Fix detection of unfinished Unicode surrogate pair at end of string.

The U&'...' and U&"..." syntaxes silently discarded a surrogate pair
start (that is, a code between U+D800 and U+DBFF) if it occurred at
the very end of the string.  This seems like an obvious oversight,
since we throw an error for every other invalid combination of surrogate
characters, including the very same situation in E'...' syntax.

This has been wrong since the pair processing was added (in 9.0),
so back-patch to all supported branches.

Discussion: https://postgr.es/m/19113.1482337898@sss.pgh.pa.us

diff --git a/src/backend/parser/scan.l b/src/backend/parser/scan.l
index 36b21c764c87bf60c04a989364fecf89a83b9e79..30242d343d1502859e982cf289468d329d1886d1 100644 (file)
--- a/src/backend/parser/scan.l
+++ b/src/backend/parser/scan.l
@@ -1395,7 +1395,15 @@ litbuf_udeescape(unsigned char escape, core_yyscan_t yyscanner)
        }
    }
 
+   /* unfinished surrogate pair? */
+   if (pair_first)
+   {
+       ADVANCE_YYLLOC(in - litbuf + 3);                /* 3 for U&" */
+       yyerror("invalid Unicode surrogate pair");
+   }
+
    *out = '\0';
+
    /*
     * We could skip pg_verifymbstr if we didn't process any non-7-bit-ASCII
     * codes; but it's probably not worth the trouble, since this isn't