Index: src/json.c ================================================================== --- src/json.c +++ src/json.c @@ -3214,10 +3214,42 @@ }else{ jsonBadPathError(ctx, zPath); } return; } + +/* +** If pArg is a blob that seems like a JSONB blob, then initialize +** p to point to that JSONB and return TRUE. If pArg does not seem like +** a JSONB blob, then return FALSE; +** +** This routine is only called if it is already known that pArg is a +** blob. The only open question is whether or not the blob appears +** to be a JSONB blob. +*/ +static int jsonArgIsJsonb(sqlite3_value *pArg, JsonParse *p){ + u32 n, sz = 0; + p->aBlob = (u8*)sqlite3_value_blob(pArg); + p->nBlob = (u32)sqlite3_value_bytes(pArg); + if( p->nBlob==0 ){ + p->aBlob = 0; + return 0; + } + if( NEVER(p->aBlob==0) ){ + return 0; + } + if( (p->aBlob[0] & 0x0f)<=JSONB_OBJECT + && (n = jsonbPayloadSize(p, 0, &sz))>0 + && sz+n==p->nBlob + && ((p->aBlob[0] & 0x0f)>JSONB_FALSE || sz==0) + ){ + return 1; + } + p->aBlob = 0; + p->nBlob = 0; + return 0; +} /* ** Generate a JsonParse object, containing valid JSONB in aBlob and nBlob, ** from the SQL function argument pArg. Return a pointer to the new ** JsonParse object. @@ -3271,33 +3303,28 @@ p->hasNonstd = pFromCache->hasNonstd; jsonParseFree(pFromCache); return p; } if( eType==SQLITE_BLOB ){ - u32 n, sz = 0; - p->aBlob = (u8*)sqlite3_value_blob(pArg); - p->nBlob = (u32)sqlite3_value_bytes(pArg); - if( p->nBlob==0 ){ - goto json_pfa_malformed; - } - if( NEVER(p->aBlob==0) ){ - goto json_pfa_oom; - } - if( (p->aBlob[0] & 0x0f)>JSONB_OBJECT ){ - goto json_pfa_malformed; - } - n = jsonbPayloadSize(p, 0, &sz); - if( n==0 - || sz+n!=p->nBlob - || ((p->aBlob[0] & 0x0f)<=JSONB_FALSE && sz>0) - ){ - goto json_pfa_malformed; - } - if( (flgs & JSON_EDITABLE)!=0 && jsonBlobMakeEditable(p, 0)==0 ){ - goto json_pfa_oom; - } - return p; + if( jsonArgIsJsonb(pArg,p) ){ + if( (flgs & JSON_EDITABLE)!=0 && jsonBlobMakeEditable(p, 0)==0 ){ + goto json_pfa_oom; + } + return p; + } + /* If the blob is not valid JSONB, fall through into trying to cast + ** the blob into text which is then interpreted as JSON. (tag-20240123-a) + ** + ** This goes against all historical documentation about how the SQLite + ** JSON functions were suppose to work. From the beginning, blob was + ** reserved for expansion and a blob value should have raised an error. + ** But it did not, due to a bug. And many applications came to depend + ** upon this buggy behavior, espeically when using the CLI and reading + ** JSON text using readfile(), which returns a blob. For this reason + ** we will continue to support the bug moving forward. + ** See for example https://sqlite.org/forum/forumpost/012136abd5292b8d + */ } p->zJson = (char*)sqlite3_value_text(pArg); p->nJson = sqlite3_value_bytes(pArg); if( p->nJson==0 ) goto json_pfa_malformed; if( NEVER(p->zJson==0) ) goto json_pfa_oom; @@ -4269,16 +4296,16 @@ sqlite3_result_int(ctx, 0); #endif return; } case SQLITE_BLOB: { - if( (flags & 0x0c)!=0 && jsonFuncArgMightBeBinary(argv[0]) ){ + if( jsonFuncArgMightBeBinary(argv[0]) ){ if( flags & 0x04 ){ /* Superficial checking only - accomplished by the ** jsonFuncArgMightBeBinary() call above. */ res = 1; - }else{ + }else if( flags & 0x08 ){ /* Strict checking. Check by translating BLOB->TEXT->BLOB. If ** no errors occur, call that a "strict check". */ JsonParse px; u32 iErr; memset(&px, 0, sizeof(px)); @@ -4285,12 +4312,15 @@ px.aBlob = (u8*)sqlite3_value_blob(argv[0]); px.nBlob = sqlite3_value_bytes(argv[0]); iErr = jsonbValidityCheck(&px, 0, px.nBlob, 1); res = iErr==0; } + break; } - break; + /* Fall through into interpreting the input as text. See note + ** above at tag-20240123-a. */ + /* no break */ deliberate_fall_through } default: { JsonParse px; if( (flags & 0x3)==0 ) break; memset(&px, 0, sizeof(px)); @@ -5022,17 +5052,13 @@ jsonEachCursorReset(p); if( idxNum==0 ) return SQLITE_OK; memset(&p->sParse, 0, sizeof(p->sParse)); p->sParse.nJPRef = 1; p->sParse.db = p->db; - if( sqlite3_value_type(argv[0])==SQLITE_BLOB ){ - if( jsonFuncArgMightBeBinary(argv[0]) ){ - p->sParse.nBlob = sqlite3_value_bytes(argv[0]); - p->sParse.aBlob = (u8*)sqlite3_value_blob(argv[0]); - }else{ - goto json_each_malformed_input; - } + if( jsonFuncArgMightBeBinary(argv[0]) ){ + p->sParse.nBlob = sqlite3_value_bytes(argv[0]); + p->sParse.aBlob = (u8*)sqlite3_value_blob(argv[0]); }else{ p->sParse.zJson = (char*)sqlite3_value_text(argv[0]); p->sParse.nJson = sqlite3_value_bytes(argv[0]); if( p->sParse.zJson==0 ){ p->i = p->iEnd = 0; ADDED test/json107.test Index: test/json107.test ================================================================== --- /dev/null +++ test/json107.test @@ -0,0 +1,86 @@ +# 2024-01-23 +# +# 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. +# +#*********************************************************************** +# +# Legacy JSON bug: If the input is a BLOB that when cast into TEXT looks +# like valid JSON, then treat it as valid JSON. +# +# The original intent of the JSON functions was to raise an error on any +# BLOB input. That intent was clearly documented, but the code failed to +# to implement it. Subsequently, many applications began to depend on the +# incorrect behavior, especially apps that used readfile() to read JSON +# content, since readfile() returns a BLOB. So we need to support the +# bug moving forward. +# +# The tests in this fail verify that the original buggy behavior is +# preserved. +# + +set testdir [file dirname $argv0] +source $testdir/tester.tcl +set testprefix json107 + +if {[db one {PRAGMA encoding}]!="UTF-8"} { + # These tests only work for a UTF-8 encoding. + finish_test + return +} + +do_execsql_test 1.1 { + SELECT json_valid( CAST('{"a":1}' AS BLOB) ); +} 1 +do_execsql_test 1.1.1 { + SELECT json_valid( CAST('{"a":1}' AS BLOB), 1); +} 1 +do_execsql_test 1.1.2 { + SELECT json_valid( CAST('{"a":1}' AS BLOB), 2); +} 1 +do_execsql_test 1.1.4 { + SELECT json_valid( CAST('{"a":1}' AS BLOB), 4); +} 0 +do_execsql_test 1.1.8 { + SELECT json_valid( CAST('{"a":1}' AS BLOB), 8); +} 0 + +do_execsql_test 1.2.1 { + SELECT CAST('{"a":123}' AS blob) -> 'a'; +} 123 +do_execsql_test 1.2.2 { + SELECT CAST('{"a":123}' AS blob) ->> 'a'; +} 123 +do_execsql_test 1.2.3 { + SELECT json_extract(CAST('{"a":123}' AS blob), '$.a'); +} 123 +do_execsql_test 1.3 { + SELECT json_insert(CAST('{"a":123}' AS blob),'$.b',456); +} {{{"a":123,"b":456}}} +do_execsql_test 1.4 { + SELECT json_remove(CAST('{"a":123,"b":456}' AS blob),'$.a'); +} {{{"b":456}}} +do_execsql_test 1.5 { + SELECT json_set(CAST('{"a":123,"b":456}' AS blob),'$.a',789); +} {{{"a":789,"b":456}}} +do_execsql_test 1.6 { + SELECT json_replace(CAST('{"a":123,"b":456}' AS blob),'$.a',789); +} {{{"a":789,"b":456}}} +do_execsql_test 1.7 { + SELECT json_type(CAST('{"a":123,"b":456}' AS blob)); +} object +do_execsql_test 1.8 { + SELECT json(CAST('{"a":123,"b":456}' AS blob)); +} {{{"a":123,"b":456}}} + +ifcapable vtab { + do_execsql_test 2.1 { + SELECT key, value FROM json_tree( CAST('{"a":123,"b":456}' AS blob) ) + WHERE atom; + } {a 123 b 456} +} +finish_test