Well, one of my most important questions has been indirectly answered. It seems like Map is still the main point of interest, and Jamie Brandon's <a href="http://www.haskell.org/pipermail/haskell-cafe/2009-March/058270.html">list of remaining objectives</a> include modifying the API to work with the edison or collections API, and benchmarking. I haven't attempted a project this large before, so I'm not sure what is feasible in the allotted time. A complete Data Structures overhaul seems out of the question, but polishing up a single one seems far too sparse. Then again, the benchmarking could represent a significant chunk of work.<div>
<br></div><div>-Nathan<br><br><div><br></div><div><br><div class="gmail_quote">On Wed, Mar 31, 2010 at 11:28 AM, wren ng thornton <span dir="ltr"><<a href="mailto:wren@freegeek.org" target="_blank">wren@freegeek.org</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left:1px solid rgb(204, 204, 204);margin:0pt 0pt 0pt 0.8ex;padding-left:1ex">
Nathan Hunter wrote:<br>
<blockquote class="gmail_quote" style="border-left:1px solid rgb(204, 204, 204);margin:0pt 0pt 0pt 0.8ex;padding-left:1ex"><div>
Hello.<br>
<br>
I am hoping to take on the Data Structures project proposed two years ago by<br></div>
Don Stewart here<<a href="http://hackage.haskell.org/trac/summer-of-code/ticket/1549" target="_blank">http://hackage.haskell.org/trac/summer-of-code/ticket/1549</a>>,<div><br>
this summer.<br>
Before I write up my proposal to Google, I wanted to gauge the reaction of<br>
the Haskell community to this project.<br>
Particularly:<br>
<br>
-What Data Structures in the current libraries are in most dire need of<br>
improvement?<br>
-How necessary do you think a Containers Library revision is?<br>
</div></blockquote>
<br>
One thing I've seen come up repeatedly is the issue of presenting a unified and general interface for Data.Map, Data.IntMap, and related things like Data.Set, bytestring-trie, pqueue, etc which intended to mimic their interface. That alone isn't big enough for a GSoC, but it would be a very nice thing to have. Every few months there's a request on the libraries@ list to alter, generalize, or reunify the map interface in some way.<br>
<br>
** Just to be clear, I do not mean coming up with a typeclass nor doing things like the generalized-trie tricks, I just mean a good old fashioned standard API. **<br>
<br>
There are countervailing forces for making a good API. On the one hand we want functions to do whatever we need, on the other hand we want the API to be small enough to be usable/memorable. In the bytestring-trie library I attempted to resolve this conflict by offering a small set of highly efficient ueber-combinators in the internals module, a medium sized set of functions for standard use in the main module, and then pushed most everything else off into a convenience module.<br>
<br>
The containers library would do good to follow this sort of design. The Data.Map and Data.IntMap structures don't provide the necessary ueber-combinators, which has led to the proliferation of convenience functions which are more general than the standard use functions but not general enough to make the interface complete. Also, these generalized functions are implemented via code duplication rather than having a single implementation, which has been known to lead to cut&paste bugs and maintenance issues. Provided the correct ueber-combinators are chosen, there is no performance benefit for this code duplication either (so far as I've discovered with bytestring-trie).<br>
<br>
Additionally, it'd be nice if some of the guts were made available for public use (e.g., the bit-twiddling tricks of Data.IntMap so I don't have to duplicate them in bytestring-trie).<br>
<br>
Also it would be nice to develop a cohesive test and benchmarking suite, which would certainly be a large enough task for GSoC, though perhaps not fundable.<br>
<br>
I would be willing to co-mentor API and algorithm design for cleaning the cobwebs out of the containers library. I wouldn't have the time for mentoring the choice of datastructures or benchmarking however.<br>
<br>
-- <br>
Live well,<br><font color="#888888">
~wren<br>
</font></blockquote></div><br>
</div></div>