From 6298d7d40c8a9891527bcb2084fef1a0e0a90cc1 Mon Sep 17 00:00:00 2001
From: Markus Scherer <markus.icu@gmail.com>
Date: Fri, 27 Feb 2026 16:31:34 -0800
Subject: [PATCH] 11. unidata-18.0.0d11.txt

A very minor update to the input file, changing the synthetic
decomp for 1DF1F from <0064, 0064, F8F0> to <0064, 00F0>.
That is the d-eth digraph, and the shorter decomposition is
more correct, but was precluded originally by a bug in how
the sifter tested for infinite recursion in decompositions.
I have now corrected that test for infinite recursion, so
the sifter now allows decompositions of the sort:

x -> a, b
b -> a, c

as long as "a" is itself atomic and does not decompose further.

Corresponding change to the sifter code is also posted:

unisift.c

No generated output files are posted for this delta, because both
allkeys.txt and decomps.txt produce the same output as for
delta 10, except for the generation date.
---
 c/uca/sifter/unidata.txt |  6 +++---
 c/uca/sifter/unisift.c   | 38 +++++++++++++++++++++++++++++++++-----
 2 files changed, 36 insertions(+), 8 deletions(-)

diff --git a/c/uca/sifter/unidata.txt b/c/uca/sifter/unidata.txt
index 415607d60..ff08f2c3a 100644
--- a/c/uca/sifter/unidata.txt
+++ b/c/uca/sifter/unidata.txt
@@ -1,5 +1,5 @@
 # unidata-18.0.0.txt
-# Date: 2026-02-21, 00:00:00 GMT [KW]
+# Date: 2026-02-26, 00:00:00 GMT [KW]
 # © 2026 Unicode®, Inc.
 # Unicode and the Unicode Logo are registered trademarks of Unicode, Inc. in the U.S. and other countries.
 # For terms of use and license, see https://www.unicode.org/copyright.html
@@ -10,7 +10,7 @@
 # Default Unicode Collation Element Table (DUCET) for
 # the Unicode Collation Algorithm.
 #
-# Version 18.0.0 draft 10 (Unicode Version: 18.0.0)
+# Version 18.0.0 draft 11 (Unicode Version: 18.0.0)
 # based on Unicode data file UnicodeData-18.0.0d8.txt
 # Ordering for Unicode 18.0
 #
@@ -14962,7 +14962,7 @@ AB66;LATIN SMALL LETTER DZ DIGRAPH WITH RETROFLEX HOOK;Ll;<sort> 0064 0290;;;;;
 10788;MODIFIER LETTER SMALL DZ DIGRAPH WITH RETROFLEX HOOK;Lm;<super> AB66;;;;;
 0238;LATIN SMALL LETTER DB DIGRAPH;Ll;<sort> 0064 0062;;;;;
 1DFDE;MODIFIER LETTER SMALL DB DIGRAPH;Lm;<super> 0238;;;;;
-1DF1F;LATIN SMALL LETTER D-ETH DIGRAPH;Ll;<sort> 0064 0064 F8F0;;;;;
+1DF1F;LATIN SMALL LETTER D-ETH DIGRAPH;Ll;<sort> 0064 00F0;;;;;
 1DF20;LATIN SMALL LETTER D-LEZH DIGRAPH;Ll;<sort> 0064 026E;;;;;
 1DF21;LATIN SMALL LETTER D-LEZH DIGRAPH WITH RETROFLEX HOOK;Ll;<sort> 0064 1DF05;;;;;
 
diff --git a/c/uca/sifter/unisift.c b/c/uca/sifter/unisift.c
index 4cf00058e..9eeca017a 100644
--- a/c/uca/sifter/unisift.c
+++ b/c/uca/sifter/unisift.c
@@ -64,6 +64,7 @@
  *   2026-Feb-19 Add 208F to isAlphabeticException.
  *               Add 1B168 to kana mapping in unisift_GetKatakanaBase.
  *   2026-Feb-21 Add implicit weights for Jurchen and Seal.
+ *   2026-Feb-26 Tweak test for infinite recursion in doRecursiveDecomp.
  */
 
 /*
@@ -188,7 +189,7 @@
 #define PATHNAMELEN (256)
 #define LONGESTARG  (256)
 
-static char versionString[] = "Sifter version 18.0.0d2, 2026-02-21\n";
+static char versionString[] = "Sifter version 18.0.0d3, 2026-02-26\n";
 
 static char unidatafilename[] = "unidata-18.0.0.txt";
 static char allkeysfilename[] = "allkeys-18.0.0.txt";
@@ -1026,12 +1027,39 @@ UInt32 buf3[60]; /* temporary hold for constructed decomp */
                     printf ( "Bad Value: %s\n", sp );
                     break;
                 }
+                /*
+                 * Check if the token is the same as the scalarValue
+                 * passed in. If this token itself has a non-trivial
+                 * decomposition, this can lead to infinite recursion.
+                 * Test that case and bail out if found.
+                 *
+                 * If cc is Atomic (or Implicit), let this just fall through.
+                 * That allows for cases like:
+                 *   x -> a b
+                 *   b -> a c ==> x -> a a c
+                 *
+                 * This condition is not encountered for the normative
+                 * decompositions, but might be encountered in synthetic
+                 * decompositions added to unidata.txt for collation if
+                 * not properly constructed.
+                 */
                 if ( cc == scalarValue )
                 {
-                    badValues++;
-                    printf ( "Infinite recursion: %08X %s\n", scalarValue, 
-                             sp );
-                    return ( 0 );
+                WALNUTPTR t1;
+
+                    t1 = getSiftDataPtr ( cc );
+                    if ( t1 == NULL )
+                    {
+                        printf ( "Bad decomposition: %04X\n", cc );
+                        return ( 0 );
+                    }
+                    if ( ( t1->decompType != Atomic ) && ( t1->decompType != Implicit ) )
+                    {
+                        badValues++;
+                        printf ( "Infinite recursion: %08X %s\n", scalarValue, 
+                                 sp );
+                        return ( 0 );
+                    }
                 }
                 buf[numCharTokens] = cc;
                 numCharTokens++;