From a2f806b656f4f5f0693292a814dc992d5f083b46 Mon Sep 17 00:00:00 2001
From: Ben Pfaff <blp@cs.stanford.edu>
Date: Mon, 10 Feb 2014 20:59:49 -0800
Subject: [PATCH] i18n: Correctly recode windows-1258 and others with combining
 diacritics.

Without this change, recoding a string from windows-1258 often drops the
last character.
---
 src/libpspp/i18n.c    | 90 ++++++++++++++++++++++++++-----------------
 tests/libpspp/i18n.at |  5 +++
 2 files changed, 60 insertions(+), 35 deletions(-)

diff --git a/src/libpspp/i18n.c b/src/libpspp/i18n.c
index cdcf57003b..10b3927f9e 100644
--- a/src/libpspp/i18n.c
+++ b/src/libpspp/i18n.c
@@ -1,5 +1,5 @@
 /* PSPP - a program for statistical analysis.
-   Copyright (C) 2006, 2009, 2010, 2011, 2012, 2013 Free Software Foundation, Inc.
+   Copyright (C) 2006, 2009, 2010, 2011, 2012, 2013, 2014 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -149,52 +149,72 @@ recode_string_len (const char *to, const char *from,
    small. */
 static ssize_t
 try_recode (iconv_t conv,
-            const char *ip, size_t inbytes,
-            char *op_, size_t outbytes)
+            const char *in, size_t inbytes,
+            char *out_, size_t outbytes)
 {
   /* FIXME: Need to ensure that this char is valid in the target encoding */
   const char fallbackchar = '?';
-  char *op = op_;
+  char *out = out_;
+  int i;
 
   /* Put the converter into the initial shift state, in case there was any
      state information left over from its last usage. */
   iconv (conv, NULL, 0, NULL, 0);
 
-  while (iconv (conv, (ICONV_CONST char **) &ip, &inbytes,
-                &op, &outbytes) == -1)
-    switch (errno)
-      {
-      case EINVAL:
-        if (outbytes < 2)
-          return -1;
-        *op++ = fallbackchar;
-        *op = '\0';
-        return op - op_;
-
-      case EILSEQ:
-        if (outbytes == 0)
-          return -1;
-        *op++ = fallbackchar;
-        outbytes--;
-        ip++;
-        inbytes--;
-        break;
-
-      case E2BIG:
-        return -1;
-
-      default:
-        /* should never happen */
-        fprintf (stderr, "Character conversion error: %s\n", strerror (errno));
-        NOT_REACHED ();
-        break;
-      }
+  /* Do two rounds of iconv() calls:
+
+     - The first round does the bulk of the conversion using the
+       caller-supplied input data..
+
+     - The second round flushes any leftover output.  This has a real effect
+       with input encodings that use combining diacritics, e.g. without the
+       second round the last character tends to gets dropped when converting
+       from windows-1258 to other encodings.
+  */
+  for (i = 0; i < 2; i++)
+    {
+      ICONV_CONST char **inp = i ? NULL : (ICONV_CONST char **) &in;
+      size_t *inbytesp = i ? NULL : &inbytes;
+
+      while (iconv (conv, inp, inbytesp, &out, &outbytes) == -1)
+        switch (errno)
+          {
+          case EINVAL:
+            if (outbytes < 2)
+              return -1;
+            *out++ = fallbackchar;
+            *out = '\0';
+            return out - out_;
+
+          case EILSEQ:
+            if (outbytes == 0)
+              return -1;
+            *out++ = fallbackchar;
+            outbytes--;
+            if (inp)
+              {
+                in++;
+                inbytes--;
+              }
+            break;
+
+          case E2BIG:
+            return -1;
+
+          default:
+            /* should never happen */
+            fprintf (stderr, "Character conversion error: %s\n",
+                     strerror (errno));
+            NOT_REACHED ();
+            break;
+          }
+    }
 
   if (outbytes == 0)
     return -1;
 
-  *op = '\0';
-  return op - op_;
+  *out = '\0';
+  return out - out_;
 }
 
 /* Converts the string TEXT, which should be encoded in FROM-encoding, to a
diff --git a/tests/libpspp/i18n.at b/tests/libpspp/i18n.at
index 5f4bb65961..974e76f590 100644
--- a/tests/libpspp/i18n.at
+++ b/tests/libpspp/i18n.at
@@ -32,6 +32,11 @@ CHECK_I18N_RECODE([invalid UTF-8 to ISO-8859-1], [UTF-8], [ISO-8859-1],
 CHECK_I18N_RECODE([truncated UTF-8 to ISO-8559-1], [UTF-8], [ISO-8859-1],
                   [xy\302], [xy?])
 
+# Checks for a bug that caused the last character to be dropped in conversions
+# from encodings that have combining diacritics (e.g. windows-1258).
+CHECK_I18N_RECODE([dropped final character in windows-1258], [windows-1258],
+                  [UTF8], [aeiou], [aeiou])
+
 dnl The input to this test is 7 bytes long and the expected output is 9 bytes.
 dnl So it should exercise the E2BIG case 
 CHECK_I18N_RECODE([from ISO-8859-1 to UTF-8 with overflow], 
-- 
2.30.2