brentonashworth / clj-diff Goto Github PK

Unless I am mistaken, the Shortest Edit Script function (ses) is not linear in the length of the input. Instead it uses quadratic space for its (mis-named) fp map, in the worst case.

By using the divide-and-conquer idea from Hirschberg, the original Myers algorithm uses linear space both for computing the edit distance D, and for computing the Shortest Edit Script, and this is also what Wu et al. claim is possible with their improved O(NP)-time algorithm.

Tests to reproduce and a possible fix attached to PR #8

That kind of call (strings of 1 to 3 chars long) fail:

(levenshtein-distance "a" "abc")

with such a stack.

Wrong number of args (0) passed to: core$max
[Thrown class clojure.lang.ArityException]

Restarts:
0: [QUIT] Quit to the SLIME top level

Backtrace:
0: clojure.lang.AFn.throwArity(AFn.java:437)
1: clojure.lang.RestFn.invoke(RestFn.java:399)
2: clojure.lang.AFn.applyToHelper(AFn.java:159)
3: clojure.lang.RestFn.applyTo(RestFn.java:132)
4: clojure.core$apply.invoke(core.clj:600)
5: clj_diff.core$levenshtein_distance.invoke(core.clj:98)
6: clj_diff.core$levenshtein_distance.invoke(core.clj:91)
7: levenshtein1.core$eval1946.invoke(NO_SOURCE_FILE:1)
8: clojure.lang.Compiler.eval(Compiler.java:6465)
9: clojure.lang.Compiler.eval(Compiler.java:6431)
10: clojure.core$eval.invoke(core.clj:2795)
11: swank.commands.basic$eval_region.invoke(basic.clj:47)
12: swank.commands.basic$eval_region.invoke(basic.clj:37)
13: swank.commands.basic$eval768$interactive_eval__769.invoke(basic.clj:66)
14: clojure.lang.Var.invoke(Var.java:401)
15: levenshtein1.core$eval1944.invoke(NO_SOURCE_FILE)
16: clojure.lang.Compiler.eval(Compiler.java:6465)
17: clojure.lang.Compiler.eval(Compiler.java:6431)
18: clojure.core$eval.invoke(core.clj:2795)
19: swank.core$eval_in_emacs_package.invoke(core.clj:92)
20: swank.core$eval_for_emacs.invoke(core.clj:239)
21: clojure.lang.Var.invoke(Var.java:409)
22: clojure.lang.AFn.applyToHelper(AFn.java:167)
23: clojure.lang.Var.applyTo(Var.java:518)
24: clojure.core$apply.invoke(core.clj:600)
25: swank.core$eval_from_control.invoke(core.clj:99)
26: swank.core$spawn_worker_thread$fn__501$fn__502.invoke(core.clj:298)
27: clojure.lang.AFn.applyToHelper(AFn.java:159)
28: clojure.lang.AFn.applyTo(AFn.java:151)
29: clojure.core$apply.invoke(core.clj:600)
30: swank.core$spawn_worker_thread$fn__501.doInvoke(core.clj:294)
31: clojure.lang.RestFn.invoke(RestFn.java:397)
32: clojure.lang.AFn.run(AFn.java:24)
33: java.lang.Thread.run(Thread.java:636)

I'm sorry but I could not dig further into the code to find an explanation.

(diff ...) works well, (edit-distance ...) too, it's just (levenshtein-distance...).

Notice: the error was reproducible on a Windows and Linux installation, thus not depending on a single weird installation...

Hi Brenton,

How do you feel about making a pre-1.0.0 non-snapshot release? Are there any issues that would make this inappropriate for production use as is?

Alex