From 745c3f552e7cbd13e07cacdd7577468cc3349a6f Mon Sep 17 00:00:00 2001
From: Michael Davidsaver <mdavidsaver@gmail.com>
Date: Fri, 5 Feb 2021 07:05:22 -0800
Subject: [PATCH] Com: add epicsStrSimilarity()

---
 modules/libcom/src/misc/epicsString.c | 70 +++++++++++++++++++++++++++
 modules/libcom/src/misc/epicsString.h | 10 ++++
 modules/libcom/test/epicsStringTest.c | 34 ++++++++++++-
 3 files changed, 113 insertions(+), 1 deletion(-)

diff --git a/modules/libcom/src/misc/epicsString.c b/modules/libcom/src/misc/epicsString.c
index 766af20b6..e18d9214a 100644
--- a/modules/libcom/src/misc/epicsString.c
+++ b/modules/libcom/src/misc/epicsString.c
@@ -23,9 +23,11 @@
 #include <errno.h>
 #include <ctype.h>
 
+#include "epicsAssert.h"
 #include "epicsStdio.h"
 #include "cantProceed.h"
 #include "epicsString.h"
+#include "epicsMath.h"
 
 /* Deprecated, use epicsStrnRawFromEscaped() instead */
 int dbTranslateEscape(char *dst, const char *src)
@@ -358,3 +360,71 @@ unsigned int epicsMemHash(const char *str, size_t length, unsigned int seed)
     }
     return hash;
 }
+
+/* Compute normalized Levenshtein distance
+ *
+ * https://en.wikipedia.org/wiki/Levenshtein_distance
+ *
+ * We modify this to give half weight to case insensitive substitution.
+ * All normal integer weights are multiplied by two, with case
+ * insensitive added in as one.
+ */
+double epicsStrSimilarity(const char *A, const char *B)
+{
+    double ret = 0;
+    size_t lA, lB, a, b;
+    size_t norm;
+    size_t *dist0, *dist1, *stemp;
+
+    lA = strlen(A);
+    lB = strlen(B);
+
+    /* max number of edits to change A into B is max(lA, lB) */
+    norm = lA > lB ? lA : lB;
+    /* take into account our weighting */
+    norm *= 2u;
+
+    dist0 = calloc(1+lB, sizeof(*dist0));
+    dist1 = calloc(1+lB, sizeof(*dist1));
+    if(!dist0 || !dist1) {
+        ret = -1.0;
+        goto done;
+    }
+
+    for(b=0; b<1+lB; b++)
+        dist0[b] = 2*b;
+
+    for(a=0; a<lA; a++) {
+        dist1[0] = 2*(a+1);
+
+        for(b=0; b<lB; b++) {
+            size_t delcost = dist0[b+1] + 2,
+                   inscost = dist1[b] + 2,
+                   subcost = dist0[b],
+                   mincost = delcost;
+            char ca = A[a], cb = B[b];
+
+            if(ca!=cb)
+                subcost++;
+            if(toupper((int)ca)!=toupper((int)cb))
+                subcost++;
+
+            if(mincost > inscost)
+                mincost = inscost;
+            if(mincost > subcost)
+                mincost = subcost;
+
+            dist1[b+1] = mincost;
+        }
+
+        stemp = dist0;
+        dist0 = dist1;
+        dist1 = stemp;
+    }
+
+    ret = norm ? (norm - dist0[lB]) / (double)norm : 1.0;
+done:
+    free(dist0);
+    free(dist1);
+    return ret;
+}
diff --git a/modules/libcom/src/misc/epicsString.h b/modules/libcom/src/misc/epicsString.h
index 2ef675bc9..3c08e5277 100644
--- a/modules/libcom/src/misc/epicsString.h
+++ b/modules/libcom/src/misc/epicsString.h
@@ -41,6 +41,16 @@ LIBCOM_API char * epicsStrtok_r(char *s, const char *delim, char **lasts);
 LIBCOM_API unsigned int epicsStrHash(const char *str, unsigned int seed);
 LIBCOM_API unsigned int epicsMemHash(const char *str, size_t length,
                                          unsigned int seed);
+/** Compare two strings and return a number in the range [0.0, 1.0] or -1.0 on error.
+ *
+ * Computes a normalized edit distance representing the similarity between two strings.
+ *
+ * @returns 1.0 when A and B are identical, down to 0.0 when A and B are unrelated,
+ *          or < 0.0 on error.
+ *
+ * @since UNRELEASED
+ */
+LIBCOM_API double epicsStrSimilarity(const char *A, const char *B);
 
 /* dbTranslateEscape is deprecated, use epicsStrnRawFromEscaped instead */
 LIBCOM_API int dbTranslateEscape(char *s, const char *ct);
diff --git a/modules/libcom/test/epicsStringTest.c b/modules/libcom/test/epicsStringTest.c
index 054f0c043..e4e95c998 100644
--- a/modules/libcom/test/epicsStringTest.c
+++ b/modules/libcom/test/epicsStringTest.c
@@ -17,6 +17,7 @@
 
 #include "epicsUnitTest.h"
 #include "epicsString.h"
+#include "epicsMath.h"
 #include "testMain.h"
 
 static
@@ -74,6 +75,35 @@ void testGlob(void) {
     testOk1(epicsStrGlobMatch("hello","*"));
 }
 
+static
+void testDistance(void) {
+    double dist;
+    testDiag("testDistance()");
+
+#define TEST(EXPECT, A, B) dist = epicsStrSimilarity(A, B); testOk(fabs(dist-(EXPECT))<0.01, "distance \"%s\", \"%s\" %f ~= %f", A, B, dist, EXPECT)
+
+    TEST(1.00, "", "");
+    TEST(1.00, "A", "A");
+    TEST(0.00, "A", "B");
+    TEST(1.00, "exact", "exact");
+    TEST(0.90, "10 second", "10 seconds");
+    TEST(0.71, "Passive", "Pensive");
+    TEST(0.11, "10 second", "Pensive");
+    TEST(0.97, "Set output to IVOV", "Set output To IVOV");
+
+    /* we modify Levenshtein to give half weight to case insensitive matches */
+
+    /* totally unrelated except for 'i' ~= 'I' */
+    TEST(0.06, "Passive", "I/O Intr");
+    TEST(0.06, "I/O Intr", "Pensive");
+    /* 2x subst and 1x case subst, max distance 2xlen("YES") */
+    TEST(0.50, "YES", "yes");
+    TEST(0.00, "YES", "NO");
+    TEST(0.67, "YES", "Yes");
+    TEST(0.67, "Tes", "yes");
+#undef TEST
+}
+
 MAIN(epicsStringTest)
 {
     const char * const empty = "";
@@ -88,7 +118,7 @@ MAIN(epicsStringTest)
     char *s;
     int status;
 
-    testPlan(387);
+    testPlan(401);
 
     testChars();
 
@@ -284,5 +314,7 @@ MAIN(epicsStringTest)
     testOk(result[1] == 'g', "  Terminator char got '%c'", result[1]);
     testOk(result[status] == 0, "  0-terminated");
 
+    testDistance();
+
     return testDone();
 }