomas, citing regrettably faulty information to explain their previous misapprehensions. Expect to hear more from Thomas soon, now that he has finished graduating and moving."
5 successful, 1 unsuccessful, 1 unknown. (These assessments are the
most untrustworthy of all, since relatively little time has passed;
it's much easier to see whether 2006 projects have had a lasting
effect than these 2008 projects.)
##### See also
- The Monad.Reader's [Issue
Yet to come.. The student apps are due during 23 March – 3 April. Look
forward to it!
#### Haskell roundup
So, what lessons can we learn from the 3 SoCs? It seems to me like
there are roughly 3 groups explanations of unsuccess or failure fall
into. They are:
1) _Hubris_. GuiHaskell is probably a good example; it is essentially
a bare-bones IDE, from its description. It is expecting a bit much of
a single student in a single summer to write *that*!
2) _Unclear use_. HsJudy is my example here. There are already so many
arrays and array types in Haskell! What does HsJudy bring to the table
that justifies a FFI dependency? Who's going to use it? Pugs initially
did apparently, but perhaps that's just because it was there - when I
looked at Pugs/HsJudy in 2007, certainly Pugs had no need of it. (The
data parallel physics engine is probably another good example. Is it
just a benchmark for the GHC developers? Is it intended for actual
games? If the former, why is it a SoC project, and if the latter,
isn't that a little hubristic?)
3) _Lack of propaganda_. One of the reasons Don Stewart's bytestring
library is so great is his relentless evangelizing of their benefits,
which convinces people to actually take the effort to learn and use
them, and eventually by network effects the whole Haskell community is
affected & improved. Some of these SoC projects suffer from a distinct
lack of community buy-in - who used HaskellNet? Who used Hat when it
was updated? Indifference can be fatal, and can defeat the point of a
project. What good is a library that no one uses? These aren't
academic research projects which accomplish their task just by
existing, after all. They're supposed to be useful to real Haskellers.
[^complaints]: I can hear the wankers in the peanut gallery - "Yeah,
and it's been buggy ever since!" Hush you.
More information about the Haskell-Cafe