Fix to_char() to use ASCII-only case-folding rules where appropriate.
authorTom Lane <tgl@sss.pgh.pa.us>
Tue, 5 Mar 2013 18:02:38 +0000 (13:02 -0500)
committerTom Lane <tgl@sss.pgh.pa.us>
Tue, 5 Mar 2013 18:02:38 +0000 (13:02 -0500)
formatting.c used locale-dependent case folding rules in some code paths
where the result isn't supposed to be locale-dependent, for example
to_char(timestamp, 'DAY').  Since the source data is always just ASCII
in these cases, that usually didn't matter ... but it does matter in
Turkish locales, which have unusual treatment of "i" and "I".  To confuse
matters even more, the misbehavior was only visible in UTF8 encoding,
because in single-byte encodings we used pg_toupper/pg_tolower which
don't have locale-specific behavior for ASCII characters.  Fix by providing
intentionally ASCII-only case-folding functions and using these where
appropriate.  Per bug #7913 from Adnan Dursun.  Back-patch to all active
branches, since it's been like this for a long time.

src/backend/utils/adt/formatting.c
src/include/utils/formatting.h

index d69d27343b03cba4b4b60a7ab3eb71cc1dd90c65..6f267db0256cd0c5aae8a38595f503ab3e01ac19 100644 (file)
@@ -1491,12 +1491,7 @@ str_tolower(const char *buff, size_t nbytes, Oid collid)
    /* C/POSIX collations use this path regardless of database encoding */
    if (lc_ctype_is_c(collid))
    {
-       char       *p;
-
-       result = pnstrdup(buff, nbytes);
-
-       for (p = result; *p; p++)
-           *p = pg_ascii_tolower((unsigned char) *p);
+       result = asc_tolower(buff, nbytes);
    }
 #ifdef USE_WIDE_UPPER_LOWER
    else if (pg_database_encoding_max_length() > 1)
@@ -1612,12 +1607,7 @@ str_toupper(const char *buff, size_t nbytes, Oid collid)
    /* C/POSIX collations use this path regardless of database encoding */
    if (lc_ctype_is_c(collid))
    {
-       char       *p;
-
-       result = pnstrdup(buff, nbytes);
-
-       for (p = result; *p; p++)
-           *p = pg_ascii_toupper((unsigned char) *p);
+       result = asc_toupper(buff, nbytes);
    }
 #ifdef USE_WIDE_UPPER_LOWER
    else if (pg_database_encoding_max_length() > 1)
@@ -1734,23 +1724,7 @@ str_initcap(const char *buff, size_t nbytes, Oid collid)
    /* C/POSIX collations use this path regardless of database encoding */
    if (lc_ctype_is_c(collid))
    {
-       char       *p;
-
-       result = pnstrdup(buff, nbytes);
-
-       for (p = result; *p; p++)
-       {
-           char        c;
-
-           if (wasalnum)
-               *p = c = pg_ascii_tolower((unsigned char) *p);
-           else
-               *p = c = pg_ascii_toupper((unsigned char) *p);
-           /* we don't trust isalnum() here */
-           wasalnum = ((c >= 'A' && c <= 'Z') ||
-                       (c >= 'a' && c <= 'z') ||
-                       (c >= '0' && c <= '9'));
-       }
+       result = asc_initcap(buff, nbytes);
    }
 #ifdef USE_WIDE_UPPER_LOWER
    else if (pg_database_encoding_max_length() > 1)
@@ -1873,6 +1847,87 @@ str_initcap(const char *buff, size_t nbytes, Oid collid)
    return result;
 }
 
+/*
+ * ASCII-only lower function
+ *
+ * We pass the number of bytes so we can pass varlena and char*
+ * to this function.  The result is a palloc'd, null-terminated string.
+ */
+char *
+asc_tolower(const char *buff, size_t nbytes)
+{
+   char       *result;
+   char       *p;
+
+   if (!buff)
+       return NULL;
+
+   result = pnstrdup(buff, nbytes);
+
+   for (p = result; *p; p++)
+       *p = pg_ascii_tolower((unsigned char) *p);
+
+   return result;
+}
+
+/*
+ * ASCII-only upper function
+ *
+ * We pass the number of bytes so we can pass varlena and char*
+ * to this function.  The result is a palloc'd, null-terminated string.
+ */
+char *
+asc_toupper(const char *buff, size_t nbytes)
+{
+   char       *result;
+   char       *p;
+
+   if (!buff)
+       return NULL;
+
+   result = pnstrdup(buff, nbytes);
+
+   for (p = result; *p; p++)
+       *p = pg_ascii_toupper((unsigned char) *p);
+
+   return result;
+}
+
+/*
+ * ASCII-only initcap function
+ *
+ * We pass the number of bytes so we can pass varlena and char*
+ * to this function.  The result is a palloc'd, null-terminated string.
+ */
+char *
+asc_initcap(const char *buff, size_t nbytes)
+{
+   char       *result;
+   char       *p;
+   int         wasalnum = false;
+
+   if (!buff)
+       return NULL;
+
+   result = pnstrdup(buff, nbytes);
+
+   for (p = result; *p; p++)
+   {
+       char        c;
+
+       if (wasalnum)
+           *p = c = pg_ascii_tolower((unsigned char) *p);
+       else
+           *p = c = pg_ascii_toupper((unsigned char) *p);
+       /* we don't trust isalnum() here */
+       wasalnum = ((c >= 'A' && c <= 'Z') ||
+                   (c >= 'a' && c <= 'z') ||
+                   (c >= '0' && c <= '9'));
+   }
+
+   return result;
+}
+
 /* convenience routines for when the input is null-terminated */
 
 static char *
@@ -1893,6 +1948,20 @@ str_initcap_z(const char *buff, Oid collid)
    return str_initcap(buff, strlen(buff), collid);
 }
 
+static char *
+asc_tolower_z(const char *buff)
+{
+   return asc_tolower(buff, strlen(buff));
+}
+
+static char *
+asc_toupper_z(const char *buff)
+{
+   return asc_toupper(buff, strlen(buff));
+}
+
+/* asc_initcap_z is not currently needed */
+
 
 /* ----------
  * Skip TM / th in FROM_CHAR
@@ -2380,7 +2449,8 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                INVALID_FOR_INTERVAL;
                if (tmtcTzn(in))
                {
-                   char       *p = str_tolower_z(tmtcTzn(in), collid);
+                   /* We assume here that timezone names aren't localized */
+                   char       *p = asc_tolower_z(tmtcTzn(in));
 
                    strcpy(s, p);
                    pfree(p);
@@ -2427,7 +2497,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                    strcpy(s, str_toupper_z(localized_full_months[tm->tm_mon - 1], collid));
                else
                    sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
-                        str_toupper_z(months_full[tm->tm_mon - 1], collid));
+                        asc_toupper_z(months_full[tm->tm_mon - 1]));
                s += strlen(s);
                break;
            case DCH_Month:
@@ -2437,7 +2507,8 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_initcap_z(localized_full_months[tm->tm_mon - 1], collid));
                else
-                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9, months_full[tm->tm_mon - 1]);
+                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
+                           months_full[tm->tm_mon - 1]);
                s += strlen(s);
                break;
            case DCH_month:
@@ -2447,10 +2518,8 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_tolower_z(localized_full_months[tm->tm_mon - 1], collid));
                else
-               {
-                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9, months_full[tm->tm_mon - 1]);
-                   *s = pg_tolower((unsigned char) *s);
-               }
+                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
+                           asc_tolower_z(months_full[tm->tm_mon - 1]));
                s += strlen(s);
                break;
            case DCH_MON:
@@ -2460,7 +2529,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_toupper_z(localized_abbrev_months[tm->tm_mon - 1], collid));
                else
-                   strcpy(s, str_toupper_z(months[tm->tm_mon - 1], collid));
+                   strcpy(s, asc_toupper_z(months[tm->tm_mon - 1]));
                s += strlen(s);
                break;
            case DCH_Mon:
@@ -2480,10 +2549,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_tolower_z(localized_abbrev_months[tm->tm_mon - 1], collid));
                else
-               {
-                   strcpy(s, months[tm->tm_mon - 1]);
-                   *s = pg_tolower((unsigned char) *s);
-               }
+                   strcpy(s, asc_tolower_z(months[tm->tm_mon - 1]));
                s += strlen(s);
                break;
            case DCH_MM:
@@ -2498,7 +2564,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                    strcpy(s, str_toupper_z(localized_full_days[tm->tm_wday], collid));
                else
                    sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
-                           str_toupper_z(days[tm->tm_wday], collid));
+                           asc_toupper_z(days[tm->tm_wday]));
                s += strlen(s);
                break;
            case DCH_Day:
@@ -2506,7 +2572,8 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_initcap_z(localized_full_days[tm->tm_wday], collid));
                else
-                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9, days[tm->tm_wday]);
+                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
+                           days[tm->tm_wday]);
                s += strlen(s);
                break;
            case DCH_day:
@@ -2514,10 +2581,8 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_tolower_z(localized_full_days[tm->tm_wday], collid));
                else
-               {
-                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9, days[tm->tm_wday]);
-                   *s = pg_tolower((unsigned char) *s);
-               }
+                   sprintf(s, "%*s", S_FM(n->suffix) ? 0 : -9,
+                           asc_tolower_z(days[tm->tm_wday]));
                s += strlen(s);
                break;
            case DCH_DY:
@@ -2525,7 +2590,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_toupper_z(localized_abbrev_days[tm->tm_wday], collid));
                else
-                   strcpy(s, str_toupper_z(days_short[tm->tm_wday], collid));
+                   strcpy(s, asc_toupper_z(days_short[tm->tm_wday]));
                s += strlen(s);
                break;
            case DCH_Dy:
@@ -2541,10 +2606,7 @@ DCH_to_char(FormatNode *node, bool is_interval, TmToChar *in, char *out, Oid col
                if (S_TM(n->suffix))
                    strcpy(s, str_tolower_z(localized_abbrev_days[tm->tm_wday], collid));
                else
-               {
-                   strcpy(s, days_short[tm->tm_wday]);
-                   *s = pg_tolower((unsigned char) *s);
-               }
+                   strcpy(s, asc_tolower_z(days_short[tm->tm_wday]));
                s += strlen(s);
                break;
            case DCH_DDD:
@@ -4651,12 +4713,12 @@ NUM_processor(FormatNode *node, NUMDesc *Num, char *inout, char *number,
                case NUM_rn:
                    if (IS_FILLMODE(Np->Num))
                    {
-                       strcpy(Np->inout_p, str_tolower_z(Np->number_p, collid));
+                       strcpy(Np->inout_p, asc_tolower_z(Np->number_p));
                        Np->inout_p += strlen(Np->inout_p) - 1;
                    }
                    else
                    {
-                       sprintf(Np->inout_p, "%15s", str_tolower_z(Np->number_p, collid));
+                       sprintf(Np->inout_p, "%15s", asc_tolower_z(Np->number_p));
                        Np->inout_p += strlen(Np->inout_p) - 1;
                    }
                    break;
index 175d396297cd39cbbd5da979699c1092b4e1d8e8..fa836051337c49243286d31db608ba97ce485e23 100644 (file)
@@ -24,6 +24,10 @@ extern char *str_tolower(const char *buff, size_t nbytes, Oid collid);
 extern char *str_toupper(const char *buff, size_t nbytes, Oid collid);
 extern char *str_initcap(const char *buff, size_t nbytes, Oid collid);
 
+extern char *asc_tolower(const char *buff, size_t nbytes);
+extern char *asc_toupper(const char *buff, size_t nbytes);
+extern char *asc_initcap(const char *buff, size_t nbytes);
+
 extern Datum timestamp_to_char(PG_FUNCTION_ARGS);
 extern Datum timestamptz_to_char(PG_FUNCTION_ARGS);
 extern Datum interval_to_char(PG_FUNCTION_ARGS);