writing graphs with do-notation
Hi! Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example: example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b] In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time. The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar). By the way, it is possible to define cyclic graphs using mdo (RecursiveDo). I haven't come across something similar, so i thought, i'd share it. What do you think? Sönke
Neat Surely there is somewhere in the haskell Twiki that something like this should live? Neil On 12 Dec 2009, at 21:00, Soenke Hahn wrote:
Hi!
Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example:
example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b]
In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time.
The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar).
By the way, it is possible to define cyclic graphs using mdo (RecursiveDo).
I haven't come across something similar, so i thought, i'd share it. What do you think?
Sönke
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Hi! This technique has been used to define netlists in hardware description languages. The original Lava [1] used a monad, but later switched to using observable sharing [2]. Wired [3] uses a monad similar to yours (but more complicated). I think it would be nice to have a single library for defining such graphs (or maybe there is one already?). The graph structure in Wired could probably be divided into a purely structural part and a hardware-specific part. [1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.46.5221 [2] http://www.cs.chalmers.se/~dave/papers/observable-sharing.pdf [3] http://hackage.haskell.org/package/Wired / Emil Soenke Hahn skrev:
Hi!
Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example:
example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b]
In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time.
The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar).
By the way, it is possible to define cyclic graphs using mdo (RecursiveDo).
I haven't come across something similar, so i thought, i'd share it. What do you think?
Sönke
Andy Gill wrote a very nice recent paper on this topic which can serve as the basis for a generic implementation: http://www.ittc.ku.edu/~andygill/paper.php?label=DSLExtract09 As long as you do your "reification" in the IO monad, Andy's library gives you the graph conversion for (almost-) free. -Levent. On Dec 13, 2009, at 10:48 PM, Emil Axelsson wrote:
Hi!
This technique has been used to define netlists in hardware description languages. The original Lava [1] used a monad, but later switched to using observable sharing [2]. Wired [3] uses a monad similar to yours (but more complicated).
I think it would be nice to have a single library for defining such graphs (or maybe there is one already?). The graph structure in Wired could probably be divided into a purely structural part and a hardware-specific part.
[1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.46.5221 [2] http://www.cs.chalmers.se/~dave/papers/observable-sharing.pdf [3] http://hackage.haskell.org/package/Wired
/ Emil
Soenke Hahn skrev:
Hi! Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example: example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b] In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time. The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar). By the way, it is possible to define cyclic graphs using mdo (RecursiveDo). I haven't come across something similar, so i thought, i'd share it. What do you think? Sönke
Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Yes, that's probably close to what I want. It would of course be nice to also have a monadic/applicative interface for building the graphs. In libraries like Wired where you're in a monad anyway, this would get rid of the need for IO. Koen Claessen has made a sketch of a generic graph library that we were planning to use as a basis for the EDSLs at Chalmers. But as far as I remember it looked a lot like the graph in data-reify, so maybe we should just use that instead. / Emil Levent Erkok skrev:
Andy Gill wrote a very nice recent paper on this topic which can serve as the basis for a generic implementation:
http://www.ittc.ku.edu/~andygill/paper.php?label=DSLExtract09
As long as you do your "reification" in the IO monad, Andy's library gives you the graph conversion for (almost-) free.
-Levent.
On Dec 13, 2009, at 10:48 PM, Emil Axelsson wrote:
Hi!
This technique has been used to define netlists in hardware description languages. The original Lava [1] used a monad, but later switched to using observable sharing [2]. Wired [3] uses a monad similar to yours (but more complicated).
I think it would be nice to have a single library for defining such graphs (or maybe there is one already?). The graph structure in Wired could probably be divided into a purely structural part and a hardware-specific part.
[1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.46.5221 [2] http://www.cs.chalmers.se/~dave/papers/observable-sharing.pdf [3] http://hackage.haskell.org/package/Wired
/ Emil
Soenke Hahn skrev:
Hi! Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example: example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b] In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time. The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar). By the way, it is possible to define cyclic graphs using mdo (RecursiveDo). I haven't come across something similar, so i thought, i'd share it. What do you think? Sönke
Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Emil Axelsson
Yes, that's probably close to what I want. It would of course be nice to also have a monadic/applicative interface for building the graphs. In libraries like Wired where you're in a monad anyway, this would get rid of the need for IO.
IIRC, dotgen does this for Dot graphs for Graphviz.
Koen Claessen has made a sketch of a generic graph library that we were planning to use as a basis for the EDSLs at Chalmers. But as far as I remember it looked a lot like the graph in data-reify, so maybe we should just use that instead.
Cale and I started on our own generic graph class... we should probably get back to working on it...
/ Emil
Levent Erkok skrev:
Andy Gill wrote a very nice recent paper on this topic which can serve as the basis for a generic implementation:
http://www.ittc.ku.edu/~andygill/paper.php?label=DSLExtract09
As long as you do your "reification" in the IO monad, Andy's library gives you the graph conversion for (almost-) free.
-Levent.
On Dec 13, 2009, at 10:48 PM, Emil Axelsson wrote:
Hi!
This technique has been used to define netlists in hardware description languages. The original Lava [1] used a monad, but later switched to using observable sharing [2]. Wired [3] uses a monad similar to yours (but more complicated).
I think it would be nice to have a single library for defining such graphs (or maybe there is one already?). The graph structure in Wired could probably be divided into a purely structural part and a hardware-specific part.
[1] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.46.5221 [2] http://www.cs.chalmers.se/~dave/papers/observable-sharing.pdf [3] http://hackage.haskell.org/package/Wired
/ Emil
Soenke Hahn skrev:
Hi! Some time ago, i needed to write down graphs in Haskell. I wanted to be able to write them down without to much noise, to make them easily maintainable. I came up with a way to define graphs using monads and the do notation. I thought this might be interesting to someone, so i wrote a small script to illustrate the idea. Here's an example: example :: Graph String example = buildGraph $ do a <- mkNode "A" [] b <- mkNode "B" [a] mkNode "C" [a, b] In this graph there are three nodes identified by ["A", "B", "C"] and three edges ([("A", "B"), ("A", "C"), ("B", "C")]). Think of the variables a and b as outputs of the nodes "A" and "B". Note that each node identifier needs to be mentioned only once. Also the definition of edges (references to other nodes via the outputs) can be checked at compile time. The attachment is a little script that defines a Graph-type (nothing elaborate), the "buildGraph" function and an example graph that is a little more complex than the above. The main function of the script prints the example graph to stdout to be read by dot (or similar). By the way, it is possible to define cyclic graphs using mdo (RecursiveDo). I haven't come across something similar, so i thought, i'd share it. What do you think? Sönke
Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
-- Ivan Lazar Miljenovic Ivan.Miljenovic@gmail.com IvanMiljenovic.wordpress.com
participants (5)
-
Emil Axelsson -
Ivan Lazar Miljenovic -
Levent Erkok -
Neil Davies -
Soenke Hahn