From 6b0bd4d97b3ffd97e7c21ccd69369750d5d21c1e Mon Sep 17 00:00:00 2001
From: Tim Edwards <tim@opencircuitdesign.com>
Date: Tue, 6 Feb 2024 15:02:50 -0500
Subject: [PATCH] Found an error with the property sorting in which float
 values were not compared for sorting in the same way they are compared for
 property matching.  The "slop" value was treated as absolute, not a
 percentage, so for example a slop of 0.01 on a dimension of microns would
 cause all dimensions to be treated as round-off error, and no sorting would
 occur.

---
 VERSION       |  2 +-
 base/netcmp.c | 34 +++++++++++++++++++++++++++++++---
 2 files changed, 32 insertions(+), 4 deletions(-)

diff --git a/VERSION b/VERSION
index 54f6bb2..1666b1d 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.5.267
+1.5.268
diff --git a/base/netcmp.c b/base/netcmp.c
index eee161a..4bb09f6 100644
--- a/base/netcmp.c
+++ b/base/netcmp.c
@@ -4204,6 +4204,28 @@ static int compsort(const void *p1, const void *p2)
        return (s1->value > s2->value) ? 1 : 0;
 }
 
+/*--------------------------------------------------------------*/
+/* compdsort() is like compsort() but uses the definition for	*/
+/* "slop" used for type double instead of type integer.		*/
+/*--------------------------------------------------------------*/
+
+static int compdsort(const void *p1, const void *p2)
+{
+    propsort *s1, *s2;
+    double smax, ddiff;
+
+    s1 = (propsort *)p1;
+    s2 = (propsort *)p2;
+
+    smax = fmax(s1->slop, s2->slop);
+    ddiff = fabs(s1->value - s2->value) /
+		fabs(s1->value + s2->value);
+    if (ddiff <= smax)
+       return (s1->avalue > s2->avalue) ? 1 : 0;
+    else
+       return (s1->value > s2->value) ? 1 : 0;
+}
+
 /*--------------------------------------------------------------*/
 /* Sort properties of ob1 starting at property idx1 up to	*/
 /* property (idx1 + run).  Use series critical property for	*/
@@ -4221,7 +4243,8 @@ void series_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
    struct valuelist *vl, *sl;
    int i, p, sval, merge_type;
    // double cval, slop;
-   int has_crit;
+   int (*sortfunc)() = compsort;
+   int has_crit, is_float;
    char ca, co;
    double tval, tslop;
    double aval, pval, oval, aslop, pslop;
@@ -4241,6 +4264,7 @@ void series_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
    for (i = 0; i < run; i++) {
       has_crit = FALSE;
       merge_type = MERGE_NONE;
+      is_float = FALSE;
       ca = co = (char)0;
 
       for (p = 0;; p++) {
@@ -4273,6 +4297,8 @@ void series_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
 	 else {
 	    tval = vl->value.dval;
 	    tslop = kl->slop.dval;
+	    sortfunc = compdsort;
+	    is_float = TRUE;
 	 }
 
 	 if (kl->merge & MERGE_S_CRIT) {
@@ -4327,7 +4353,7 @@ void series_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
 
    obn = obp;	/* Link from last property */
 
-   qsort(&proplist[0], run, sizeof(propsort), compsort);
+   qsort(&proplist[0], run, sizeof(propsort), sortfunc);
 
    // Re-sort list
    obp = ob1;
@@ -4417,6 +4443,7 @@ void parallel_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
    propsort *proplist;
    struct property *kl;
    struct valuelist *vl;
+   int (*sortfunc)() = compsort;
    int i, p, mval, merge_type;
    int has_crit;
    char ca, co;
@@ -4477,6 +4504,7 @@ void parallel_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
 	 else {
 	    tval = vl->value.dval;
 	    tslop = kl->slop.dval;
+	    sortfunc = compdsort;
 	 }
 
 	 if (kl->merge & MERGE_P_CRIT) {
@@ -4531,7 +4559,7 @@ void parallel_sort(struct objlist *ob1, struct nlist *tp1, int idx1, int run)
 
    obn = obp;	/* Link from last property */
 
-   qsort(&proplist[0], run, sizeof(propsort), compsort);
+   qsort(&proplist[0], run, sizeof(propsort), sortfunc);
 
    // Re-sort list
    obp = ob1;