mbuiter: Implement multi-byte per encoding error (MEE) consistently.

* lib/mbuiter.h: Include mbiter-aux.h.
(struct mbuiter_multi): Add field is_utf8.
(mbuiter_multi_next): Invoke mbiter_is_utf8,
mbiter_utf8_maximal_subpart.
(mbui_init): Initialize the field is_utf8.
* modules/mbuiter (Depends-on): Add mbiter-aux.
* tests/test-mbsstr2.c (OR): New macro, copied from
tests/test-mbsnlen.c.
(main): Add test cases with incomplete characters.
* tests/test-mbsstr1.c: Update comments.
* tests/test-mbsstr3.c: Likewise.
* tests/test-mbscasestr2.c (OR): New macro, copied from
tests/test-mbsnlen.c.
(main): Add test cases with incomplete characters.
* tests/test-mbscasestr1.c: Update comments.
* tests/test-mbscasestr3.c: Likewise.
* tests/test-mbscasestr4.c: Likewise.

ChangeLog

@@ -1,3 +1,24 @@
+2026-05-25  Bruno Haible  <bruno@clisp.org>
+
+	mbuiter: Implement multi-byte per encoding error (MEE) consistently.
+	* lib/mbuiter.h: Include mbiter-aux.h.
+	(struct mbuiter_multi): Add field is_utf8.
+	(mbuiter_multi_next): Invoke mbiter_is_utf8,
+	mbiter_utf8_maximal_subpart.
+	(mbui_init): Initialize the field is_utf8.
+	* modules/mbuiter (Depends-on): Add mbiter-aux.
+	* tests/test-mbsstr2.c (OR): New macro, copied from
+	tests/test-mbsnlen.c.
+	(main): Add test cases with incomplete characters.
+	* tests/test-mbsstr1.c: Update comments.
+	* tests/test-mbsstr3.c: Likewise.
+	* tests/test-mbscasestr2.c (OR): New macro, copied from
+	tests/test-mbsnlen.c.
+	(main): Add test cases with incomplete characters.
+	* tests/test-mbscasestr1.c: Update comments.
+	* tests/test-mbscasestr3.c: Likewise.
+	* tests/test-mbscasestr4.c: Likewise.
+
 2026-05-25  Bruno Haible  <bruno@clisp.org>
 
 	mbiterf: Implement multi-byte per encoding error (MEE) consistently.

lib/mbuiter.h

@@ -103,6 +103,7 @@
 #include <wchar.h>
 
 #include "mbchar.h"
+#include "mbiter-aux.h"
 #include "strnlen1.h"
 
 _GL_INLINE_HEADER_BEGIN
@@ -128,6 +129,7 @@ struct mbuiter_multi
                          */
   bool next_done;       /* true if mbui_avail has already filled the following */
   unsigned int cur_max; /* A cache of MB_CUR_MAX.  */
+  int is_utf8;          /* A cache of mbiter_is_utf8.  */
   struct mbchar cur;    /* the current character:
         const char *cur.ptr          pointer to current character
         The following are only valid after mbui_avail.
@@ -164,15 +166,18 @@ mbuiter_multi_next (struct mbuiter_multi *iter)
       assert (mbsinit (&iter->state));
       #if !GNULIB_MBRTOC32_REGULAR
       iter->in_shift = true;
-    with_shift:
+    with_shift:;
       #endif
-      iter->cur.bytes = mbrtoc32 (&iter->cur.wc, iter->cur.ptr,
-                                  strnlen1 (iter->cur.ptr, iter->cur_max),
+      size_t avail_bytes = strnlen1 (iter->cur.ptr, iter->cur_max);
+      iter->cur.bytes = mbrtoc32 (&iter->cur.wc, iter->cur.ptr, avail_bytes,
                                   &iter->state);
       if (iter->cur.bytes == (size_t) -1)
         {
           /* An invalid multibyte sequence was encountered.  */
-          iter->cur.bytes = 1;
+          iter->cur.bytes =
+            (mbiter_is_utf8 (&iter->is_utf8)
+             ? mbiter_utf8_maximal_subpart (iter->cur.ptr, avail_bytes)
+             : 1);
           iter->cur.wc_valid = false;
           /* Allow the next invocation to continue from a sane state.  */
           #if !GNULIB_MBRTOC32_REGULAR
@@ -243,14 +248,14 @@ typedef struct mbuiter_multi mbui_iterator_t;
   ((iter).cur.ptr = (startptr), \
    (iter).in_shift = false, mbszero (&(iter).state), \
    (iter).next_done = false, \
-   (iter).cur_max = MB_CUR_MAX)
+   (iter).cur_max = MB_CUR_MAX, (iter).is_utf8 = -1)
 #else
 /* Optimized: no in_shift.  */
 #define mbui_init(iter, startptr) \
   ((iter).cur.ptr = (startptr), \
    mbszero (&(iter).state), \
    (iter).next_done = false, \
-   (iter).cur_max = MB_CUR_MAX)
+   (iter).cur_max = MB_CUR_MAX, (iter).is_utf8 = -1)
 #endif
 #define mbui_avail(iter) \
   (mbuiter_multi_next (&(iter)), !mb_isnul ((iter).cur))

modules/mbuiter

@@ -13,6 +13,7 @@ mbchar
 mbrtoc32
 mbsinit
 mbszero
+mbiter-aux
 uchar-h
 bool
 strnlen1

tests/test-mbscasestr1.c

@@ -1,4 +1,4 @@
-/* Test of case-insensitive searching in a string.
+/* Test of case-insensitive searching in a string in the "C" locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify

tests/test-mbscasestr2.c

@@ -1,4 +1,4 @@
-/* Test of searching in a string.
+/* Test of searching in a string in a UTF-8 locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify
@@ -25,6 +25,20 @@
 
 #include "macros.h"
 
+/* The mcel-based implementation of mbsnlen behaves differently than the
+   original one.  Namely, for invalid/incomplete byte sequences:
+   Where we ideally should have multi-byte-per-encoding-error (MEE) behaviour
+   everywhere, mcel implements single-byte-per-encoding-error (SEE) behaviour.
+   See <https://lists.gnu.org/archive/html/bug-gnulib/2023-07/msg00131.html>,
+       <https://lists.gnu.org/archive/html/bug-gnulib/2023-07/msg00145.html>.
+   Therefore, here we have different expected results, depending on the
+   implementation.  */
+#if GNULIB_MCEL_PREFER
+# define OR(a,b) b
+#else
+# define OR(a,b) a
+#endif
+
 int
 main ()
 {
@@ -50,6 +64,74 @@ main ()
     ASSERT (result == NULL);
   }
 
+  /* Incomplete characters.  See
+     https://www.unicode.org/versions/Unicode15.0.0/ch03.pdf
+     page 128 table 3-11.  */
+
+  /* "\341\200\240" = 0xE1 0x80 0xA0 = U+1020.  */
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbscasestr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbscasestr (input, "\341\200");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbscasestr (input, "\200\341");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+
+  /* "\360\221\222\240" = 0xF0 0x91 0x92 0xA0 = U+114A0.  */
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "\360\221\222");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "\221\222\360\221");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "\221\222\360");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "\222\360\221");
+    ASSERT (result == OR (NULL, input + 3));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbscasestr (input, "\222\360");
+    ASSERT (result == OR (NULL, input + 3));
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbscasestr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbscasestr (input, "\360\221");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbscasestr (input, "\221\360");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+
   {
     const char input[] = "\303\204BC \303\204BCD\303\204B \303\204BCD\303\204BCD\303\204BDE"; /* "ÄBC ÄBCDÄB ÄBCDÄBCDÄBDE" */
     const char *result = mbscasestr (input, "\303\244BCD\303\204BD"); /* "äBCDÄBD" */

tests/test-mbscasestr3.c

@@ -1,4 +1,4 @@
-/* Test of case-insensitive searching in a string.
+/* Test of case-insensitive searching in a string in a GB18030 locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify

tests/test-mbscasestr4.c

@@ -1,4 +1,4 @@
-/* Test of case-insensitive searching in a string.
+/* Test of case-insensitive searching in a string in a Turkish locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify

tests/test-mbsstr1.c

@@ -1,4 +1,4 @@
-/* Test of searching in a string.
+/* Test of searching in a string in the "C" locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify

tests/test-mbsstr2.c

@@ -1,4 +1,4 @@
-/* Test of searching in a string.
+/* Test of searching in a string in a UTF-8 locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify
@@ -25,6 +25,20 @@
 
 #include "macros.h"
 
+/* The mcel-based implementation of mbsnlen behaves differently than the
+   original one.  Namely, for invalid/incomplete byte sequences:
+   Where we ideally should have multi-byte-per-encoding-error (MEE) behaviour
+   everywhere, mcel implements single-byte-per-encoding-error (SEE) behaviour.
+   See <https://lists.gnu.org/archive/html/bug-gnulib/2023-07/msg00131.html>,
+       <https://lists.gnu.org/archive/html/bug-gnulib/2023-07/msg00145.html>.
+   Therefore, here we have different expected results, depending on the
+   implementation.  */
+#if GNULIB_MCEL_PREFER
+# define OR(a,b) b
+#else
+# define OR(a,b) a
+#endif
+
 int
 main ()
 {
@@ -50,6 +64,74 @@ main ()
     ASSERT (result == NULL);
   }
 
+  /* Incomplete characters.  See
+     https://www.unicode.org/versions/Unicode15.0.0/ch03.pdf
+     page 128 table 3-11.  */
+
+  /* "\341\200\240" = 0xE1 0x80 0xA0 = U+1020.  */
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbsstr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbsstr (input, "\341\200");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\341\200\341\200";
+    const char *result = mbsstr (input, "\200\341");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+
+  /* "\360\221\222\240" = 0xF0 0x91 0x92 0xA0 = U+114A0.  */
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "\360\221\222");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "\221\222\360\221");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "\221\222\360");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "\222\360\221");
+    ASSERT (result == OR (NULL, input + 3));
+  }
+  {
+    const char input[] = "f\360\221\222\360\221\222";
+    const char *result = mbsstr (input, "\222\360");
+    ASSERT (result == OR (NULL, input + 3));
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbsstr (input, "");
+    ASSERT (result == input);
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbsstr (input, "\360\221");
+    ASSERT (result == input + 1);
+  }
+  {
+    const char input[] = "f\360\221\360\221";
+    const char *result = mbsstr (input, "\221\360");
+    ASSERT (result == OR (NULL, input + 2));
+  }
+
   {
     const char input[] = "\303\204BC \303\204BCD\303\204B \303\204BCD\303\204BCD\303\204BDE"; /* "ÄBC ÄBCDÄB ÄBCDÄBCDÄBDE" */
     const char *result = mbsstr (input, "\303\204BCD\303\204BD"); /* "ÄBCDÄBD" */

tests/test-mbsstr3.c

@@ -1,4 +1,4 @@
-/* Test of searching in a string.
+/* Test of searching in a string in a GB18030 locale.
    Copyright (C) 2007-2026 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify