/*
** 2015-08-12
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
******************************************************************************
**
** This SQLite extension implements JSON functions. The interface is
** modeled after MySQL JSON functions:
**
** https://dev.mysql.com/doc/refman/5.7/en/json.html
**
** For the time being, all JSON is stored as pure text. (We might add
** a JSONB type in the future which stores a binary encoding of JSON in
** a BLOB, but there is no support for JSONB in the current implementation.
** This implementation parses JSON text at 250 MB/s, so it is hard to see
** how JSONB might improve on that.)
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_JSON1)
#if !defined(SQLITEINT_H)
#include "sqlite3ext.h"
#endif
SQLITE_EXTENSION_INIT1
/* If compiling this extension separately (why would anybody do that when
** it is built into the amalgamation?) we must set NDEBUG if SQLITE_DEBUG
** is not defined *before* including <assert.h>, in order to disable asserts().
*/
#if !defined(SQLITE_AMALGAMATION) && !defined(SQLITE_DEBUG)
# define NDEBUG 1
#endif
#include <assert.h>
#include <string.h>
#include <stdlib.h>
#include <stdarg.h>
/* Mark a function parameter as unused, to suppress nuisance compiler
** warnings. */
#ifndef UNUSED_PARAM
# define UNUSED_PARAM(X) (void)(X)
#endif
#ifndef LARGEST_INT64
# define LARGEST_INT64 (0xffffffff|(((sqlite3_int64)0x7fffffff)<<32))
# define SMALLEST_INT64 (((sqlite3_int64)-1) - LARGEST_INT64)
#endif
#ifndef deliberate_fall_through
# define deliberate_fall_through
#endif
/*
** Versions of isspace(), isalnum() and isdigit() to which it is safe
** to pass signed char values.
*/
#ifdef sqlite3Isdigit
/* Use the SQLite core versions if this routine is part of the
** SQLite amalgamation */
# define safe_isdigit(x) sqlite3Isdigit(x)
# define safe_isalnum(x) sqlite3Isalnum(x)
# define safe_isxdigit(x) sqlite3Isxdigit(x)
#else
/* Use the standard library for separate compilation */
#include <ctype.h> /* amalgamator: keep */
# define safe_isdigit(x) isdigit((unsigned char)(x))
# define safe_isalnum(x) isalnum((unsigned char)(x))
# define safe_isxdigit(x) isxdigit((unsigned char)(x))
#endif
/*
** Growing our own isspace() routine this way is twice as fast as
** the library isspace() function, resulting in a 7% overall performance
** increase for the parser. (Ubuntu14.10 gcc 4.8.4 x64 with -Os).
*/
static const char jsonIsSpace[] = {
0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
};
#define safe_isspace(x) (jsonIsSpace[(unsigned char)x])
#ifndef SQLITE_AMALGAMATION
/* Unsigned integer types. These are already defined in the sqliteInt.h,
** but the definitions need to be repeated for separate compilation. */
typedef sqlite3_uint64 u64;
typedef unsigned int u32;
typedef unsigned short int u16;
typedef unsigned char u8;
# if defined(SQLITE_COVERAGE_TEST) || defined(SQLITE_MUTATION_TEST)
# define SQLITE_OMIT_AUXILIARY_SAFETY_CHECKS 1
# endif
# if defined(SQLITE_OMIT_AUXILIARY_SAFETY_CHECKS)
# define ALWAYS(X) (1)
# define NEVER(X) (0)
# elif !defined(NDEBUG)
# define ALWAYS(X) ((X)?1:(assert(0),0))
# define NEVER(X) ((X)?(assert(0),1):0)
# else
# define ALWAYS(X) (X)
# define NEVER(X) (X)
# endif
# define testcase(X)
#endif
#if !defined(SQLITE_DEBUG) && !defined(SQLITE_COVERAGE_TEST)
# define VVA(X)
#else
# define VVA(X) X
#endif
/*
** Some of the testcase() macros in this file are problematic for gcov
** in that they generate false-miss errors randomly. This is a gcov problem,
** not a problem in this case. But to work around it, we disable the
** problematic test cases for production builds.
*/
#define json_testcase(X)
/* Objects */
typedef struct JsonString JsonString;
typedef struct JsonNode JsonNode;
typedef struct JsonParse JsonParse;
/* An instance of this object represents a JSON string
** under construction. Really, this is a generic string accumulator
** that can be and is used to create strings other than JSON.
*/
struct JsonString {
sqlite3_context *pCtx; /* Function context - put error messages here */
char *zBuf; /* Append JSON content here */
u64 nAlloc; /* Bytes of storage available in zBuf[] */
u64 nUsed; /* Bytes of zBuf[] currently used */
u8 bStatic; /* True if zBuf is static space */
u8 bErr; /* True if an error has been encountered */
char zSpace[100]; /* Initial static space */
};
/* JSON type values
*/
#define JSON_NULL 0
#define JSON_TRUE 1
#define JSON_FALSE 2
#define JSON_INT 3
#define JSON_REAL 4
#define JSON_STRING 5
#define JSON_ARRAY 6
#define JSON_OBJECT 7
/* The "subtype" set for JSON values */
#define JSON_SUBTYPE 74 /* Ascii for "J" */
/*
** Names of the various JSON types:
*/
static const char * const jsonType[] = {
"null", "true", "false", "integer", "real", "text", "array", "object"
};
/* Bit values for the JsonNode.jnFlag field
*/
#define JNODE_RAW 0x01 /* Content is raw, not JSON encoded */
#define JNODE_ESCAPE 0x02 /* Content is text with \ escapes */
#define JNODE_REMOVE 0x04 /* Do not output */
#define JNODE_REPLACE 0x08 /* Replace with JsonNode.u.iReplace */
#define JNODE_PATCH 0x10 /* Patch with JsonNode.u.pPatch */
#define JNODE_APPEND 0x20 /* More ARRAY/OBJECT entries at u.iAppend */
#define JNODE_LABEL 0x40 /* Is a label of an object */
/* A single node of parsed JSON
*/
struct JsonNode {
u8 eType; /* One of the JSON_ type values */
u8 jnFlags; /* JNODE flags */
u8 eU; /* Which union element to use */
u32 n; /* Bytes of content, or number of sub-nodes */
union {
const char *zJContent; /* 1: Content for INT, REAL, and STRING */
u32 iAppend; /* 2: More terms for ARRAY and OBJECT */
u32 iKey; /* 3: Key for ARRAY objects in json_tree() */
u32 iReplace; /* 4: Replacement content for JNODE_REPLACE */
JsonNode *pPatch; /* 5: Node chain of patch for JNODE_PATCH */
} u;
};
/* A completely parsed JSON string
*/
struct JsonParse {
u32 nNode; /* Number of slots of aNode[] used */
u32 nAlloc; /* Number of slots of aNode[] allocated */
JsonNode *aNode; /* Array of nodes containing the parse */
const char *zJson; /* Original JSON string */
u32 *aUp; /* Index of parent of each node */
u8 oom; /* Set to true if out of memory */
u8 nErr; /* Number of errors seen */
u16 iDepth; /* Nesting depth */
int nJson; /* Length of the zJson string in bytes */
u32 iHold; /* Replace cache line with the lowest iHold value */
};
/*
** Maximum nesting depth of JSON for this implementation.
**
** This limit is needed to avoid a stack overflow in the recursive
** descent parser. A depth of 2000 is far deeper than any sane JSON
** should go.
*/
#define JSON_MAX_DEPTH 2000
/**************************************************************************
** Utility routines for dealing with JsonString objects
**************************************************************************/
/* Set the JsonString object to an empty string
*/
static void jsonZero(JsonString *p){
p->zBuf = p->zSpace;
p->nAlloc = sizeof(p->zSpace);
p->nUsed = 0;
p->bStatic = 1;
}
/* Initialize the JsonString object
*/
static void jsonInit(JsonString *p, sqlite3_context *pCtx){
p->pCtx = pCtx;
p->bErr = 0;
jsonZero(p);
}
/* Free all allocated memory and reset the JsonString object back to its
** initial state.
*/
static void jsonReset(JsonString *p){
if( !p->bStatic ) sqlite3_free(p->zBuf);
jsonZero(p);
}
/* Report an out-of-memory (OOM) condition
*/
static void jsonOom(JsonString *p){
p->bErr = 1;
sqlite3_result_error_nomem(p->pCtx);
jsonReset(p);
}
/* Enlarge pJson->zBuf so that it can hold at least N more bytes.
** Return zero on success. Return non-zero on an OOM error
*/
static int jsonGrow(JsonString *p, u32 N){
u64 nTotal = N<p->nAlloc ? p->nAlloc*2 : p->nAlloc+N+10;
char *zNew;
if( p->bStatic ){
if( p->bErr ) return 1;
zNew = sqlite3_malloc64(nTotal);
if( zNew==0 ){
jsonOom(p);
return SQLITE_NOMEM;
}
memcpy(zNew, p->zBuf, (size_t)p->nUsed);
p->zBuf = zNew;
p->bStatic = 0;
}else{
zNew = sqlite3_realloc64(p->zBuf, nTotal);
if( zNew==0 ){
jsonOom(p);
return SQLITE_NOMEM;
}
p->zBuf = zNew;
}
p->nAlloc = nTotal;
return SQLITE_OK;
}
/* Append N bytes from zIn onto the end of the JsonString string.
*/
static void jsonAppendRaw(JsonString *p, const char *zIn, u32 N){
if( N==0 ) return;
if( (N+p->nUsed >= p->nAlloc) && jsonGrow(p,N)!=0 ) return;
memcpy(p->zBuf+p->nUsed, zIn, N);
p->nUsed += N;
}
/* Append formatted text (not to exceed N bytes) to the JsonString.
*/
static void jsonPrintf(int N, JsonString *p, const char *zFormat, ...){
va_list ap;
if( (p->nUsed + N >= p->nAlloc) && jsonGrow(p, N) ) return;
va_start(ap, zFormat);
sqlite3_vsnprintf(N, p->zBuf+p->nUsed, zFormat, ap);
va_end(ap);
p->nUsed += (int)strlen(p->zBuf+p->nUsed);
}
/* Append a single character
*/
static void jsonAppendChar(JsonString *p, char c){
if( p->nUsed>=p->nAlloc && jsonGrow(p,1)!=0 ) return;
p->zBuf[p->nUsed++] = c;
}
/* Append a comma separator to the output buffer, if the previous
** character is not '[' or '{'.
*/
static void jsonAppendSeparator(JsonString *p){
char c;
if( p->nUsed==0 ) return;
c = p->zBuf[p->nUsed-1];
if( c!='[' && c!='{' ) jsonAppendChar(p, ',');
}
/* Append the N-byte string in zIn to the end of the JsonString string
** under construction. Enclose the string in "..." and escape
** any double-quotes or backslash characters contained within the
** string.
*/
static void jsonAppendString(JsonString *p, const char *zIn, u32 N){
u32 i;
if( zIn==0 || ((N+p->nUsed+2 >= p->nAlloc) && jsonGrow(p,N+2)!=0) ) return;
p->zBuf[p->nUsed++] = '"';
for(i=0; i<N; i++){
unsigned char c = ((unsigned const char*)zIn)[i];
if( c=='"' || c=='\\' ){
json_simple_escape:
if( (p->nUsed+N+3-i > p->nAlloc) && jsonGrow(p,N+3-i)!=0 ) return;
p->zBuf[p->nUsed++] = '\\';
}else if( c<=0x1f ){
static const char aSpecial[] = {
0, 0, 0, 0, 0, 0, 0, 0, 'b', 't', 'n', 0, 'f', 'r', 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};
assert( sizeof(aSpecial)==32 );
assert( aSpecial['\b']=='b' );
assert( aSpecial['\f']=='f' );
assert( aSpecial['\n']=='n' );
assert( aSpecial['\r']=='r' );
assert( aSpecial['\t']=='t' );
if( aSpecial[c] ){
c = aSpecial[c];
goto json_simple_escape;
}
if( (p->nUsed+N+7+i > p->nAlloc) && jsonGrow(p,N+7-i)!=0 ) return;
p->zBuf[p->nUsed++] = '\\';
p->zBuf[p->nUsed++] = 'u';
p->zBuf[p->nUsed++] = '0';
p->zBuf[p->nUsed++] = '0';
p->zBuf[p->nUsed++] = '0' + (c>>4);
c = "0123456789abcdef"[c&0xf];
}
p->zBuf[p->nUsed++] = c;
}
p->zBuf[p->nUsed++] = '"';
assert( p->nUsed<p->nAlloc );
}
/*
** Append a function parameter value to the JSON string under
** construction.
*/
static void jsonAppendValue(
JsonString *p, /* Append to this JSON string */
sqlite3_value *pValue /* Value to append */
){
switch( sqlite3_value_type(pValue) ){
case SQLITE_NULL: {
jsonAppendRaw(p, "null", 4);
break;
}
case SQLITE_INTEGER:
case SQLITE_FLOAT: {
const char *z = (const char*)sqlite3_value_text(pValue);
u32 n = (u32)sqlite3_value_bytes(pValue);
jsonAppendRaw(p, z, n);
break;
}
case SQLITE_TEXT: {
const char *z = (const char*)sqlite3_value_text(pValue);
u32 n = (u32)sqlite3_value_bytes(pValue);
if( sqlite3_value_subtype(pValue)==JSON_SUBTYPE ){
jsonAppendRaw(p, z, n);
}else{
jsonAppendString(p, z, n);
}
break;
}
default: {
if( p->bErr==0 ){
sqlite3_result_error(p->pCtx, "JSON cannot hold BLOB values", -1);
p->bErr = 2;
jsonReset(p);
}
break;
}
}
}
/* Make the JSON in p the result of the SQL function.
*/
static void jsonResult(JsonString *p){
if( p->bErr==0 ){
sqlite3_result_text64(p->pCtx, p->zBuf, p->nUsed,
p->bStatic ? SQLITE_TRANSIENT : sqlite3_free,
SQLITE_UTF8);
jsonZero(p);
}
assert( p->bStatic );
}
/**************************************************************************
** Utility routines for dealing with JsonNode and JsonParse objects
**************************************************************************/
/*
** Return the number of consecutive JsonNode slots need to represent
** the parsed JSON at pNode. The minimum answer is 1. For ARRAY and
** OBJECT types, the number might be larger.
**
** Appended elements are not counted. The value returned is the number
** by which the JsonNode counter should increment in order to go to the
** next peer value.
*/
static u32 jsonNodeSize(JsonNode *pNode){
return pNode->eType>=JSON_ARRAY ? pNode->n+1 : 1;
}
/*
** Reclaim all memory allocated by a JsonParse object. But do not
** delete the JsonParse object itself.
*/
static void jsonParseReset(JsonParse *pParse){
sqlite3_free(pParse->aNode);
pParse->aNode = 0;
pParse->nNode = 0;
pParse->nAlloc = 0;
sqlite3_free(pParse->aUp);
pParse->aUp = 0;
}
/*
** Free a JsonParse object that was obtained from sqlite3_malloc().
*/
static void jsonParseFree(JsonParse *pParse){
jsonParseReset(pParse);
sqlite3_free(pParse);
}
/*
** Convert the JsonNode pNode into a pure JSON string and
** append to pOut. Subsubstructure is also included. Return
** the number of JsonNode objects that are encoded.
*/
static void jsonRenderNode(
JsonNode *pNode, /* The node to render */
JsonString *pOut, /* Write JSON here */
sqlite3_value **aReplace /* Replacement values */
){
assert( pNode!=0 );
if( pNode->jnFlags & (JNODE_REPLACE|JNODE_PATCH) ){
if( (pNode->jnFlags & JNODE_REPLACE)!=0 && ALWAYS(aReplace!=0) ){
assert( pNode->eU==4 );
jsonAppendValue(pOut, aReplace[pNode->u.iReplace]);
return;
}
assert( pNode->eU==5 );
pNode = pNode->u.pPatch;
}
switch( pNode->eType ){
default: {
assert( pNode->eType==JSON_NULL );
jsonAppendRaw(pOut, "null", 4);
break;
}
case JSON_TRUE: {
jsonAppendRaw(pOut, "true", 4);
break;
}
case JSON_FALSE: {
jsonAppendRaw(pOut, "false", 5);
break;
}
case JSON_STRING: {
if( pNode->jnFlags & JNODE_RAW ){
assert( pNode->eU==1 );
jsonAppendString(pOut, pNode->u.zJContent, pNode->n);
break;
}
/* no break */ deliberate_fall_through
}
case JSON_REAL:
case JSON_INT: {
assert( pNode->eU==1 );
jsonAppendRaw(pOut, pNode->u.zJContent, pNode->n);
break;
}
case JSON_ARRAY: {
u32 j = 1;
jsonAppendChar(pOut, '[');
for(;;){
while( j<=pNode->n ){
if( (pNode[j].jnFlags & JNODE_REMOVE)==0 ){
jsonAppendSeparator(pOut);
jsonRenderNode(&pNode[j], pOut, aReplace);
}
j += jsonNodeSize(&pNode[j]);
}
if( (pNode->jnFlags & JNODE_APPEND)==0 ) break;
assert( pNode->eU==2 );
pNode = &pNode[pNode->u.iAppend];
j = 1;
}
jsonAppendChar(pOut, ']');
break;
}
case JSON_OBJECT: {
u32 j = 1;
jsonAppendChar(pOut, '{');
for(;;){
while( j<=pNode->n ){
if( (pNode[j+1].jnFlags & JNODE_REMOVE)==0 ){
jsonAppendSeparator(pOut);
jsonRenderNode(&pNode[j], pOut, aReplace);
jsonAppendChar(pOut, ':');
jsonRenderNode(&pNode[j+1], pOut, aReplace);
}
j += 1 + jsonNodeSize(&pNode[j+1]);
}
if( (pNode->jnFlags & JNODE_APPEND)==0 ) break;
assert( pNode->eU==2 );
pNode = &pNode[pNode->u.iAppend];
j = 1;
}
jsonAppendChar(pOut, '}');
break;
}
}
}
/*
** Return a JsonNode and all its descendents as a JSON string.
*/
static void jsonReturnJson(
JsonNode *pNode, /* Node to return */
sqlite3_context *pCtx, /* Return value for this function */
sqlite3_value **aReplace /* Array of replacement values */
){
JsonString s;
jsonInit(&s, pCtx);
jsonRenderNode(pNode, &s, aReplace);
jsonResult(&s);
sqlite3_result_subtype(pCtx, JSON_SUBTYPE);
}
/*
** Translate a single byte of Hex into an integer.
** This routine only works if h really is a valid hexadecimal
** character: 0..9a..fA..F
*/
static u8 jsonHexToInt(int h){
assert( (h>='0' && h<='9') || (h>='a' && h<='f') || (h>='A' && h<='F') );
#ifdef SQLITE_EBCDIC
h += 9*(1&~(h>>4));
#else
h += 9*(1&(h>>6));
#endif
return (u8)(h & 0xf);
}
/*
** Convert a 4-byte hex string into an integer
*/
static u32 jsonHexToInt4(const char *z){
u32 v;
assert( safe_isxdigit(z[0]) );
assert( safe_isxdigit(z[1]) );
assert( safe_isxdigit(z[2]) );
assert( safe_isxdigit(z[3]) );
v = (jsonHexToInt(z[0])<<12)
+ (jsonHexToInt(z[1])<<8)
+ (jsonHexToInt(z[2])<<4)
+ jsonHexToInt(z[3]);
return v;
}
/*
** Make the JsonNode the return value of the function.
*/
static void jsonReturn(
JsonNode *pNode, /* Node to return */
sqlite3_context *pCtx, /* Return value for this function */
sqlite3_value **aReplace /* Array of replacement values */
){
switch( pNode->eType ){
default: {
assert( pNode->eType==JSON_NULL );
sqlite3_result_null(pCtx);
break;
}
case JSON_TRUE: {
sqlite3_result_int(pCtx, 1);
break;
}
case JSON_FALSE: {
sqlite3_result_int(pCtx, 0);
break;
}
case JSON_INT: {
sqlite3_int64 i = 0;
const char *z;
assert( pNode->eU==1 );
z = pNode->u.zJContent;
if( z[0]=='-' ){ z++; }
while( z[0]>='0' && z[0]<='9' ){
unsigned v = *(z++) - '0';
if( i>=LARGEST_INT64/10 ){
if( i>LARGEST_INT64/10 ) goto int_as_real;
if( z[0]>='0' && z[0]<='9' ) goto int_as_real;
if( v==9 ) goto int_as_real;
if( v==8 ){
if( pNode->u.zJContent[0]=='-' ){
sqlite3_result_int64(pCtx, SMALLEST_INT64);
goto int_done;
}else{
goto int_as_real;
}
}
}
i = i*10 + v;
}
if( pNode->u.zJContent[0]=='-' ){ i = -i; }
sqlite3_result_int64(pCtx, i);
int_done:
break;
int_as_real: ; /* no break */ deliberate_fall_through
}
case JSON_REAL: {
double r;
#ifdef SQLITE_AMALGAMATION
const char *z;
assert( pNode->eU==1 );
z = pNode->u.zJContent;
sqlite3AtoF(z, &r, sqlite3Strlen30(z), SQLITE_UTF8);
#else
assert( pNode->eU==1 );
r = strtod(pNode->u.zJContent, 0);
#endif
sqlite3_result_double(pCtx, r);
break;
}
case JSON_STRING: {
#if 0 /* Never happens because JNODE_RAW is only set by json_set(),
** json_insert() and json_replace() and those routines do not
** call jsonReturn() */
if( pNode->jnFlags & JNODE_RAW ){
assert( pNode->eU==1 );
sqlite3_result_text(pCtx, pNode->u.zJContent, pNode->n,
SQLITE_TRANSIENT);
}else
#endif
assert( (pNode->jnFlags & JNODE_RAW)==0 );
if( (pNode->jnFlags & JNODE_ESCAPE)==0 ){
/* JSON formatted without any backslash-escapes */
assert( pNode->eU==1 );
sqlite3_result_text(pCtx, pNode->u.zJContent+1, pNode->n-2,
SQLITE_TRANSIENT);
}else{
/* Translate JSON formatted string into raw text */
u32 i;
u32 n = pNode->n;
const char *z;
char *zOut;
u32 j;
assert( pNode->eU==1 );
z = pNode->u.zJContent;
zOut = sqlite3_malloc( n+1 );
if( zOut==0 ){
sqlite3_result_error_nomem(pCtx);
break;
}
for(i=1, j=0; i<n-1; i++){
char c = z[i];
if( c!='\\' ){
zOut[j++] = c;
}else{
c = z[++i];
if( c=='u' ){
u32 v = jsonHexToInt4(z+i+1);
i += 4;
if( v==0 ) break;
if( v<=0x7f ){
zOut[j++] = (char)v;
}else if( v<=0x7ff ){
zOut[j++] = (char)(0xc0 | (v>>6));
zOut[j++] = 0x80 | (v&0x3f);
}else{
u32 vlo;
if( (v&0xfc00)==0xd800
&& i<n-6
&& z[i+1]=='\\'
&& z[i+2]=='u'
&& ((vlo = jsonHexToInt4(z+i+3))&0xfc00)==0xdc00
){
/* We have a surrogate pair */
v = ((v&0x3ff)<<10) + (vlo&0x3ff) + 0x10000;
i += 6;
zOut[j++] = 0xf0 | (v>>18);
zOut[j++] = 0x80 | ((v>>12)&0x3f);
zOut[j++] = 0x80 | ((v>>6)&0x3f);
zOut[j++] = 0x80 | (v&0x3f);
}else{
zOut[j++] = 0xe0 | (v>>12);
zOut[j++] = 0x80 | ((v>>6)&0x3f);
zOut[j++] = 0x80 | (v&0x3f);
}
}
}else{
if( c=='b' ){
c = '\b';
}else if( c=='f' ){
c = '\f';
}else if( c=='n' ){
c = '\n';
}else if( c=='r' ){
c = '\r';
}else if( c=='t' ){
c = '\t';
}
zOut[j++] = c;
}
}
}
zOut[j] = 0;
sqlite3_result_text(pCtx, zOut, j, sqlite3_free);
}
break;
}
case JSON_ARRAY:
case JSON_OBJECT: {
jsonReturnJson(pNode, pCtx, aReplace);
break;
}
}
}
/* Forward reference */
static int jsonParseAddNode(JsonParse*,u32,u32,const char*);
/*
** A macro to hint to the compiler that a function should not be
** inlined.
*/
#if defined(__GNUC__)
# define JSON_NOINLINE __attribute__((noinline))
#elif defined(_MSC_VER) && _MSC_VER>=1310
# define JSON_NOINLINE __declspec(noinline)
#else
# define JSON_NOINLINE
#endif
static JSON_NOINLINE int jsonParseAddNodeExpand(
JsonParse *pParse, /* Append the node to this object */
u32 eType, /* Node type */
u32 n, /* Content size or sub-node count */
const char *zContent /* Content */
){
u32 nNew;
JsonNode *pNew;
assert( pParse->nNode>=pParse->nAlloc );
if( pParse->oom ) return -1;
nNew = pParse->nAlloc*2 + 10;
pNew = sqlite3_realloc64(pParse->aNode, sizeof(JsonNode)*nNew);
if( pNew==0 ){
pParse->oom = 1;
return -1;
}
pParse->nAlloc = nNew;
pParse->aNode = pNew;
assert( pParse->nNode<pParse->nAlloc );
return jsonParseAddNode(pParse, eType, n, zContent);
}
/*
** Create a new JsonNode instance based on the arguments and append that
** instance to the JsonParse. Return the index in pParse->aNode[] of the
** new node, or -1 if a memory allocation fails.
*/
static int jsonParseAddNode(
JsonParse *pParse, /* Append the node to this object */
u32 eType, /* Node type */
u32 n, /* Content size or sub-node count */
const char *zContent /* Content */
){
JsonNode *p;
if( pParse->aNode==0 || pParse->nNode>=pParse->nAlloc ){
return jsonParseAddNodeExpand(pParse, eType, n, zContent);
}
p = &pParse->aNode[pParse->nNode];
p->eType = (u8)eType;
p->jnFlags = 0;
VVA( p->eU = zContent ? 1 : 0 );
p->n = n;
p->u.zJContent = zContent;
return pParse->nNode++;
}
/*
** Return true if z[] begins with 4 (or more) hexadecimal digits
*/
static int jsonIs4Hex(const char *z){
int i;
for(i=0; i<4; i++) if( !safe_isxdigit(z[i]) ) return 0;
return 1;
}
/*
** Parse a single JSON value which begins at pParse->zJson[i]. Return the
** index of the first character past the end of the value parsed.
**
** Return negative for a syntax error. Special cases: return -2 if the
** first non-whitespace character is '}' and return -3 if the first
** non-whitespace character is ']'.
*/
static int jsonParseValue(JsonParse *pParse, u32 i){
char c;
u32 j;
int iThis;
int x;
JsonNode *pNode;
const char *z = pParse->zJson;
while( safe_isspace(z[i]) ){ i++; }
if( (c = z[i])=='{' ){
/* Parse object */
iThis = jsonParseAddNode(pParse, JSON_OBJECT, 0, 0);
if( iThis<0 ) return -1;
for(j=i+1;;j++){
while( safe_isspace(z[j]) ){ j++; }
if( ++pParse->iDepth > JSON_MAX_DEPTH ) return -1;
x = jsonParseValue(pParse, j);
if( x<0 ){
pParse->iDepth--;
if( x==(-2) && pParse->nNode==(u32)iThis+1 ) return j+1;
return -1;
}
if( pParse->oom ) return -1;
pNode = &pParse->aNode[pParse->nNode-1];
if( pNode->eType!=JSON_STRING ) return -1;
pNode->jnFlags |= JNODE_LABEL;
j = x;
while( safe_isspace(z[j]) ){ j++; }
if( z[j]!=':' ) return -1;
j++;
x = jsonParseValue(pParse, j);
pParse->iDepth--;
if( x<0 ) return -1;
j = x;
while( safe_isspace(z[j]) ){ j++; }
c = z[j];
if( c==',' ) continue;
if( c!='}' ) return -1;
break;
}
pParse->aNode[iThis].n = pParse->nNode - (u32)iThis - 1;
return j+1;
}else if( c=='[' ){
/* Parse array */
iThis = jsonParseAddNode(pParse, JSON_ARRAY, 0, 0);
if( iThis<0 ) return -1;
memset(&pParse->aNode[iThis].u, 0, sizeof(pParse->aNode[iThis].u));
for(j=i+1;;j++){
while( safe_isspace(z[j]) ){ j++; }
if( ++pParse->iDepth > JSON_MAX_DEPTH ) return -1;
x = jsonParseValue(pParse, j);
pParse->iDepth--;
if( x<0 ){
if( x==(-3) && pParse->nNode==(u32)iThis+1 ) return j+1;
return -1;
}
j = x;
while( safe_isspace(z[j]) ){ j++; }
c = z[j];
if( c==',' ) continue;
if( c!=']' ) return -1;
break;
}
pParse->aNode[iThis].n = pParse->nNode - (u32)iThis - 1;
return j+1;
}else if( c=='"' ){
/* Parse string */
u8 jnFlags = 0;
j = i+1;
for(;;){
c = z[j];
if( (c & ~0x1f)==0 ){
/* Control characters are not allowed in strings */
return -1;
}
if( c=='\\' ){
c = z[++j];
if( c=='"' || c=='\\' || c=='/' || c=='b' || c=='f'
|| c=='n' || c=='r' || c=='t'
|| (c=='u' && jsonIs4Hex(z+j+1)) ){
jnFlags = JNODE_ESCAPE;
}else{
return -1;
}
}else if( c=='"' ){
break;
}
j++;
}
jsonParseAddNode(pParse, JSON_STRING, j+1-i, &z[i]);
if( !pParse->oom ) pParse->aNode[pParse->nNode-1].jnFlags = jnFlags;
return j+1;
}else if( c=='n'
&& strncmp(z+i,"null",4)==0
&& !safe_isalnum(z[i+4]) ){
jsonParseAddNode(pParse, JSON_NULL, 0, 0);
return i+4;
}else if( c=='t'
&& strncmp(z+i,"true",4)==0
&& !safe_isalnum(z[i+4]) ){
jsonParseAddNode(pParse, JSON_TRUE, 0, 0);
return i+4;
}else if( c=='f'
&& strncmp(z+i,"false",5)==0
&& !safe_isalnum(z[i+5]) ){
jsonParseAddNode(pParse, JSON_FALSE, 0, 0);
return i+5;
}else if( c=='-' || (c>='0' && c<='9') ){
/* Parse number */
u8 seenDP = 0;
u8 seenE = 0;
assert( '-' < '0' );
if( c<='0' ){
j = c=='-' ? i+1 : i;
if( z[j]=='0' && z[j+1]>='0' && z[j+1]<='9' ) return -1;
}
j = i+1;
for(;; j++){
c = z[j];
if( c>='0' && c<='9' ) continue;
if( c=='.' ){
if( z[j-1]=='-' ) return -1;
if( seenDP ) return -1;
seenDP = 1;
continue;
}
if( c=='e' || c=='E' ){
if( z[j-1]<'0' ) return -1;
if( seenE ) return -1;
seenDP = seenE = 1;
c = z[j+1];
if( c=='+' || c=='-' ){
j++;
c = z[j+1];
}
if( c<'0' || c>'9' ) return -1;
continue;
}
break;
}
if( z[j-1]<'0' ) return -1;
jsonParseAddNode(pParse, seenDP ? JSON_REAL : JSON_INT,
j - i, &z[i]);
return j;
}else if( c=='}' ){
return -2; /* End of {...} */
}else if( c==']' ){
return -3; /* End of [...] */
}else if( c==0 ){
return 0; /* End of file */
}else{
return -1; /* Syntax error */
}
}
/*
** Parse a complete JSON string. Return 0 on success or non-zero if there
** are any errors. If an error occurs, free all memory associated with
** pParse.
**
** pParse is uninitialized when this routine is called.
*/
static int jsonParse(
JsonParse *pParse, /* Initialize and fill this JsonParse object */
sqlite3_context *pCtx, /* Report errors here */
const char *zJson /* Input JSON text to be parsed */
){
int i;
memset(pParse, 0, sizeof(*pParse));
if( zJson==0 ) return 1;
pParse->zJson = zJson;
i = jsonParseValue(pParse, 0);
if( pParse->oom ) i = -1;
if( i>0 ){
assert( pParse->iDepth==0 );
while( safe_isspace(zJson[i]) ) i++;
if( zJson[i] ) i = -1;
}
if( i<=0 ){
if( pCtx!=0 ){
if( pParse->oom ){
sqlite3_result_error_nomem(pCtx);
}else{
sqlite3_result_error(pCtx, "malformed JSON", -1);
}
}
jsonParseReset(pParse);
return 1;
}
return 0;
}
/* Mark node i of pParse as being a child of iParent. Call recursively
** to fill in all the descendants of node i.
*/
static void jsonParseFillInParentage(JsonParse *pParse, u32 i, u32 iParent){
JsonNode *pNode = &pParse->aNode[i];
u32 j;
pParse->aUp[i] = iParent;
switch( pNode->eType ){
case JSON_ARRAY: {
for(j=1; j<=pNode->n; j += jsonNodeSize(pNode+j)){
jsonParseFillInParentage(pParse, i+j, i);
}
break;
}
case JSON_OBJECT: {
for(j=1; j<=pNode->n; j += jsonNodeSize(pNode+j+1)+1){
pParse->aUp[i+j] = i;
jsonParseFillInParentage(pParse, i+j+1, i);
}
break;
}
default: {
break;
}
}
}
/*
** Compute the parentage of all nodes in a completed parse.
*/
static int jsonParseFindParents(JsonParse *pParse){
u32 *aUp;
assert( pParse->aUp==0 );
aUp = pParse->aUp = sqlite3_malloc64( sizeof(u32)*pParse->nNode );
if( aUp==0 ){
pParse->oom = 1;
return SQLITE_NOMEM;
}
jsonParseFillInParentage(pParse, 0, 0);
return SQLITE_OK;
}
/*
** Magic number used for the JSON parse cache in sqlite3_get_auxdata()
*/
#define JSON_CACHE_ID (-429938) /* First cache entry */
#define JSON_CACHE_SZ 4 /* Max number of cache entries */
/*
** Obtain a complete parse of the JSON found in the first argument
** of the argv array. Use the sqlite3_get_auxdata() cache for this
** parse if it is available. If the cache is not available or if it
** is no longer valid, parse the JSON again and return the new parse,
** and also register the new parse so that it will be available for
** future sqlite3_get_auxdata() calls.
*/
static JsonParse *jsonParseCached(
sqlite3_context *pCtx,
sqlite3_value **argv,
sqlite3_context *pErrCtx
){
const char *zJson = (const char*)sqlite3_value_text(argv[0]);
int nJson = sqlite3_value_bytes(argv[0]);
JsonParse *p;
JsonParse *pMatch = 0;
int iKey;
int iMinKey = 0;
u32 iMinHold = 0xffffffff;
u32 iMaxHold = 0;
if( zJson==0 ) return 0;
for(iKey=0; iKey<JSON_CACHE_SZ; iKey++){
p = (JsonParse*)sqlite3_get_auxdata(pCtx, JSON_CACHE_ID+iKey);
if( p==0 ){
iMinKey = iKey;
break;
}
if( pMatch==0
&& p->nJson==nJson
&& memcmp(p->zJson,zJson,nJson)==0
){
p->nErr = 0;
pMatch = p;
}else if( p->iHold<iMinHold ){
iMinHold = p->iHold;
iMinKey = iKey;
}
if( p->iHold>iMaxHold ){
iMaxHold = p->iHold;
}
}
if( pMatch ){
pMatch->nErr = 0;
pMatch->iHold = iMaxHold+1;
return pMatch;
}
p = sqlite3_malloc64( sizeof(*p) + nJson + 1 );
if( p==0 ){
sqlite3_result_error_nomem(pCtx);
return 0;
}
memset(p, 0, sizeof(*p));
p->zJson = (char*)&p[1];
memcpy((char*)p->zJson, zJson, nJson+1);
if( jsonParse(p, pErrCtx, p->zJson) ){
sqlite3_free(p);
return 0;
}
p->nJson = nJson;
p->iHold = iMaxHold+1;
sqlite3_set_auxdata(pCtx, JSON_CACHE_ID+iMinKey, p,
(void(*)(void*))jsonParseFree);
return (JsonParse*)sqlite3_get_auxdata(pCtx, JSON_CACHE_ID+iMinKey);
}
/*
** Compare the OBJECT label at pNode against zKey,nKey. Return true on
** a match.
*/
static int jsonLabelCompare(JsonNode *pNode, const char *zKey, u32 nKey){
assert( pNode->eU==1 );
if( pNode->jnFlags & JNODE_RAW ){
if( pNode->n!=nKey ) return 0;
return strncmp(pNode->u.zJContent, zKey, nKey)==0;
}else{
if( pNode->n!=nKey+2 ) return 0;
return strncmp(pNode->u.zJContent+1, zKey, nKey)==0;
}
}
/* forward declaration */
static JsonNode *jsonLookupAppend(JsonParse*,const char*,int*,const char**);
/*
** Search along zPath to find the node specified. Return a pointer
** to that node, or NULL if zPath is malformed or if there is no such
** node.
**
** If pApnd!=0, then try to append new nodes to complete zPath if it is
** possible to do so and if no existing node corresponds to zPath. If
** new nodes are appended *pApnd is set to 1.
*/
static JsonNode *jsonLookupStep(
JsonParse *pParse, /* The JSON to search */
u32 iRoot, /* Begin the search at this node */
const char *zPath, /* The path to search */
int *pApnd, /* Append nodes to complete path if not NULL */
const char **pzErr /* Make *pzErr point to any syntax error in zPath */
){
u32 i, j, nKey;
const char *zKey;
JsonNode *pRoot = &pParse->aNode[iRoot];
if( zPath[0]==0 ) return pRoot;
if( pRoot->jnFlags & JNODE_REPLACE ) return 0;
if( zPath[0]=='.' ){
if( pRoot->eType!=JSON_OBJECT ) return 0;
zPath++;
if( zPath[0]=='"' ){
zKey = zPath + 1;
for(i=1; zPath[i] && zPath[i]!='"'; i++){}
nKey = i-1;
if( zPath[i] ){
i++;
}else{
*pzErr = zPath;
return 0;
}
}else{
zKey = zPath;
for(i=0; zPath[i] && zPath[i]!='.' && zPath[i]!='['; i++){}
nKey = i;
}
if( nKey==0 ){
*pzErr = zPath;
return 0;
}
j = 1;
for(;;){
while( j<=pRoot->n ){
if( jsonLabelCompare(pRoot+j, zKey, nKey) ){
return jsonLookupStep(pParse, iRoot+j+1, &zPath[i], pApnd, pzErr);
}
j++;
j += jsonNodeSize(&pRoot[j]);
}
if( (pRoot->jnFlags & JNODE_APPEND)==0 ) break;
assert( pRoot->eU==2 );
iRoot += pRoot->u.iAppend;
pRoot = &pParse->aNode[iRoot];
j = 1;
}
if( pApnd ){
u32 iStart, iLabel;
JsonNode *pNode;
iStart = jsonParseAddNode(pParse, JSON_OBJECT, 2, 0);
iLabel = jsonParseAddNode(pParse, JSON_STRING, nKey, zKey);
zPath += i;
pNode = jsonLookupAppend(pParse, zPath, pApnd, pzErr);
if( pParse->oom ) return 0;
if( pNode ){
pRoot = &pParse->aNode[iRoot];
assert( pRoot->eU==0 );
pRoot->u.iAppend = iStart - iRoot;
pRoot->jnFlags |= JNODE_APPEND;
VVA( pRoot->eU = 2 );
pParse->aNode[iLabel].jnFlags |= JNODE_RAW;
}
return pNode;
}
}else if( zPath[0]=='[' ){
i = 0;
j = 1;
while( safe_isdigit(zPath[j]) ){
i = i*10 + zPath[j] - '0';
j++;
}
if( j<2 || zPath[j]!=']' ){
if( zPath[1]=='#' ){
JsonNode *pBase = pRoot;
int iBase = iRoot;
if( pRoot->eType!=JSON_ARRAY ) return 0;
for(;;){
while( j<=pBase->n ){
if( (pBase[j].jnFlags & JNODE_REMOVE)==0 ) i++;
j += jsonNodeSize(&pBase[j]);
}
if( (pBase->jnFlags & JNODE_APPEND)==0 ) break;
assert( pBase->eU==2 );
iBase += pBase->u.iAppend;
pBase = &pParse->aNode[iBase];
j = 1;
}
j = 2;
if( zPath[2]=='-' && safe_isdigit(zPath[3]) ){
unsigned int x = 0;
j = 3;
do{
x = x*10 + zPath[j] - '0';
j++;
}while( safe_isdigit(zPath[j]) );
if( x>i ) return 0;
i -= x;
}
if( zPath[j]!=']' ){
*pzErr = zPath;
return 0;
}
}else{
*pzErr = zPath;
return 0;
}
}
if( pRoot->eType!=JSON_ARRAY ) return 0;
zPath += j + 1;
j = 1;
for(;;){
while( j<=pRoot->n && (i>0 || (pRoot[j].jnFlags & JNODE_REMOVE)!=0) ){
if( (pRoot[j].jnFlags & JNODE_REMOVE)==0 ) i--;
j += jsonNodeSize(&pRoot[j]);
}
if( (pRoot->jnFlags & JNODE_APPEND)==0 ) break;
assert( pRoot->eU==2 );
iRoot += pRoot->u.iAppend;
pRoot = &pParse->aNode[iRoot];
j = 1;
}
if( j<=pRoot->n ){
return jsonLookupStep(pParse, iRoot+j, zPath, pApnd, pzErr);
}
if( i==0 && pApnd ){
u32 iStart;
JsonNode *pNode;
iStart = jsonParseAddNode(pParse, JSON_ARRAY, 1, 0);
pNode = jsonLookupAppend(pParse, zPath, pApnd, pzErr);
if( pParse->oom ) return 0;
if( pNode ){
pRoot = &pParse->aNode[iRoot];
assert( pRoot->eU==0 );
pRoot->u.iAppend = iStart - iRoot;
pRoot->jnFlags |= JNODE_APPEND;
VVA( pRoot->eU = 2 );
}
return pNode;
}
}else{
*pzErr = zPath;
}
return 0;
}
/*
** Append content to pParse that will complete zPath. Return a pointer
** to the inserted node, or return NULL if the append fails.
*/
static JsonNode *jsonLookupAppend(
JsonParse *pParse, /* Append content to the JSON parse */
const char *zPath, /* Description of content to append */
int *pApnd, /* Set this flag to 1 */
const char **pzErr /* Make this point to any syntax error */
){
*pApnd = 1;
if( zPath[0]==0 ){
jsonParseAddNode(pParse, JSON_NULL, 0, 0);
return pParse->oom ? 0 : &pParse->aNode[pParse->nNode-1];
}
if( zPath[0]=='.' ){
jsonParseAddNode(pParse, JSON_OBJECT, 0, 0);
}else if( strncmp(zPath,"[0]",3)==0 ){
jsonParseAddNode(pParse, JSON_ARRAY, 0, 0);
}else{
return 0;
}
if( pParse->oom ) return 0;
return jsonLookupStep(pParse, pParse->nNode-1, zPath, pApnd, pzErr);
}
/*
** Return the text of a syntax error message on a JSON path. Space is
** obtained from sqlite3_malloc().
*/
static char *jsonPathSyntaxError(const char *zErr){
return sqlite3_mprintf("JSON path error near '%q'", zErr);
}
/*
** Do a node lookup using zPath. Return a pointer to the node on success.
** Return NULL if not found or if there is an error.
**
** On an error, write an error message into pCtx and increment the
** pParse->nErr counter.
**
** If pApnd!=NULL then try to append missing nodes and set *pApnd = 1 if
** nodes are appended.
*/
static JsonNode *jsonLookup(
JsonParse *pParse, /* The JSON to search */
const char *zPath, /* The path to search */
int *pApnd, /* Append nodes to complete path if not NULL */
sqlite3_context *pCtx /* Report errors here, if not NULL */
){
const char *zErr = 0;
JsonNode *pNode = 0;
char *zMsg;
if( zPath==0 ) return 0;
if( zPath[0]!='$' ){
zErr = zPath;
goto lookup_err;
}
zPath++;
pNode = jsonLookupStep(pParse, 0, zPath, pApnd, &zErr);
if( zErr==0 ) return pNode;
lookup_err:
pParse->nErr++;
assert( zErr!=0 && pCtx!=0 );
zMsg = jsonPathSyntaxError(zErr);
if( zMsg ){
sqlite3_result_error(pCtx, zMsg, -1);
sqlite3_free(zMsg);
}else{
sqlite3_result_error_nomem(pCtx);
}
return 0;
}
/*
** Report the wrong number of arguments for json_insert(), json_replace()
** or json_set().
*/
static void jsonWrongNumArgs(
sqlite3_context *pCtx,
const char *zFuncName
){
char *zMsg = sqlite3_mprintf("json_%s() needs an odd number of arguments",
zFuncName);
sqlite3_result_error(pCtx, zMsg, -1);
sqlite3_free(zMsg);
}
/*
** Mark all NULL entries in the Object passed in as JNODE_REMOVE.
*/
static void jsonRemoveAllNulls(JsonNode *pNode){
int i, n;
assert( pNode->eType==JSON_OBJECT );
n = pNode->n;
for(i=2; i<=n; i += jsonNodeSize(&pNode[i])+1){
switch( pNode[i].eType ){
case JSON_NULL:
pNode[i].jnFlags |= JNODE_REMOVE;
break;
case JSON_OBJECT:
jsonRemoveAllNulls(&pNode[i]);
break;
}
}
}
/****************************************************************************
** SQL functions used for testing and debugging
****************************************************************************/
#ifdef SQLITE_DEBUG
/*
** The json_parse(JSON) function returns a string which describes
** a parse of the JSON provided. Or it returns NULL if JSON is not
** well-formed.
*/
static void jsonParseFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonString s; /* Output string - not real JSON */
JsonParse x; /* The parse */
u32 i;
assert( argc==1 );
if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return;
jsonParseFindParents(&x);
jsonInit(&s, ctx);
for(i=0; i<x.nNode; i++){
const char *zType;
if( x.aNode[i].jnFlags & JNODE_LABEL ){
assert( x.aNode[i].eType==JSON_STRING );
zType = "label";
}else{
zType = jsonType[x.aNode[i].eType];
}
jsonPrintf(100, &s,"node %3u: %7s n=%-4d up=%-4d",
i, zType, x.aNode[i].n, x.aUp[i]);
assert( x.aNode[i].eU==0 || x.aNode[i].eU==1 );
if( x.aNode[i].u.zJContent!=0 ){
assert( x.aNode[i].eU==1 );
jsonAppendRaw(&s, " ", 1);
jsonAppendRaw(&s, x.aNode[i].u.zJContent, x.aNode[i].n);
}else{
assert( x.aNode[i].eU==0 );
}
jsonAppendRaw(&s, "\n", 1);
}
jsonParseReset(&x);
jsonResult(&s);
}
/*
** The json_test1(JSON) function return true (1) if the input is JSON
** text generated by another json function. It returns (0) if the input
** is not known to be JSON.
*/
static void jsonTest1Func(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
UNUSED_PARAM(argc);
sqlite3_result_int(ctx, sqlite3_value_subtype(argv[0])==JSON_SUBTYPE);
}
#endif /* SQLITE_DEBUG */
/****************************************************************************
** Scalar SQL function implementations
****************************************************************************/
/*
** Implementation of the json_QUOTE(VALUE) function. Return a JSON value
** corresponding to the SQL value input. Mostly this means putting
** double-quotes around strings and returning the unquoted string "null"
** when given a NULL input.
*/
static void jsonQuoteFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonString jx;
UNUSED_PARAM(argc);
jsonInit(&jx, ctx);
jsonAppendValue(&jx, argv[0]);
jsonResult(&jx);
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
/*
** Implementation of the json_array(VALUE,...) function. Return a JSON
** array that contains all values given in arguments. Or if any argument
** is a BLOB, throw an error.
*/
static void jsonArrayFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
int i;
JsonString jx;
jsonInit(&jx, ctx);
jsonAppendChar(&jx, '[');
for(i=0; i<argc; i++){
jsonAppendSeparator(&jx);
jsonAppendValue(&jx, argv[i]);
}
jsonAppendChar(&jx, ']');
jsonResult(&jx);
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
/*
** json_array_length(JSON)
** json_array_length(JSON, PATH)
**
** Return the number of elements in the top-level JSON array.
** Return 0 if the input is not a well-formed JSON array.
*/
static void jsonArrayLengthFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse *p; /* The parse */
sqlite3_int64 n = 0;
u32 i;
JsonNode *pNode;
p = jsonParseCached(ctx, argv, ctx);
if( p==0 ) return;
assert( p->nNode );
if( argc==2 ){
const char *zPath = (const char*)sqlite3_value_text(argv[1]);
pNode = jsonLookup(p, zPath, 0, ctx);
}else{
pNode = p->aNode;
}
if( pNode==0 ){
return;
}
if( pNode->eType==JSON_ARRAY ){
assert( (pNode->jnFlags & JNODE_APPEND)==0 );
for(i=1; i<=pNode->n; n++){
i += jsonNodeSize(&pNode[i]);
}
}
sqlite3_result_int64(ctx, n);
}
/*
** Bit values for the flags passed into jsonExtractFunc() or
** jsonSetFunc() via the user-data value.
*/
#define JSON_NULLERR 0x01 /* Return NULL if input is not JSON */
#define JSON_ABPATH 0x02 /* Allow abbreviated JSON path specs */
#define JSON_ISSET 0x04 /* json_set(), not json_insert() */
/*
** json_extract(JSON, PATH, ...)
** json_nextract(JSON, PATH, ...)
** "->"(JSON,PATH)
** "->>"(JSON,PATH)
**
** Return the element described by PATH. Return NULL if that PATH element
** is not found. For leaf nodes of the JSON, the value returned is a pure
** SQL value. In other words, quotes have been removed from strings.
**
** If there are multiple PATHs, then the value returned is a JSON array
** with one entry in the array for each PATH term.
**
** Throw an error if any PATH is malformed.
**
** If JSON is not well-formed JSON then:
**
** (1) raise an error if the JSON_NULLERR flag is not set.
**
** (2) Otherwise (if the JSON_NULLERR flags is set and) if there
** is a single PATH argument with the value '$', simply quote
** the JSON input as if by json_quote(). In other words, treat
** the JSON input as a string and convert it into a valid JSON
** string.
**
** (3) Otherwise (if JSON_NULLERR is set and the PATH is not '$')
** return NULL
**
** If the JSON_ABPATH flag is set and there is only a single PATH, then
** allow abbreviated PATH specs that omit the leading "$".
*/
static void jsonExtractFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse *p; /* The parse */
JsonNode *pNode;
const char *zPath;
int flags = *(int*)sqlite3_user_data(ctx);
JsonString jx;
if( argc<2 ) return;
p = jsonParseCached(ctx, argv, (flags & JSON_NULLERR)!=0 ? 0 : ctx);
if( p==0 ){
/* If the form is "json_nextract(IN,'$')" and IN is not well-formed JSON,
** then return IN as a quoted JSON string. */
if( (flags & JSON_NULLERR)!=0
&& argc==2
&& (zPath = (const char*)sqlite3_value_text(argv[1]))!=0
&& zPath[0]=='$' && zPath[1]==0
){
jsonQuoteFunc(ctx, argc, argv);
}
return;
}
if( argc==2 ){
/* With a single PATH argument, the return is the unquoted SQL value */
zPath = (const char*)sqlite3_value_text(argv[1]);
if( zPath && zPath[0]!='$' && zPath[0]!=0 && (flags & JSON_ABPATH)!=0 ){
/* The -> and ->> operators accept abbreviated PATH arguments. This
** is mostly for compatibility with PostgreSQL, but also for convenience.
**
** NUMBER ==> $[NUMBER] // PG compatible
** LABEL ==> $.LABEL // PG compatible
** [NUMBER] ==> $[NUMBER] // Not PG. Purely for convenience
*/
jsonInit(&jx, ctx);
if( safe_isdigit(zPath[0]) ){
jsonAppendRaw(&jx, "$[", 2);
jsonAppendRaw(&jx, zPath, (int)strlen(zPath));
jsonAppendRaw(&jx, "]", 2);
}else{
jsonAppendRaw(&jx, "$.", 1 + (zPath[0]!='['));
jsonAppendRaw(&jx, zPath, (int)strlen(zPath));
jsonAppendChar(&jx, 0);
}
pNode = jx.bErr ? 0 : jsonLookup(p, jx.zBuf, 0, ctx);
jsonReset(&jx);
}else{
pNode = jsonLookup(p, zPath, 0, ctx);
}
if( p->nErr ) return;
if( pNode ) jsonReturn(pNode, ctx, 0);
}else{
/* Two or more PATH arguments results in a JSON array with each
** element of the array being the value selected by one of the PATHs */
int i;
jsonInit(&jx, ctx);
jsonAppendChar(&jx, '[');
for(i=1; i<argc; i++){
zPath = (const char*)sqlite3_value_text(argv[i]);
pNode = jsonLookup(p, zPath, 0, ctx);
if( p->nErr ) break;
jsonAppendSeparator(&jx);
if( pNode ){
jsonRenderNode(pNode, &jx, 0);
}else{
jsonAppendRaw(&jx, "null", 4);
}
}
if( i==argc ){
jsonAppendChar(&jx, ']');
jsonResult(&jx);
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
jsonReset(&jx);
}
}
/* This is the RFC 7396 MergePatch algorithm.
*/
static JsonNode *jsonMergePatch(
JsonParse *pParse, /* The JSON parser that contains the TARGET */
u32 iTarget, /* Node of the TARGET in pParse */
JsonNode *pPatch /* The PATCH */
){
u32 i, j;
u32 iRoot;
JsonNode *pTarget;
if( pPatch->eType!=JSON_OBJECT ){
return pPatch;
}
assert( iTarget>=0 && iTarget<pParse->nNode );
pTarget = &pParse->aNode[iTarget];
assert( (pPatch->jnFlags & JNODE_APPEND)==0 );
if( pTarget->eType!=JSON_OBJECT ){
jsonRemoveAllNulls(pPatch);
return pPatch;
}
iRoot = iTarget;
for(i=1; i<pPatch->n; i += jsonNodeSize(&pPatch[i+1])+1){
u32 nKey;
const char *zKey;
assert( pPatch[i].eType==JSON_STRING );
assert( pPatch[i].jnFlags & JNODE_LABEL );
assert( pPatch[i].eU==1 );
nKey = pPatch[i].n;
zKey = pPatch[i].u.zJContent;
assert( (pPatch[i].jnFlags & JNODE_RAW)==0 );
for(j=1; j<pTarget->n; j += jsonNodeSize(&pTarget[j+1])+1 ){
assert( pTarget[j].eType==JSON_STRING );
assert( pTarget[j].jnFlags & JNODE_LABEL );
assert( (pPatch[i].jnFlags & JNODE_RAW)==0 );
if( pTarget[j].n==nKey && strncmp(pTarget[j].u.zJContent,zKey,nKey)==0 ){
if( pTarget[j+1].jnFlags & (JNODE_REMOVE|JNODE_PATCH) ) break;
if( pPatch[i+1].eType==JSON_NULL ){
pTarget[j+1].jnFlags |= JNODE_REMOVE;
}else{
JsonNode *pNew = jsonMergePatch(pParse, iTarget+j+1, &pPatch[i+1]);
if( pNew==0 ) return 0;
pTarget = &pParse->aNode[iTarget];
if( pNew!=&pTarget[j+1] ){
assert( pTarget[j+1].eU==0
|| pTarget[j+1].eU==1
|| pTarget[j+1].eU==2 );
testcase( pTarget[j+1].eU==1 );
testcase( pTarget[j+1].eU==2 );
VVA( pTarget[j+1].eU = 5 );
pTarget[j+1].u.pPatch = pNew;
pTarget[j+1].jnFlags |= JNODE_PATCH;
}
}
break;
}
}
if( j>=pTarget->n && pPatch[i+1].eType!=JSON_NULL ){
int iStart, iPatch;
iStart = jsonParseAddNode(pParse, JSON_OBJECT, 2, 0);
jsonParseAddNode(pParse, JSON_STRING, nKey, zKey);
iPatch = jsonParseAddNode(pParse, JSON_TRUE, 0, 0);
if( pParse->oom ) return 0;
jsonRemoveAllNulls(pPatch);
pTarget = &pParse->aNode[iTarget];
assert( pParse->aNode[iRoot].eU==0 || pParse->aNode[iRoot].eU==2 );
testcase( pParse->aNode[iRoot].eU==2 );
pParse->aNode[iRoot].jnFlags |= JNODE_APPEND;
VVA( pParse->aNode[iRoot].eU = 2 );
pParse->aNode[iRoot].u.iAppend = iStart - iRoot;
iRoot = iStart;
assert( pParse->aNode[iPatch].eU==0 );
VVA( pParse->aNode[iPatch].eU = 5 );
pParse->aNode[iPatch].jnFlags |= JNODE_PATCH;
pParse->aNode[iPatch].u.pPatch = &pPatch[i+1];
}
}
return pTarget;
}
/*
** Implementation of the json_mergepatch(JSON1,JSON2) function. Return a JSON
** object that is the result of running the RFC 7396 MergePatch() algorithm
** on the two arguments.
*/
static void jsonPatchFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse x; /* The JSON that is being patched */
JsonParse y; /* The patch */
JsonNode *pResult; /* The result of the merge */
UNUSED_PARAM(argc);
if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return;
if( jsonParse(&y, ctx, (const char*)sqlite3_value_text(argv[1])) ){
jsonParseReset(&x);
return;
}
pResult = jsonMergePatch(&x, 0, y.aNode);
assert( pResult!=0 || x.oom );
if( pResult ){
jsonReturnJson(pResult, ctx, 0);
}else{
sqlite3_result_error_nomem(ctx);
}
jsonParseReset(&x);
jsonParseReset(&y);
}
/*
** Implementation of the json_object(NAME,VALUE,...) function. Return a JSON
** object that contains all name/value given in arguments. Or if any name
** is not a string or if any value is a BLOB, throw an error.
*/
static void jsonObjectFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
int i;
JsonString jx;
const char *z;
u32 n;
if( argc&1 ){
sqlite3_result_error(ctx, "json_object() requires an even number "
"of arguments", -1);
return;
}
jsonInit(&jx, ctx);
jsonAppendChar(&jx, '{');
for(i=0; i<argc; i+=2){
if( sqlite3_value_type(argv[i])!=SQLITE_TEXT ){
sqlite3_result_error(ctx, "json_object() labels must be TEXT", -1);
jsonReset(&jx);
return;
}
jsonAppendSeparator(&jx);
z = (const char*)sqlite3_value_text(argv[i]);
n = (u32)sqlite3_value_bytes(argv[i]);
jsonAppendString(&jx, z, n);
jsonAppendChar(&jx, ':');
jsonAppendValue(&jx, argv[i+1]);
}
jsonAppendChar(&jx, '}');
jsonResult(&jx);
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
/*
** json_remove(JSON, PATH, ...)
**
** Remove the named elements from JSON and return the result. malformed
** JSON or PATH arguments result in an error.
*/
static void jsonRemoveFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse x; /* The parse */
JsonNode *pNode;
const char *zPath;
u32 i;
if( argc<1 ) return;
if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return;
assert( x.nNode );
for(i=1; i<(u32)argc; i++){
zPath = (const char*)sqlite3_value_text(argv[i]);
if( zPath==0 ) goto remove_done;
pNode = jsonLookup(&x, zPath, 0, ctx);
if( x.nErr ) goto remove_done;
if( pNode ) pNode->jnFlags |= JNODE_REMOVE;
}
if( (x.aNode[0].jnFlags & JNODE_REMOVE)==0 ){
jsonReturnJson(x.aNode, ctx, 0);
}
remove_done:
jsonParseReset(&x);
}
/*
** json_replace(JSON, PATH, VALUE, ...)
**
** Replace the value at PATH with VALUE. If PATH does not already exist,
** this routine is a no-op. If JSON or PATH is malformed, throw an error.
*/
static void jsonReplaceFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse x; /* The parse */
JsonNode *pNode;
const char *zPath;
u32 i;
if( argc<1 ) return;
if( (argc&1)==0 ) {
jsonWrongNumArgs(ctx, "replace");
return;
}
if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return;
assert( x.nNode );
for(i=1; i<(u32)argc; i+=2){
zPath = (const char*)sqlite3_value_text(argv[i]);
pNode = jsonLookup(&x, zPath, 0, ctx);
if( x.nErr ) goto replace_err;
if( pNode ){
assert( pNode->eU==0 || pNode->eU==1 || pNode->eU==4 );
json_testcase( pNode->eU!=0 && pNode->eU!=1 );
pNode->jnFlags |= (u8)JNODE_REPLACE;
VVA( pNode->eU = 4 );
pNode->u.iReplace = i + 1;
}
}
if( x.aNode[0].jnFlags & JNODE_REPLACE ){
assert( x.aNode[0].eU==4 );
sqlite3_result_value(ctx, argv[x.aNode[0].u.iReplace]);
}else{
jsonReturnJson(x.aNode, ctx, argv);
}
replace_err:
jsonParseReset(&x);
}
/*
** json_set(JSON, PATH, VALUE, ...)
**
** Set the value at PATH to VALUE. Create the PATH if it does not already
** exist. Overwrite existing values that do exist.
** If JSON or PATH is malformed, throw an error.
**
** json_insert(JSON, PATH, VALUE, ...)
**
** Create PATH and initialize it to VALUE. If PATH already exists, this
** routine is a no-op. If JSON or PATH is malformed, throw an error.
*/
static void jsonSetFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse x; /* The parse */
JsonNode *pNode;
const char *zPath;
u32 i;
int bApnd;
int bIsSet = *(int*)sqlite3_user_data(ctx);
if( argc<1 ) return;
if( (argc&1)==0 ) {
jsonWrongNumArgs(ctx, bIsSet ? "set" : "insert");
return;
}
if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return;
assert( x.nNode );
for(i=1; i<(u32)argc; i+=2){
zPath = (const char*)sqlite3_value_text(argv[i]);
bApnd = 0;
pNode = jsonLookup(&x, zPath, &bApnd, ctx);
if( x.oom ){
sqlite3_result_error_nomem(ctx);
goto jsonSetDone;
}else if( x.nErr ){
goto jsonSetDone;
}else if( pNode && (bApnd || bIsSet) ){
json_testcase( pNode->eU!=0 && pNode->eU!=1 && pNode->eU!=4 );
assert( pNode->eU!=3 || pNode->eU!=5 );
VVA( pNode->eU = 4 );
pNode->jnFlags |= (u8)JNODE_REPLACE;
pNode->u.iReplace = i + 1;
}
}
if( x.aNode[0].jnFlags & JNODE_REPLACE ){
assert( x.aNode[0].eU==4 );
sqlite3_result_value(ctx, argv[x.aNode[0].u.iReplace]);
}else{
jsonReturnJson(x.aNode, ctx, argv);
}
jsonSetDone:
jsonParseReset(&x);
}
/*
** json_type(JSON)
** json_ntype(JSON)
** json_type(JSON, PATH)
**
** Return the top-level "type" of a JSON string. json_type() raises an
** error if either the JSON or PATH inputs are not well-formed. json_ntype()
** works like the one-argument version of json_type() except that it
** returns NULL if the JSON argument is not well-formed.
*/
static void jsonTypeFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse *p; /* The parse */
const char *zPath;
JsonNode *pNode;
p = jsonParseCached(ctx, argv, *(int*)sqlite3_user_data(ctx) ? 0 : ctx);
if( p==0 ) return;
if( argc==2 ){
zPath = (const char*)sqlite3_value_text(argv[1]);
pNode = jsonLookup(p, zPath, 0, ctx);
}else{
pNode = p->aNode;
}
if( pNode ){
sqlite3_result_text(ctx, jsonType[pNode->eType], -1, SQLITE_STATIC);
}
}
/*
** json_valid(JSON)
**
** Return 1 if JSON is a well-formed JSON string according to RFC-7159.
** Return 0 otherwise.
*/
static void jsonValidFunc(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonParse *p; /* The parse */
UNUSED_PARAM(argc);
p = jsonParseCached(ctx, argv, 0);
sqlite3_result_int(ctx, p!=0);
}
/****************************************************************************
** Aggregate SQL function implementations
****************************************************************************/
/*
** json_group_array(VALUE)
**
** Return a JSON array composed of all values in the aggregate.
*/
static void jsonArrayStep(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonString *pStr;
UNUSED_PARAM(argc);
pStr = (JsonString*)sqlite3_aggregate_context(ctx, sizeof(*pStr));
if( pStr ){
if( pStr->zBuf==0 ){
jsonInit(pStr, ctx);
jsonAppendChar(pStr, '[');
}else if( pStr->nUsed>1 ){
jsonAppendChar(pStr, ',');
}
pStr->pCtx = ctx;
jsonAppendValue(pStr, argv[0]);
}
}
static void jsonArrayCompute(sqlite3_context *ctx, int isFinal){
JsonString *pStr;
pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0);
if( pStr ){
pStr->pCtx = ctx;
jsonAppendChar(pStr, ']');
if( pStr->bErr ){
if( pStr->bErr==1 ) sqlite3_result_error_nomem(ctx);
assert( pStr->bStatic );
}else if( isFinal ){
sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed,
pStr->bStatic ? SQLITE_TRANSIENT : sqlite3_free);
pStr->bStatic = 1;
}else{
sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, SQLITE_TRANSIENT);
pStr->nUsed--;
}
}else{
sqlite3_result_text(ctx, "[]", 2, SQLITE_STATIC);
}
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
static void jsonArrayValue(sqlite3_context *ctx){
jsonArrayCompute(ctx, 0);
}
static void jsonArrayFinal(sqlite3_context *ctx){
jsonArrayCompute(ctx, 1);
}
#ifndef SQLITE_OMIT_WINDOWFUNC
/*
** This method works for both json_group_array() and json_group_object().
** It works by removing the first element of the group by searching forward
** to the first comma (",") that is not within a string and deleting all
** text through that comma.
*/
static void jsonGroupInverse(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
unsigned int i;
int inStr = 0;
int nNest = 0;
char *z;
char c;
JsonString *pStr;
UNUSED_PARAM(argc);
UNUSED_PARAM(argv);
pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0);
#ifdef NEVER
/* pStr is always non-NULL since jsonArrayStep() or jsonObjectStep() will
** always have been called to initalize it */
if( NEVER(!pStr) ) return;
#endif
z = pStr->zBuf;
for(i=1; i<pStr->nUsed && ((c = z[i])!=',' || inStr || nNest); i++){
if( c=='"' ){
inStr = !inStr;
}else if( c=='\\' ){
i++;
}else if( !inStr ){
if( c=='{' || c=='[' ) nNest++;
if( c=='}' || c==']' ) nNest--;
}
}
if( i<pStr->nUsed ){
pStr->nUsed -= i;
memmove(&z[1], &z[i+1], (size_t)pStr->nUsed-1);
z[pStr->nUsed] = 0;
}else{
pStr->nUsed = 1;
}
}
#else
# define jsonGroupInverse 0
#endif
/*
** json_group_obj(NAME,VALUE)
**
** Return a JSON object composed of all names and values in the aggregate.
*/
static void jsonObjectStep(
sqlite3_context *ctx,
int argc,
sqlite3_value **argv
){
JsonString *pStr;
const char *z;
u32 n;
UNUSED_PARAM(argc);
pStr = (JsonString*)sqlite3_aggregate_context(ctx, sizeof(*pStr));
if( pStr ){
if( pStr->zBuf==0 ){
jsonInit(pStr, ctx);
jsonAppendChar(pStr, '{');
}else if( pStr->nUsed>1 ){
jsonAppendChar(pStr, ',');
}
pStr->pCtx = ctx;
z = (const char*)sqlite3_value_text(argv[0]);
n = (u32)sqlite3_value_bytes(argv[0]);
jsonAppendString(pStr, z, n);
jsonAppendChar(pStr, ':');
jsonAppendValue(pStr, argv[1]);
}
}
static void jsonObjectCompute(sqlite3_context *ctx, int isFinal){
JsonString *pStr;
pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0);
if( pStr ){
jsonAppendChar(pStr, '}');
if( pStr->bErr ){
if( pStr->bErr==1 ) sqlite3_result_error_nomem(ctx);
assert( pStr->bStatic );
}else if( isFinal ){
sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed,
pStr->bStatic ? SQLITE_TRANSIENT : sqlite3_free);
pStr->bStatic = 1;
}else{
sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, SQLITE_TRANSIENT);
pStr->nUsed--;
}
}else{
sqlite3_result_text(ctx, "{}", 2, SQLITE_STATIC);
}
sqlite3_result_subtype(ctx, JSON_SUBTYPE);
}
static void jsonObjectValue(sqlite3_context *ctx){
jsonObjectCompute(ctx, 0);
}
static void jsonObjectFinal(sqlite3_context *ctx){
jsonObjectCompute(ctx, 1);
}
#ifndef SQLITE_OMIT_VIRTUALTABLE
/****************************************************************************
** The json_each virtual table
****************************************************************************/
typedef struct JsonEachCursor JsonEachCursor;
struct JsonEachCursor {
sqlite3_vtab_cursor base; /* Base class - must be first */
u32 iRowid; /* The rowid */
u32 iBegin; /* The first node of the scan */
u32 i; /* Index in sParse.aNode[] of current row */
u32 iEnd; /* EOF when i equals or exceeds this value */
u8 eType; /* Type of top-level element */
u8 bRecursive; /* True for json_tree(). False for json_each() */
char *zJson; /* Input JSON */
char *zRoot; /* Path by which to filter zJson */
JsonParse sParse; /* Parse of the input JSON */
};
/* Constructor for the json_each virtual table */
static int jsonEachConnect(
sqlite3 *db,
void *pAux,
int argc, const char *const*argv,
sqlite3_vtab **ppVtab,
char **pzErr
){
sqlite3_vtab *pNew;
int rc;
/* Column numbers */
#define JEACH_KEY 0
#define JEACH_VALUE 1
#define JEACH_TYPE 2
#define JEACH_ATOM 3
#define JEACH_ID 4
#define JEACH_PARENT 5
#define JEACH_FULLKEY 6
#define JEACH_PATH 7
/* The xBestIndex method assumes that the JSON and ROOT columns are
** the last two columns in the table. Should this ever changes, be
** sure to update the xBestIndex method. */
#define JEACH_JSON 8
#define JEACH_ROOT 9
UNUSED_PARAM(pzErr);
UNUSED_PARAM(argv);
UNUSED_PARAM(argc);
UNUSED_PARAM(pAux);
rc = sqlite3_declare_vtab(db,
"CREATE TABLE x(key,value,type,atom,id,parent,fullkey,path,"
"json HIDDEN,root HIDDEN)");
if( rc==SQLITE_OK ){
pNew = *ppVtab = sqlite3_malloc( sizeof(*pNew) );
if( pNew==0 ) return SQLITE_NOMEM;
memset(pNew, 0, sizeof(*pNew));
sqlite3_vtab_config(db, SQLITE_VTAB_INNOCUOUS);
}
return rc;
}
/* destructor for json_each virtual table */
static int jsonEachDisconnect(sqlite3_vtab *pVtab){
sqlite3_free(pVtab);
return SQLITE_OK;
}
/* constructor for a JsonEachCursor object for json_each(). */
static int jsonEachOpenEach(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){
JsonEachCursor *pCur;
UNUSED_PARAM(p);
pCur = sqlite3_malloc( sizeof(*pCur) );
if( pCur==0 ) return SQLITE_NOMEM;
memset(pCur, 0, sizeof(*pCur));
*ppCursor = &pCur->base;
return SQLITE_OK;
}
/* constructor for a JsonEachCursor object for json_tree(). */
static int jsonEachOpenTree(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){
int rc = jsonEachOpenEach(p, ppCursor);
if( rc==SQLITE_OK ){
JsonEachCursor *pCur = (JsonEachCursor*)*ppCursor;
pCur->bRecursive = 1;
}
return rc;
}
/* Reset a JsonEachCursor back to its original state. Free any memory
** held. */
static void jsonEachCursorReset(JsonEachCursor *p){
sqlite3_free(p->zJson);
sqlite3_free(p->zRoot);
jsonParseReset(&p->sParse);
p->iRowid = 0;
p->i = 0;
p->iEnd = 0;
p->eType = 0;
p->zJson = 0;
p->zRoot = 0;
}
/* Destructor for a jsonEachCursor object */
static int jsonEachClose(sqlite3_vtab_cursor *cur){
JsonEachCursor *p = (JsonEachCursor*)cur;
jsonEachCursorReset(p);
sqlite3_free(cur);
return SQLITE_OK;
}
/* Return TRUE if the jsonEachCursor object has been advanced off the end
** of the JSON object */
static int jsonEachEof(sqlite3_vtab_cursor *cur){
JsonEachCursor *p = (JsonEachCursor*)cur;
return p->i >= p->iEnd;
}
/* Advance the cursor to the next element for json_tree() */
static int jsonEachNext(sqlite3_vtab_cursor *cur){
JsonEachCursor *p = (JsonEachCursor*)cur;
if( p->bRecursive ){
if( p->sParse.aNode[p->i].jnFlags & JNODE_LABEL ) p->i++;
p->i++;
p->iRowid++;
if( p->i<p->iEnd ){
u32 iUp = p->sParse.aUp[p->i];
JsonNode *pUp = &p->sParse.aNode[iUp];
p->eType = pUp->eType;
if( pUp->eType==JSON_ARRAY ){
assert( pUp->eU==0 || pUp->eU==3 );
json_testcase( pUp->eU==3 );
VVA( pUp->eU = 3 );
if( iUp==p->i-1 ){
pUp->u.iKey = 0;
}else{
pUp->u.iKey++;
}
}
}
}else{
switch( p->eType ){
case JSON_ARRAY: {
p->i += jsonNodeSize(&p->sParse.aNode[p->i]);
p->iRowid++;
break;
}
case JSON_OBJECT: {
p->i += 1 + jsonNodeSize(&p->sParse.aNode[p->i+1]);
p->iRowid++;
break;
}
default: {
p->i = p->iEnd;
break;
}
}
}
return SQLITE_OK;
}
/* Append the name of the path for element i to pStr
*/
static void jsonEachComputePath(
JsonEachCursor *p, /* The cursor */
JsonString *pStr, /* Write the path here */
u32 i /* Path to this element */
){
JsonNode *pNode, *pUp;
u32 iUp;
if( i==0 ){
jsonAppendChar(pStr, '$');
return;
}
iUp = p->sParse.aUp[i];
jsonEachComputePath(p, pStr, iUp);
pNode = &p->sParse.aNode[i];
pUp = &p->sParse.aNode[iUp];
if( pUp->eType==JSON_ARRAY ){
assert( pUp->eU==3 || (pUp->eU==0 && pUp->u.iKey==0) );
testcase( pUp->eU==0 );
jsonPrintf(30, pStr, "[%d]", pUp->u.iKey);
}else{
assert( pUp->eType==JSON_OBJECT );
if( (pNode->jnFlags & JNODE_LABEL)==0 ) pNode--;
assert( pNode->eType==JSON_STRING );
assert( pNode->jnFlags & JNODE_LABEL );
assert( pNode->eU==1 );
jsonPrintf(pNode->n+1, pStr, ".%.*s", pNode->n-2, pNode->u.zJContent+1);
}
}
/* Return the value of a column */
static int jsonEachColumn(
sqlite3_vtab_cursor *cur, /* The cursor */
sqlite3_context *ctx, /* First argument to sqlite3_result_...() */
int i /* Which column to return */
){
JsonEachCursor *p = (JsonEachCursor*)cur;
JsonNode *pThis = &p->sParse.aNode[p->i];
switch( i ){
case JEACH_KEY: {
if( p->i==0 ) break;
if( p->eType==JSON_OBJECT ){
jsonReturn(pThis, ctx, 0);
}else if( p->eType==JSON_ARRAY ){
u32 iKey;
if( p->bRecursive ){
if( p->iRowid==0 ) break;
assert( p->sParse.aNode[p->sParse.aUp[p->i]].eU==3 );
iKey = p->sParse.aNode[p->sParse.aUp[p->i]].u.iKey;
}else{
iKey = p->iRowid;
}
sqlite3_result_int64(ctx, (sqlite3_int64)iKey);
}
break;
}
case JEACH_VALUE: {
if( pThis->jnFlags & JNODE_LABEL ) pThis++;
jsonReturn(pThis, ctx, 0);
break;
}
case JEACH_TYPE: {
if( pThis->jnFlags & JNODE_LABEL ) pThis++;
sqlite3_result_text(ctx, jsonType[pThis->eType], -1, SQLITE_STATIC);
break;
}
case JEACH_ATOM: {
if( pThis->jnFlags & JNODE_LABEL ) pThis++;
if( pThis->eType>=JSON_ARRAY ) break;
jsonReturn(pThis, ctx, 0);
break;
}
case JEACH_ID: {
sqlite3_result_int64(ctx,
(sqlite3_int64)p->i + ((pThis->jnFlags & JNODE_LABEL)!=0));
break;
}
case JEACH_PARENT: {
if( p->i>p->iBegin && p->bRecursive ){
sqlite3_result_int64(ctx, (sqlite3_int64)p->sParse.aUp[p->i]);
}
break;
}
case JEACH_FULLKEY: {
JsonString x;
jsonInit(&x, ctx);
if( p->bRecursive ){
jsonEachComputePath(p, &x, p->i);
}else{
if( p->zRoot ){
jsonAppendRaw(&x, p->zRoot, (int)strlen(p->zRoot));
}else{
jsonAppendChar(&x, '$');
}
if( p->eType==JSON_ARRAY ){
jsonPrintf(30, &x, "[%d]", p->iRowid);
}else if( p->eType==JSON_OBJECT ){
assert( pThis->eU==1 );
jsonPrintf(pThis->n, &x, ".%.*s", pThis->n-2, pThis->u.zJContent+1);
}
}
jsonResult(&x);
break;
}
case JEACH_PATH: {
if( p->bRecursive ){
JsonString x;
jsonInit(&x, ctx);
jsonEachComputePath(p, &x, p->sParse.aUp[p->i]);
jsonResult(&x);
break;
}
/* For json_each() path and root are the same so fall through
** into the root case */
/* no break */ deliberate_fall_through
}
default: {
const char *zRoot = p->zRoot;
if( zRoot==0 ) zRoot = "$";
sqlite3_result_text(ctx, zRoot, -1, SQLITE_STATIC);
break;
}
case JEACH_JSON: {
assert( i==JEACH_JSON );
sqlite3_result_text(ctx, p->sParse.zJson, -1, SQLITE_STATIC);
break;
}
}
return SQLITE_OK;
}
/* Return the current rowid value */
static int jsonEachRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
JsonEachCursor *p = (JsonEachCursor*)cur;
*pRowid = p->iRowid;
return SQLITE_OK;
}
/* The query strategy is to look for an equality constraint on the json
** column. Without such a constraint, the table cannot operate. idxNum is
** 1 if the constraint is found, 3 if the constraint and zRoot are found,
** and 0 otherwise.
*/
static int jsonEachBestIndex(
sqlite3_vtab *tab,
sqlite3_index_info *pIdxInfo
){
int i; /* Loop counter or computed array index */
int aIdx[2]; /* Index of constraints for JSON and ROOT */
int unusableMask = 0; /* Mask of unusable JSON and ROOT constraints */
int idxMask = 0; /* Mask of usable == constraints JSON and ROOT */
const struct sqlite3_index_constraint *pConstraint;
/* This implementation assumes that JSON and ROOT are the last two
** columns in the table */
assert( JEACH_ROOT == JEACH_JSON+1 );
UNUSED_PARAM(tab);
aIdx[0] = aIdx[1] = -1;
pConstraint = pIdxInfo->aConstraint;
for(i=0; i<pIdxInfo->nConstraint; i++, pConstraint++){
int iCol;
int iMask;
if( pConstraint->iColumn < JEACH_JSON ) continue;
iCol = pConstraint->iColumn - JEACH_JSON;
assert( iCol==0 || iCol==1 );
testcase( iCol==0 );
iMask = 1 << iCol;
if( pConstraint->usable==0 ){
unusableMask |= iMask;
}else if( pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){
aIdx[iCol] = i;
idxMask |= iMask;
}
}
if( (unusableMask & ~idxMask)!=0 ){
/* If there are any unusable constraints on JSON or ROOT, then reject
** this entire plan */
return SQLITE_CONSTRAINT;
}
if( aIdx[0]<0 ){
/* No JSON input. Leave estimatedCost at the huge value that it was
** initialized to to discourage the query planner from selecting this
** plan. */
pIdxInfo->idxNum = 0;
}else{
pIdxInfo->estimatedCost = 1.0;
i = aIdx[0];
pIdxInfo->aConstraintUsage[i].argvIndex = 1;
pIdxInfo->aConstraintUsage[i].omit = 1;
if( aIdx[1]<0 ){
pIdxInfo->idxNum = 1; /* Only JSON supplied. Plan 1 */
}else{
i = aIdx[1];
pIdxInfo->aConstraintUsage[i].argvIndex = 2;
pIdxInfo->aConstraintUsage[i].omit = 1;
pIdxInfo->idxNum = 3; /* Both JSON and ROOT are supplied. Plan 3 */
}
}
return SQLITE_OK;
}
/* Start a search on a new JSON string */
static int jsonEachFilter(
sqlite3_vtab_cursor *cur,
int idxNum, const char *idxStr,
int argc, sqlite3_value **argv
){
JsonEachCursor *p = (JsonEachCursor*)cur;
const char *z;
const char *zRoot = 0;
sqlite3_int64 n;
UNUSED_PARAM(idxStr);
UNUSED_PARAM(argc);
jsonEachCursorReset(p);
if( idxNum==0 ) return SQLITE_OK;
z = (const char*)sqlite3_value_text(argv[0]);
if( z==0 ) return SQLITE_OK;
n = sqlite3_value_bytes(argv[0]);
p->zJson = sqlite3_malloc64( n+1 );
if( p->zJson==0 ) return SQLITE_NOMEM;
memcpy(p->zJson, z, (size_t)n+1);
if( jsonParse(&p->sParse, 0, p->zJson) ){
int rc = SQLITE_NOMEM;
if( p->sParse.oom==0 ){
sqlite3_free(cur->pVtab->zErrMsg);
cur->pVtab->zErrMsg = sqlite3_mprintf("malformed JSON");
if( cur->pVtab->zErrMsg ) rc = SQLITE_ERROR;
}
jsonEachCursorReset(p);
return rc;
}else if( p->bRecursive && jsonParseFindParents(&p->sParse) ){
jsonEachCursorReset(p);
return SQLITE_NOMEM;
}else{
JsonNode *pNode = 0;
if( idxNum==3 ){
const char *zErr = 0;
zRoot = (const char*)sqlite3_value_text(argv[1]);
if( zRoot==0 ) return SQLITE_OK;
n = sqlite3_value_bytes(argv[1]);
p->zRoot = sqlite3_malloc64( n+1 );
if( p->zRoot==0 ) return SQLITE_NOMEM;
memcpy(p->zRoot, zRoot, (size_t)n+1);
if( zRoot[0]!='$' ){
zErr = zRoot;
}else{
pNode = jsonLookupStep(&p->sParse, 0, p->zRoot+1, 0, &zErr);
}
if( zErr ){
sqlite3_free(cur->pVtab->zErrMsg);
cur->pVtab->zErrMsg = jsonPathSyntaxError(zErr);
jsonEachCursorReset(p);
return cur->pVtab->zErrMsg ? SQLITE_ERROR : SQLITE_NOMEM;
}else if( pNode==0 ){
return SQLITE_OK;
}
}else{
pNode = p->sParse.aNode;
}
p->iBegin = p->i = (int)(pNode - p->sParse.aNode);
p->eType = pNode->eType;
if( p->eType>=JSON_ARRAY ){
assert( pNode->eU==0 );
VVA( pNode->eU = 3 );
pNode->u.iKey = 0;
p->iEnd = p->i + pNode->n + 1;
if( p->bRecursive ){
p->eType = p->sParse.aNode[p->sParse.aUp[p->i]].eType;
if( p->i>0 && (p->sParse.aNode[p->i-1].jnFlags & JNODE_LABEL)!=0 ){
p->i--;
}
}else{
p->i++;
}
}else{
p->iEnd = p->i+1;
}
}
return SQLITE_OK;
}
/* The methods of the json_each virtual table */
static sqlite3_module jsonEachModule = {
0, /* iVersion */
0, /* xCreate */
jsonEachConnect, /* xConnect */
jsonEachBestIndex, /* xBestIndex */
jsonEachDisconnect, /* xDisconnect */
0, /* xDestroy */
jsonEachOpenEach, /* xOpen - open a cursor */
jsonEachClose, /* xClose - close a cursor */
jsonEachFilter, /* xFilter - configure scan constraints */
jsonEachNext, /* xNext - advance a cursor */
jsonEachEof, /* xEof - check for end of scan */
jsonEachColumn, /* xColumn - read data */
jsonEachRowid, /* xRowid - read data */
0, /* xUpdate */
0, /* xBegin */
0, /* xSync */
0, /* xCommit */
0, /* xRollback */
0, /* xFindMethod */
0, /* xRename */
0, /* xSavepoint */
0, /* xRelease */
0, /* xRollbackTo */
0 /* xShadowName */
};
/* The methods of the json_tree virtual table. */
static sqlite3_module jsonTreeModule = {
0, /* iVersion */
0, /* xCreate */
jsonEachConnect, /* xConnect */
jsonEachBestIndex, /* xBestIndex */
jsonEachDisconnect, /* xDisconnect */
0, /* xDestroy */
jsonEachOpenTree, /* xOpen - open a cursor */
jsonEachClose, /* xClose - close a cursor */
jsonEachFilter, /* xFilter - configure scan constraints */
jsonEachNext, /* xNext - advance a cursor */
jsonEachEof, /* xEof - check for end of scan */
jsonEachColumn, /* xColumn - read data */
jsonEachRowid, /* xRowid - read data */
0, /* xUpdate */
0, /* xBegin */
0, /* xSync */
0, /* xCommit */
0, /* xRollback */
0, /* xFindMethod */
0, /* xRename */
0, /* xSavepoint */
0, /* xRelease */
0, /* xRollbackTo */
0 /* xShadowName */
};
#endif /* SQLITE_OMIT_VIRTUALTABLE */
/****************************************************************************
** The following routines are the only publically visible identifiers in this
** file. Call the following routines in order to register the various SQL
** functions and the virtual table implemented by this file.
****************************************************************************/
int sqlite3Json1Init(sqlite3 *db){
int rc = SQLITE_OK;
unsigned int i;
static const struct {
const char *zName;
void (*xFunc)(sqlite3_context*,int,sqlite3_value**);
int nArg;
int flag;
} aFunc[] = {
{ "json", jsonRemoveFunc, 1, 0 },
{ "json_array", jsonArrayFunc, -1, 0 },
{ "json_array_length", jsonArrayLengthFunc, 1, 0 },
{ "json_array_length", jsonArrayLengthFunc, 2, 0 },
{ "json_extract", jsonExtractFunc, -1, 0 },
{ "json_nextract", jsonExtractFunc, -1, JSON_NULLERR },
{ "->", jsonExtractFunc, 2, JSON_NULLERR|JSON_ABPATH },
{ "->>", jsonExtractFunc, 2, JSON_ABPATH },
{ "json_insert", jsonSetFunc, -1, 0 },
{ "json_object", jsonObjectFunc, -1, 0 },
{ "json_patch", jsonPatchFunc, 2, 0 },
{ "json_quote", jsonQuoteFunc, 1, 0 },
{ "json_remove", jsonRemoveFunc, -1, 0 },
{ "json_replace", jsonReplaceFunc, -1, 0 },
{ "json_set", jsonSetFunc, -1, JSON_ISSET },
{ "json_type", jsonTypeFunc, 1, 0 },
{ "json_ntype", jsonTypeFunc, 1, JSON_NULLERR },
{ "json_type", jsonTypeFunc, 2, 0 },
{ "json_valid", jsonValidFunc, 1, 0 },
#if SQLITE_DEBUG
/* DEBUG and TESTING functions */
{ "json_parse", jsonParseFunc, 1, 0 },
{ "json_test1", jsonTest1Func, 1, 0 },
#endif
};
static const struct {
const char *zName;
int nArg;
void (*xStep)(sqlite3_context*,int,sqlite3_value**);
void (*xFinal)(sqlite3_context*);
void (*xValue)(sqlite3_context*);
} aAgg[] = {
{ "json_group_array", 1,
jsonArrayStep, jsonArrayFinal, jsonArrayValue },
{ "json_group_object", 2,
jsonObjectStep, jsonObjectFinal, jsonObjectValue },
};
#ifndef SQLITE_OMIT_VIRTUALTABLE
static const struct {
const char *zName;
sqlite3_module *pModule;
} aMod[] = {
{ "json_each", &jsonEachModule },
{ "json_tree", &jsonTreeModule },
};
#endif
static const int enc =
SQLITE_UTF8 |
SQLITE_DETERMINISTIC |
SQLITE_INNOCUOUS;
for(i=0; i<sizeof(aFunc)/sizeof(aFunc[0]) && rc==SQLITE_OK; i++){
rc = sqlite3_create_function(db, aFunc[i].zName, aFunc[i].nArg, enc,
(void*)&aFunc[i].flag,
aFunc[i].xFunc, 0, 0);
}
#ifndef SQLITE_OMIT_WINDOWFUNC
for(i=0; i<sizeof(aAgg)/sizeof(aAgg[0]) && rc==SQLITE_OK; i++){
rc = sqlite3_create_window_function(db, aAgg[i].zName, aAgg[i].nArg,
SQLITE_SUBTYPE | enc, 0,
aAgg[i].xStep, aAgg[i].xFinal,
aAgg[i].xValue, jsonGroupInverse, 0);
}
#endif
#ifndef SQLITE_OMIT_VIRTUALTABLE
for(i=0; i<sizeof(aMod)/sizeof(aMod[0]) && rc==SQLITE_OK; i++){
rc = sqlite3_create_module(db, aMod[i].zName, aMod[i].pModule, 0);
}
#endif
return rc;
}
#ifndef SQLITE_CORE
#ifdef _WIN32
__declspec(dllexport)
#endif
int sqlite3_json_init(
sqlite3 *db,
char **pzErrMsg,
const sqlite3_api_routines *pApi
){
SQLITE_EXTENSION_INIT2(pApi);
(void)pzErrMsg; /* Unused parameter */
return sqlite3Json1Init(db);
}
#endif
#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_JSON1) */