Hi folks,
First let me say 'Thankyou very much' for Clojure - it has enabled me to
finally take the plunge into learning a Lisp without feeling like I'm
abandoning a ten year investment in the Java platform and its libraries.
I bought 'Practical Common Lisp' about eighteen months ago and read it
half heartedly, never making it to a REPL; however I now have the luxury
of working on a project of my own and decided the time was right to
revisit the promised land of Lisp. I have been aware of Clojure for
about six months, and, given my experience of Java, it seemed like the
obvious place to start. I've spent the past couple of weeks devouring
lots of general Lisp related documentation, Rich's Clojure webcasts and
Stuart's 'Programming Clojure' book. I am pleased to report that my Lisp
epiphany occurred yesterday when I found I could understand this macro
pasted to lisp.org by Rich:
(defmacro defnk [sym args & body]
(let [[ps keys] (split-with (complement keyword?) args)
ks (apply array-map keys)
gkeys (gensym "gkeys__")
letk (fn [[k v]]
(let [kname (symbol (name k))]
`(~kname (~gkeys ~k ~v))))]
`(defn ~sym [...@ps & k#]
(let [~gkeys (apply hash-map k#)
~@(mapcat letk ks)]
~...@body))))
I am now spoiled forever, and although my powers are weak, I realise I
have become one of those smug Lisp types condemned to look down on all
other programming languages for the rest of their lives. Thankyou's and
cliched tales of enlightenment dispensed with, I can now move on to the
plea for aid :)
My current project involves semantic web technologies, at this point
specifically SPARQL queries over RDF triple stores (for those unfamiliar
with SPARQL it will suffice to say that it's a straightforward query
language modelled after SQL, but operating over graphs rather than
tables). Like their SQL counterpart, the Java APIs for performing these
queries are verbose and cumbersome, and I am hoping that they can be
hidden behind some cunning Lisp macros to provide an expressive and
tightly integrated semantic query capability (similar to Microsoft's
LINQ). Unfortunately my ambitions far exceed my skills at this point,
and I am hoping to garner some gentle mentoring to steer me in the right
direction from the outset.
My first inclination is to start with the simplest thing which will
work, which is to create a function which takes a SPARQL query string as
an argument and returns a list of results:
(sparql "
PREFIX foaf: <http://xmlns.com/foaf/0.1>
SELECT ?name
WHERE
{
?person foaf:mbox \"mailto:[email protected]\"; .
?person foaf:name ?name
}")
However this style is poor for several reasons: it looks ugly, quotation
marks have to be escaped manually, and interpolation of variables into
the query string is a chore which further decreases readability. What I
really want is a nice DSL:
(let [mbox "mailto:[email protected]";]
(sparql
(with-prefixes [foaf "http://xmlns.com/foaf/0.1";]
(select ?name
(where
(?person foaf:mbox mbox)
(?person foaf:name ?name)))))
Clearly this is going to require a macro, because for a start I don't
want the symbols representing SPARQL capture variables (the ones
starting with '?') to be evaluated - I want to take the name of the
symbol, '?' and all, and embed it into the query string which this DSL
will ultimately generate before calling into the Java API. On the other
hand, I do want some things evaluated - I want to embed the value
('mailto:[email protected]') bound to the symbol 'mbox' in the
query string, not the name of the symbol.
From my position of total ignorance, I can see two broad approaches to
tackling this. The first is to implement (sparql ...) as a macro which
is responsible for interpreting the entire subtree of forms below it,
building a query string by selectively evaluating some forms whilst
using others as navigational markers which give context. It would honour
the grammar which defines SPARQL queries, and either signal an error or
be guaranteed to generate syntactically correct queries. The macro would
also have insight into the format of the data which would be returned
(gleaned from the 'select' part) and so could return something useful
like a list of maps where the keys are the names of the capture
variables that appear in the select clause. I have no idea how to do
this, but it feels like the 'right' way.
The other approach, which is IMO a bit hacky, but within my reach, is to
define 'with-prefixes', 'select' and 'where' as individual macros whose
first arguments are expanded into the relevant subcomponent of the query
string and whose final argument is a string to be appended to the end.
You then compose these together into the right order to get the compete
query string:
(with-prefix [foaf "http://xmlns.com/foaf/0.1";] "...the select
statement") would evaluate to "PREFIX foaf: <http://xmlns.com/foaf/0.1>
...the select statement"
(select ?name "... the where clause") would evaluate to "SELECT ?name
WHERE ...the where clause"
and so on. In this case 'sparql' would remain a function which takes a
string argument, and you build that query string recursively by nesting
calls to with-prefix, select & where in the correct order. Whist this is
much easier to implement, it has some serious drawbacks - it is up to
the user to compose these things in the correct order (since everything
gets flattened to a string at each level of recursion, we can't check
for errors), and it's no longer possible for the 'sparql' function to
glean the structure of its return value without reparsing the query
string it is passed.
These problems aside, I decided to take a stab at implementing the
'where' macro from the second approach because I have to start somewhere
:) Here is what I have so far:
(defmacro where [& triples]
`(let [encode# (fn [x#] (cond (and (symbol? x#) (= (first (name x#))
\?)) (name x#)
(integer? x#) (str "\"" x# "\"^^xsd:integer")
(float? x#) (str "\"" x# "\"^^xsd:decimal")
(string? x#) (str "\"" x# "\"")))]
(apply str
(interpose " .\n"
(for [triple# '~triples]
(apply str
(interpose " "
(map encode# triple#))))))))
As you can see, so far it correctly encodes SPARQL capture variables,
and literal strings, integers and floats:
user=> (print (where (?a ?b 1) (?a ?b 2.0) (?a ?b "string")))
?a ?b "1"^^xsd:integer .
?a ?b "2.0"^^xsd:decimal .
?a ?b "string"
I tried adding '(list? x#) (eval x#)' to the encode cond to make it cope
with expressions like this:
(where (?a ?b (+ 1 2)))
Unfortunately that results in an unencoded literal '3' in the query
string instead of the '"3"^^xsd:integer' I was looking for. I tried
calling encode recursively (despite the obvious infinite recursion issue
if the eval returns a list) '(list? x#) (encode# eval x#)' but I got an
error at runtime:
user=> (where ((+ 1 2)))
java.lang.Exception: Unable to resolve symbol: encode__2605 in this context
clojure.lang.Compiler$CompilerException: NO_SOURCE_FILE:153: Unable to
resolve symbol: encode__2605 in this context
Clealy this is due to encode# not being bound until after the function
definition completes, but I have no idea how to fix it yet (is there a
way to refer to a function during its definition?). I am also struggling
to get access to variables bound outside the macro:
(let [v "a value"]
(where (?a ?b v))
I tried adding '(symbol? x#) (eval x#)' to the encode cond but that gets
me a complaint about 'v' being unresolvable in this context.
Another problem I face is that there is no enforcement that triples are
passed - the macro just maps all the values through encode and
interposes them with spaces, so no error is raised if you have too many
or too few values to create a valid where clause. I have no idea what is
the proper way of dealing with things like this in a functional language.
So as you can see, if you have made it heroically to the end of this
email, I am keen but facing a steep learning curve :) I am aware that
most of my troubles are well trodden issues that no doubt have thirty
year old Lisp idiomatic solutions, and for bothering you with them on a
Clojure specific mailing list I apologise. On the other hand, if you
feel like sharing some wisdom with me, either directly or through
pointers to resources I would be most grateful. I would also be most
appreciative to receive comments on the design of the DSL itself, and
suggestions for the most Lispish/Clojurish implementation thereof. In
the meantime, it's back to the REPL for me :)
Best Regards,
Adam Harrison
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"Clojure" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
-~----------~----~----~----~------~----~------~--~---
