Allow "_" characters to appear between any two digits in an integer, real or hexadecimal SQL literal.

FossilOrigin-Name: 0e6700f43f133510c8049b2c5a2610cb3be29da7ed4d39b1fa65dc22c4cf529b

4 files changed, 91 insertions, 20 deletions

diff --git a/src/parse.y b/src/parse.y
index 12621b434..37c9fa8bc 100644
--- a/src/parse.y
+++ b/src/parse.y
@@ -1935,6 +1935,12 @@ filter_clause(A) ::= FILTER LP WHERE expr(X) RP.  { A = X; }
   SPAN            /* The span operator */
   ERROR           /* An expression containing an error */
 .
+
+term(A) ::= QNUMBER(X). {
+  A=tokenExpr(pParse,@X,X);
+  sqlite3DequoteNumber(pParse, A);
+}
+
 /* There must be no more than 255 tokens defined above.  If this grammar
 ** is extended with new rules and tokens, they must either be so few in
 ** number that TK_SPAN is no more than 255, or else the new tokens must
diff --git a/src/sqliteInt.h b/src/sqliteInt.h
index 1c1055f42..f5920748b 100644
--- a/src/sqliteInt.h
+++ b/src/sqliteInt.h
@@ -609,6 +609,8 @@
 # define SQLITE_OMIT_ALTERTABLE
 #endif
 
+#define SQLITE_DIGIT_SEPARATOR '_'
+
 /*
 ** Return true (non-zero) if the input is an integer that is too large
 ** to fit in 32-bits.  This macro is used inside of various testcase()
@@ -4795,6 +4797,7 @@ int sqlite3ErrorToParser(sqlite3*,int);
 void sqlite3Dequote(char*);
 void sqlite3DequoteExpr(Expr*);
 void sqlite3DequoteToken(Token*);
+void sqlite3DequoteNumber(Parse*, Expr*);
 void sqlite3TokenInit(Token*,char*);
 int sqlite3KeywordCode(const unsigned char*, int);
 int sqlite3RunParser(Parse*, const char*);
diff --git a/src/tokenize.c b/src/tokenize.c
index f4d013dee..65d1fbf35 100644
--- a/src/tokenize.c
+++ b/src/tokenize.c
@@ -437,27 +437,58 @@ int sqlite3GetToken(const unsigned char *z, int *tokenType){
       *tokenType = TK_INTEGER;
 #ifndef SQLITE_OMIT_HEX_INTEGER
       if( z[0]=='0' && (z[1]=='x' || z[1]=='X') && sqlite3Isxdigit(z[2]) ){
-        for(i=3; sqlite3Isxdigit(z[i]); i++){}
-        return i;
-      }
+        for(i=3; 1; i++){
+          if( sqlite3Isxdigit(z[i])==0 ){
+            if( z[i]==SQLITE_DIGIT_SEPARATOR ){
+              *tokenType = TK_QNUMBER;
+            }else{
+              break;
+            }
+          }
+        }
+      }else
 #endif
-      for(i=0; sqlite3Isdigit(z[i]); i++){}
+        {
+        for(i=0; 1; i++){
+          if( sqlite3Isdigit(z[i])==0 ){
+            if( z[i]==SQLITE_DIGIT_SEPARATOR ){
+              *tokenType = TK_QNUMBER;
+            }else{
+              break;
+            }
+          }
+        }
 #ifndef SQLITE_OMIT_FLOATING_POINT
-      if( z[i]=='.' ){
-        i++;
-        while( sqlite3Isdigit(z[i]) ){ i++; }
-        *tokenType = TK_FLOAT;
-      }
-      if( (z[i]=='e' || z[i]=='E') &&
-           ( sqlite3Isdigit(z[i+1]) 
-            || ((z[i+1]=='+' || z[i+1]=='-') && sqlite3Isdigit(z[i+2]))
-           )
-      ){
-        i += 2;
-        while( sqlite3Isdigit(z[i]) ){ i++; }
-        *tokenType = TK_FLOAT;
-      }
+        if( z[i]=='.' ){
+          if( *tokenType==TK_INTEGER ) *tokenType = TK_FLOAT;
+          for(i++; 1; i++){
+            if( sqlite3Isdigit(z[i])==0 ){
+              if( z[i]==SQLITE_DIGIT_SEPARATOR ){
+                *tokenType = TK_QNUMBER;
+              }else{
+                break;
+              }
+            }
+          }
+        }
+        if( (z[i]=='e' || z[i]=='E') &&
+             ( sqlite3Isdigit(z[i+1]) 
+              || ((z[i+1]=='+' || z[i+1]=='-') && sqlite3Isdigit(z[i+2]))
+             )
+        ){
+          if( *tokenType==TK_INTEGER ) *tokenType = TK_FLOAT;
+          for(i+=2; 1; i++){
+            if( sqlite3Isdigit(z[i])==0 ){
+              if( z[i]==SQLITE_DIGIT_SEPARATOR ){
+                *tokenType = TK_QNUMBER;
+              }else{
+                break;
+              }
+            }
+          }
+        }
 #endif
+      }
       while( IdChar(z[i]) ){
         *tokenType = TK_ILLEGAL;
         i++;
@@ -622,10 +653,13 @@ int sqlite3RunParser(Parse *pParse, const char *zSql){
     if( tokenType>=TK_WINDOW ){
       assert( tokenType==TK_SPACE || tokenType==TK_OVER || tokenType==TK_FILTER
            || tokenType==TK_ILLEGAL || tokenType==TK_WINDOW 
+           || tokenType==TK_QNUMBER
       );
 #else
     if( tokenType>=TK_SPACE ){
-      assert( tokenType==TK_SPACE || tokenType==TK_ILLEGAL );
+      assert( tokenType==TK_SPACE || tokenType==TK_ILLEGAL 
+           || tokenType==TK_QNUMBER 
+      );
 #endif /* SQLITE_OMIT_WINDOWFUNC */
       if( AtomicLoad(&db->u1.isInterrupted) ){
         pParse->rc = SQLITE_INTERRUPT;
@@ -658,7 +692,7 @@ int sqlite3RunParser(Parse *pParse, const char *zSql){
         assert( n==6 );
         tokenType = analyzeFilterKeyword((const u8*)&zSql[6], lastTokenParsed);
 #endif /* SQLITE_OMIT_WINDOWFUNC */
-      }else{
+      }else if( tokenType!=TK_QNUMBER ){
         Token x;
         x.z = zSql;
         x.n = n;
diff --git a/src/util.c b/src/util.c
index 4aa82d063..bdf92d9c4 100644
--- a/src/util.c
+++ b/src/util.c
@@ -312,6 +312,34 @@ void sqlite3DequoteExpr(Expr *p){
 }
 
 /*
+** Expression p is a QNUMBER (quoted number). Dequote the value in p->u.zToken
+** and set the type to INTEGER or FLOAT. "Quoted" integers or floats are those
+** that contain '_' characters that must be removed before further processing.
+*/
+void sqlite3DequoteNumber(Parse *pParse, Expr *p){
+  if( p ){
+    const char *pIn = p->u.zToken;
+    char *pOut = p->u.zToken;
+    int bHex = (pIn[0]=='0' && (pIn[1]=='x' || pIn[1]=='X'));
+    assert( p->op==TK_QNUMBER );
+    p->op = TK_INTEGER;
+    do {
+      if( *pIn!=SQLITE_DIGIT_SEPARATOR ){
+        *pOut++ = *pIn;
+        if( *pIn=='e' || *pIn=='E' || *pIn=='.' ) p->op = TK_FLOAT;
+      }else{
+        if( (bHex==0 && (!sqlite3Isdigit(pIn[-1]) || !sqlite3Isdigit(pIn[1])))
+         || (bHex==1 && (!sqlite3Isxdigit(pIn[-1]) || !sqlite3Isxdigit(pIn[1])))
+        ){
+          sqlite3ErrorMsg(pParse, "unrecognized token: \"%s\"", p->u.zToken);
+        }
+      }
+    }while( *pIn++ );
+    if( bHex ) p->op = TK_INTEGER;
+  }
+}
+
+/*
 ** If the input token p is quoted, try to adjust the token to remove
 ** the quotes.  This is not always possible:
 **