From 552da5a6475972db2e8194ac6f204b44204c4dc8 Mon Sep 17 00:00:00 2001 From: Colin Dellow Date: Fri, 2 Mar 2018 18:59:34 -0500 Subject: [PATCH] Initial checkin of CSV table parquet.cc is a fork of the sample CSV virtual table at https://www.sqlite.org/src/artifact?ci=trunk&filename=ext/misc/csv.c So far the only changes are those needed to make it compile cleanly in C++11 mode. --- README.md | 20 +- parquet/.gitignore | 2 + parquet/cmds.txt | 3 + parquet/go | 1 + parquet/make | 25 ++ parquet/parquet.cc | 909 +++++++++++++++++++++++++++++++++++++++++++++ 6 files changed, 959 insertions(+), 1 deletion(-) create mode 100644 parquet/.gitignore create mode 100644 parquet/cmds.txt create mode 100755 parquet/go create mode 100755 parquet/make create mode 100644 parquet/parquet.cc diff --git a/README.md b/README.md index de7b1a1..99c7555 100644 --- a/README.md +++ b/README.md @@ -1,2 +1,20 @@ # parquet-vtable -A SQLite vtable extension to read Parquet files + +A SQLite [virtual table](https://sqlite.org/vtab.html) extension to expose Parquet files as SQL tables. + +## Building + +1. Install [`parquet-cpp`](https://github.com/apache/parquet-cpp) +2. Run `./build-sqlite` to fetch and build the SQLite dev bits +3. Run `./parquet/make` to build the module + 1. You will need to fixup the paths in this file to point at your local parquet-cpp folder. + +## Use + +``` +$ sqlite/sqlite3 +sqlite> .load parquet/libparquet +sqlite> create virtual table demo USING parquet('demo.parquet'); +sqlite> select * from demo limit 1; +...if all goes well, you'll see data here!... +``` diff --git a/parquet/.gitignore b/parquet/.gitignore new file mode 100644 index 0000000..deb64be --- /dev/null +++ b/parquet/.gitignore @@ -0,0 +1,2 @@ +cmds.txt +go diff --git a/parquet/cmds.txt b/parquet/cmds.txt new file mode 100644 index 0000000..4c77a73 --- /dev/null +++ b/parquet/cmds.txt @@ -0,0 +1,3 @@ +.load ./libparquet +create virtual table parquet using parquet(filename='../csv.csv'); +select * from parquet; diff --git a/parquet/go b/parquet/go new file mode 100755 index 0000000..ec520d8 --- /dev/null +++ b/parquet/go @@ -0,0 +1 @@ +./make && ../sqlite/sqlite3 -init ./cmds.txt < /dev/null diff --git a/parquet/make b/parquet/make new file mode 100755 index 0000000..1ab3a82 --- /dev/null +++ b/parquet/make @@ -0,0 +1,25 @@ +#!/bin/bash + +set -x + +PARQUET_CPP=${PARQUET_CPP:-~/src/parquet-cpp} + +# make obj file +clean() { + rm -f *.o *.so +} + +build() { + clean + g++ -std=c++11 -Wall -fPIC -c -g parquet.cc -I ../sqlite -lz -ldl -lpthread + + g++ -shared -o libparquet.so parquet.o \ + ${PARQUET_CPP}/build/release/libparquet.a \ + ${PARQUET_CPP}/thrift_ep/src/thrift_ep-install/lib/libthrift.a \ + ${PARQUET_CPP}/build/release/libarrow.so \ + /usr/lib/x86_64-linux-gnu/libboost_regex.so +} + +fn=${1:-build} + +"$fn" diff --git a/parquet/parquet.cc b/parquet/parquet.cc new file mode 100644 index 0000000..bed9088 --- /dev/null +++ b/parquet/parquet.cc @@ -0,0 +1,909 @@ +/* +** 2016-05-28 +** +** 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 file contains the implementation of an SQLite virtual table for +** reading CSV files. +** +** Usage: +** +** .load ./csv +** CREATE VIRTUAL TABLE temp.csv USING csv(filename=FILENAME); +** SELECT * FROM csv; +** +** The columns are named "c1", "c2", "c3", ... by default. But the +** application can define its own CREATE TABLE statement as an additional +** parameter. For example: +** +** CREATE VIRTUAL TABLE temp.csv2 USING csv( +** filename = "../http.log", +** schema = "CREATE TABLE x(date,ipaddr,url,referrer,userAgent)" +** ); +** +** Instead of specifying a file, the text of the CSV can be loaded using +** the data= parameter. +** +** If the columns=N parameter is supplied, then the CSV file is assumed to have +** N columns. If the columns parameter is omitted, the CSV file is opened +** as soon as the virtual table is constructed and the first row of the CSV +** is read in order to count the tables. +** +** Some extra debugging features (used for testing virtual tables) are available +** if this module is compiled with -DSQLITE_TEST. +*/ +#include +SQLITE_EXTENSION_INIT1 +#include +#include +#include +#include +#include +#include + +#include "parquet/api/reader.h" + +char const *EMPTY_STRING = ""; +#ifndef SQLITE_OMIT_VIRTUALTABLE + +/* +** A macro to hint to the compiler that a function should not be +** inlined. +*/ +#if defined(__GNUC__) +# define CSV_NOINLINE __attribute__((noinline)) +#elif defined(_MSC_VER) && _MSC_VER>=1310 +# define CSV_NOINLINE __declspec(noinline) +#else +# define CSV_NOINLINE +#endif + +void gogo() { + printf("ok"); + try { + std::unique_ptr reader = + parquet::ParquetFileReader::OpenFile("/home/cldellow/src/csv2parquet/12m.parquet.snappy"); + printf("%d\n", reader->metadata()->size()); + printf("%ld\n", reader->metadata()->num_rows()); + } catch (const std::exception& e) { + std::cerr << "Parquet error: " << e.what() << std::endl; + } +} + +/* Max size of the error message in a CsvReader */ +#define CSV_MXERR 200 + +/* Size of the CsvReader input buffer */ +#define CSV_INBUFSZ 1024 + +/* A context object used when read a CSV file. */ +typedef struct CsvReader CsvReader; +struct CsvReader { + FILE *in; /* Read the CSV text from this input stream */ + char *z; /* Accumulated text for a field */ + int n; /* Number of bytes in z */ + int nAlloc; /* Space allocated for z[] */ + int nLine; /* Current line number */ + int bNotFirst; /* True if prior text has been seen */ + int cTerm; /* Character that terminated the most recent field */ + size_t iIn; /* Next unread character in the input buffer */ + size_t nIn; /* Number of characters in the input buffer */ + char *zIn; /* The input buffer */ + char zErr[CSV_MXERR]; /* Error message */ +}; + +/* Initialize a CsvReader object */ +static void csv_reader_init(CsvReader *p){ + p->in = 0; + p->z = 0; + p->n = 0; + p->nAlloc = 0; + p->nLine = 0; + p->bNotFirst = 0; + p->nIn = 0; + p->zIn = 0; + p->zErr[0] = 0; +} + +/* Close and reset a CsvReader object */ +static void csv_reader_reset(CsvReader *p){ + if( p->in ){ + fclose(p->in); + sqlite3_free(p->zIn); + } + sqlite3_free(p->z); + csv_reader_init(p); +} + +/* Report an error on a CsvReader */ +static void csv_errmsg(CsvReader *p, const char *zFormat, ...){ + va_list ap; + va_start(ap, zFormat); + sqlite3_vsnprintf(CSV_MXERR, p->zErr, zFormat, ap); + va_end(ap); +} + +/* Open the file associated with a CsvReader +** Return the number of errors. +*/ +static int csv_reader_open( + CsvReader *p, /* The reader to open */ + const char *zFilename, /* Read from this filename */ + const char *zData /* ... or use this data */ +){ + if( zFilename ){ + p->zIn = (char*)sqlite3_malloc( CSV_INBUFSZ ); + if( p->zIn==0 ){ + csv_errmsg(p, "out of memory"); + return 1; + } + p->in = fopen(zFilename, "rb"); + if( p->in==0 ){ + csv_reader_reset(p); + csv_errmsg(p, "cannot open '%s' for reading", zFilename); + return 1; + } + }else{ + assert( p->in==0 ); + p->zIn = (char*)zData; + p->nIn = strlen(zData); + } + return 0; +} + +/* The input buffer has overflowed. Refill the input buffer, then +** return the next character +*/ +static CSV_NOINLINE int csv_getc_refill(CsvReader *p){ + size_t got; + + assert( p->iIn>=p->nIn ); /* Only called on an empty input buffer */ + assert( p->in!=0 ); /* Only called if reading froma file */ + + got = fread(p->zIn, 1, CSV_INBUFSZ, p->in); + if( got==0 ) return EOF; + p->nIn = got; + p->iIn = 1; + return p->zIn[0]; +} + +/* Return the next character of input. Return EOF at end of input. */ +static int csv_getc(CsvReader *p){ + if( p->iIn >= p->nIn ){ + if( p->in!=0 ) return csv_getc_refill(p); + return EOF; + } + return ((unsigned char*)p->zIn)[p->iIn++]; +} + +/* Increase the size of p->z and append character c to the end. +** Return 0 on success and non-zero if there is an OOM error */ +static CSV_NOINLINE int csv_resize_and_append(CsvReader *p, char c){ + char *zNew; + int nNew = p->nAlloc*2 + 100; + zNew = (char*)sqlite3_realloc64(p->z, nNew); + if( zNew ){ + p->z = zNew; + p->nAlloc = nNew; + p->z[p->n++] = c; + return 0; + }else{ + csv_errmsg(p, "out of memory"); + return 1; + } +} + +/* Append a single character to the CsvReader.z[] array. +** Return 0 on success and non-zero if there is an OOM error */ +static int csv_append(CsvReader *p, char c){ + if( p->n>=p->nAlloc-1 ) return csv_resize_and_append(p, c); + p->z[p->n++] = c; + return 0; +} + +/* Read a single field of CSV text. Compatible with rfc4180 and extended +** with the option of having a separator other than ",". +** +** + Input comes from p->in. +** + Store results in p->z of length p->n. Space to hold p->z comes +** from sqlite3_malloc64(). +** + Keep track of the line number in p->nLine. +** + Store the character that terminates the field in p->cTerm. Store +** EOF on end-of-file. +** +** Return "" at EOF. Return 0 on an OOM error. +*/ +static const char *csv_read_one_field(CsvReader *p){ + int c; + p->n = 0; + c = csv_getc(p); + if( c==EOF ){ + p->cTerm = EOF; + return EMPTY_STRING; + } + if( c=='"' ){ + int pc, ppc; + int startLine = p->nLine; + pc = ppc = 0; + while( 1 ){ + c = csv_getc(p); + if( c<='"' || pc=='"' ){ + if( c=='\n' ) p->nLine++; + if( c=='"' ){ + if( pc=='"' ){ + pc = 0; + continue; + } + } + if( (c==',' && pc=='"') + || (c=='\n' && pc=='"') + || (c=='\n' && pc=='\r' && ppc=='"') + || (c==EOF && pc=='"') + ){ + do{ p->n--; }while( p->z[p->n]!='"' ); + p->cTerm = (char)c; + break; + } + if( pc=='"' && c!='\r' ){ + csv_errmsg(p, "line %d: unescaped %c character", p->nLine, '"'); + break; + } + if( c==EOF ){ + csv_errmsg(p, "line %d: unterminated %c-quoted field\n", + startLine, '"'); + p->cTerm = (char)c; + break; + } + } + if( csv_append(p, (char)c) ) return 0; + ppc = pc; + pc = c; + } + }else{ + /* If this is the first field being parsed and it begins with the + ** UTF-8 BOM (0xEF BB BF) then skip the BOM */ + if( (c&0xff)==0xef && p->bNotFirst==0 ){ + csv_append(p, (char)c); + c = csv_getc(p); + if( (c&0xff)==0xbb ){ + csv_append(p, (char)c); + c = csv_getc(p); + if( (c&0xff)==0xbf ){ + p->bNotFirst = 1; + p->n = 0; + return csv_read_one_field(p); + } + } + } + while( c>',' || (c!=EOF && c!=',' && c!='\n') ){ + if( csv_append(p, (char)c) ) return 0; + c = csv_getc(p); + } + if( c=='\n' ){ + p->nLine++; + if( p->n>0 && p->z[p->n-1]=='\r' ) p->n--; + } + p->cTerm = (char)c; + } + if( p->z ) p->z[p->n] = 0; + p->bNotFirst = 1; + return p->z; +} + + +/* Forward references to the various virtual table methods implemented +** in this file. */ +static int csvtabCreate(sqlite3*, void*, int, const char*const*, + sqlite3_vtab**,char**); +static int csvtabConnect(sqlite3*, void*, int, const char*const*, + sqlite3_vtab**,char**); +static int csvtabBestIndex(sqlite3_vtab*,sqlite3_index_info*); +static int csvtabDisconnect(sqlite3_vtab*); +static int csvtabOpen(sqlite3_vtab*, sqlite3_vtab_cursor**); +static int csvtabClose(sqlite3_vtab_cursor*); +static int csvtabFilter(sqlite3_vtab_cursor*, int idxNum, const char *idxStr, + int argc, sqlite3_value **argv); +static int csvtabNext(sqlite3_vtab_cursor*); +static int csvtabEof(sqlite3_vtab_cursor*); +static int csvtabColumn(sqlite3_vtab_cursor*,sqlite3_context*,int); +static int csvtabRowid(sqlite3_vtab_cursor*,sqlite3_int64*); + +/* An instance of the CSV virtual table */ +typedef struct CsvTable { + sqlite3_vtab base; /* Base class. Must be first */ + char *zFilename; /* Name of the CSV file */ + char *zData; /* Raw CSV data in lieu of zFilename */ + long iStart; /* Offset to start of data in zFilename */ + unsigned int nCol; /* Number of columns in the CSV file */ + unsigned int tstFlags; /* Bit values used for testing */ +} CsvTable; + +/* Allowed values for tstFlags */ +#define CSVTEST_FIDX 0x0001 /* Pretend that constrained searchs cost less*/ + +/* A cursor for the CSV virtual table */ +typedef struct CsvCursor { + sqlite3_vtab_cursor base; /* Base class. Must be first */ + CsvReader rdr; /* The CsvReader object */ + char **azVal; /* Value of the current row */ + int *aLen; /* Length of each entry */ + sqlite3_int64 iRowid; /* The current rowid. Negative for EOF */ +} CsvCursor; + +/* Transfer error message text from a reader into a CsvTable */ +static void csv_xfer_error(CsvTable *pTab, CsvReader *pRdr){ + sqlite3_free(pTab->base.zErrMsg); + pTab->base.zErrMsg = sqlite3_mprintf("%s", pRdr->zErr); +} + +/* +** This method is the destructor fo a CsvTable object. +*/ +static int csvtabDisconnect(sqlite3_vtab *pVtab){ + CsvTable *p = (CsvTable*)pVtab; + sqlite3_free(p->zFilename); + sqlite3_free(p->zData); + sqlite3_free(p); + return SQLITE_OK; +} + +/* Skip leading whitespace. Return a pointer to the first non-whitespace +** character, or to the zero terminator if the string has only whitespace */ +static const char *csv_skip_whitespace(const char *z){ + while( isspace((unsigned char)z[0]) ) z++; + return z; +} + +/* Remove trailing whitespace from the end of string z[] */ +static void csv_trim_whitespace(char *z){ + size_t n = strlen(z); + while( n>0 && isspace((unsigned char)z[n]) ) n--; + z[n] = 0; +} + +/* Dequote the string */ +static void csv_dequote(char *z){ + int j; + char cQuote = z[0]; + size_t i, n; + + if( cQuote!='\'' && cQuote!='"' ) return; + n = strlen(z); + if( n<2 || z[n-1]!=z[0] ) return; + for(i=1, j=0; izErr. If there are no errors, p->zErr[0]==0. +*/ +static int csv_string_parameter( + CsvReader *p, /* Leave the error message here, if there is one */ + const char *zParam, /* Parameter we are checking for */ + const char *zArg, /* Raw text of the virtual table argment */ + char **pzVal /* Write the dequoted string value here */ +){ + const char *zValue; + zValue = csv_parameter(zParam,(int)strlen(zParam),zArg); + if( zValue==0 ) return 0; + p->zErr[0] = 0; + if( *pzVal ){ + csv_errmsg(p, "more than one '%s' parameter", zParam); + return 1; + } + *pzVal = sqlite3_mprintf("%s", zValue); + if( *pzVal==0 ){ + csv_errmsg(p, "out of memory"); + return 1; + } + csv_trim_whitespace(*pzVal); + csv_dequote(*pzVal); + return 1; +} + + +/* Return 0 if the argument is false and 1 if it is true. Return -1 if +** we cannot really tell. +*/ +static int csv_boolean(const char *z){ + if( sqlite3_stricmp("yes",z)==0 + || sqlite3_stricmp("on",z)==0 + || sqlite3_stricmp("true",z)==0 + || (z[0]=='1' && z[1]==0) + ){ + return 1; + } + if( sqlite3_stricmp("no",z)==0 + || sqlite3_stricmp("off",z)==0 + || sqlite3_stricmp("false",z)==0 + || (z[0]=='0' && z[1]==0) + ){ + return 0; + } + return -1; +} + + +/* +** Parameters: +** filename=FILENAME Name of file containing CSV content +** data=TEXT Direct CSV content. +** schema=SCHEMA Alternative CSV schema. +** header=YES|NO First row of CSV defines the names of +** columns if "yes". Default "no". +** columns=N Assume the CSV file contains N columns. +** +** Only available if compiled with SQLITE_TEST: +** +** testflags=N Bitmask of test flags. Optional +** +** If schema= is omitted, then the columns are named "c0", "c1", "c2", +** and so forth. If columns=N is omitted, then the file is opened and +** the number of columns in the first row is counted to determine the +** column count. If header=YES, then the first row is skipped. +*/ +static int csvtabConnect( + sqlite3 *db, + void *pAux, + int argcOrig, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + unsigned int argc = argcOrig; + CsvTable *pNew = 0; /* The CsvTable object to construct */ + int bHeader = -1; /* header= flags. -1 means not seen yet */ + int rc = SQLITE_OK; /* Result code from this routine */ + unsigned int i, j; /* Loop counters */ +#ifdef SQLITE_TEST + int tstFlags = 0; /* Value for testflags=N parameter */ +#endif + int nCol = -99; /* Value of the columns= parameter */ + CsvReader sRdr; /* A CSV file reader used to store an error + ** message and/or to count the number of columns */ + static const char *azParam[] = { + "filename", "data", "schema", + }; + char *azPValue[3]; /* Parameter values */ +# define CSV_FILENAME (azPValue[0]) +# define CSV_DATA (azPValue[1]) +# define CSV_SCHEMA (azPValue[2]) + + + assert( sizeof(azPValue)==sizeof(azParam) ); + memset(&sRdr, 0, sizeof(sRdr)); + memset(azPValue, 0, sizeof(azPValue)); + for(i=3; i=0 ){ + csv_errmsg(&sRdr, "more than one 'header' parameter"); + goto csvtab_connect_error; + } + x = csv_boolean(zValue); + if( x==1 ){ + bHeader = 1; + }else if( x==0 ){ + bHeader = 0; + }else{ + csv_errmsg(&sRdr, "unrecognized argument to 'header': %s", zValue); + goto csvtab_connect_error; + } + }else +#ifdef SQLITE_TEST + if( (zValue = csv_parameter("testflags",9,z))!=0 ){ + tstFlags = (unsigned int)atoi(zValue); + }else +#endif + if( (zValue = csv_parameter("columns",7,z))!=0 ){ + if( nCol>0 ){ + csv_errmsg(&sRdr, "more than one 'columns' parameter"); + goto csvtab_connect_error; + } + nCol = atoi(zValue); + if( nCol<=0 ){ + csv_errmsg(&sRdr, "must have at least one column"); + goto csvtab_connect_error; + } + }else + { + csv_errmsg(&sRdr, "unrecognized parameter '%s'", z); + goto csvtab_connect_error; + } + } + if( (CSV_FILENAME==0)==(CSV_DATA==0) ){ + csv_errmsg(&sRdr, "must either filename= or data= but not both"); + goto csvtab_connect_error; + } + if( nCol<=0 && csv_reader_open(&sRdr, CSV_FILENAME, CSV_DATA) ){ + goto csvtab_connect_error; + } + pNew = (CsvTable*)sqlite3_malloc( sizeof(*pNew) ); + *ppVtab = (sqlite3_vtab*)pNew; + if( pNew==0 ) goto csvtab_connect_oom; + memset(pNew, 0, sizeof(*pNew)); + if( nCol>0 ){ + pNew->nCol = nCol; + }else{ + do{ + const char *z = csv_read_one_field(&sRdr); + if( z==0 ) goto csvtab_connect_oom; + pNew->nCol++; + }while( sRdr.cTerm==',' ); + } + pNew->zFilename = CSV_FILENAME; CSV_FILENAME = 0; + pNew->zData = CSV_DATA; CSV_DATA = 0; +#ifdef SQLITE_TEST + pNew->tstFlags = tstFlags; +#endif + pNew->iStart = bHeader==1 ? ftell(sRdr.in) : 0; + csv_reader_reset(&sRdr); + if( CSV_SCHEMA==0 ){ + const char *zSep = EMPTY_STRING; + CSV_SCHEMA = sqlite3_mprintf("CREATE TABLE x("); + if( CSV_SCHEMA==0 ) goto csvtab_connect_oom; + for(i=0; inCol; i++){ + CSV_SCHEMA = sqlite3_mprintf("%z%sc%d TEXT",CSV_SCHEMA, zSep, i); + zSep = ","; + } + CSV_SCHEMA = sqlite3_mprintf("%z);", CSV_SCHEMA); + } + rc = sqlite3_declare_vtab(db, CSV_SCHEMA); + if( rc ) goto csvtab_connect_error; + for(i=0; ibase); + for(i=0; ibase.pVtab; + unsigned int i; + for(i=0; inCol; i++){ + sqlite3_free(pCur->azVal[i]); + pCur->azVal[i] = 0; + pCur->aLen[i] = 0; + } +} + +/* +** The xConnect and xCreate methods do the same thing, but they must be +** different so that the virtual table is not an eponymous virtual table. +*/ +static int csvtabCreate( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + return csvtabConnect(db, pAux, argc, argv, ppVtab, pzErr); +} + +/* +** Destructor for a CsvCursor. +*/ +static int csvtabClose(sqlite3_vtab_cursor *cur){ + CsvCursor *pCur = (CsvCursor*)cur; + csvtabCursorRowReset(pCur); + csv_reader_reset(&pCur->rdr); + sqlite3_free(cur); + return SQLITE_OK; +} + +/* +** Constructor for a new CsvTable cursor object. +*/ +static int csvtabOpen(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){ + CsvTable *pTab = (CsvTable*)p; + CsvCursor *pCur; + size_t nByte; + nByte = sizeof(*pCur) + (sizeof(char*)+sizeof(int))*pTab->nCol; + pCur = (CsvCursor*)sqlite3_malloc64( nByte ); + if( pCur==0 ) return SQLITE_NOMEM; + memset(pCur, 0, nByte); + pCur->azVal = (char**)&pCur[1]; + pCur->aLen = (int*)&pCur->azVal[pTab->nCol]; + *ppCursor = &pCur->base; + if( csv_reader_open(&pCur->rdr, pTab->zFilename, pTab->zData) ){ + csv_xfer_error(pTab, &pCur->rdr); + return SQLITE_ERROR; + } + return SQLITE_OK; +} + + +/* +** Advance a CsvCursor to its next row of input. +** Set the EOF marker if we reach the end of input. +*/ +static int csvtabNext(sqlite3_vtab_cursor *cur){ + CsvCursor *pCur = (CsvCursor*)cur; + CsvTable *pTab = (CsvTable*)cur->pVtab; + unsigned int i = 0; + const char *z; + do{ + z = csv_read_one_field(&pCur->rdr); + if( z==0 ){ + csv_xfer_error(pTab, &pCur->rdr); + break; + } + if( inCol ){ + if( pCur->aLen[i] < pCur->rdr.n+1 ){ + char *zNew = (char*)sqlite3_realloc64(pCur->azVal[i], pCur->rdr.n+1); + if( zNew==0 ){ + csv_errmsg(&pCur->rdr, "out of memory"); + csv_xfer_error(pTab, &pCur->rdr); + break; + } + pCur->azVal[i] = zNew; + pCur->aLen[i] = pCur->rdr.n+1; + } + memcpy(pCur->azVal[i], z, pCur->rdr.n+1); + i++; + } + }while( pCur->rdr.cTerm==',' ); + if( z==0 || (pCur->rdr.cTerm==EOF && inCol) ){ + pCur->iRowid = -1; + }else{ + pCur->iRowid++; + while( inCol ){ + sqlite3_free(pCur->azVal[i]); + pCur->azVal[i] = 0; + pCur->aLen[i] = 0; + i++; + } + } + return SQLITE_OK; +} + +/* +** Return values of columns for the row at which the CsvCursor +** is currently pointing. +*/ +static int csvtabColumn( + sqlite3_vtab_cursor *cur, /* The cursor */ + sqlite3_context *ctx, /* First argument to sqlite3_result_...() */ + int iOrig /* Which column to return */ +){ + unsigned int i = iOrig; + CsvCursor *pCur = (CsvCursor*)cur; + CsvTable *pTab = (CsvTable*)cur->pVtab; + if( i>=0 && inCol && pCur->azVal[i]!=0 ){ + sqlite3_result_text(ctx, pCur->azVal[i], -1, SQLITE_STATIC); + } + return SQLITE_OK; +} + +/* +** Return the rowid for the current row. +*/ +static int csvtabRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ + CsvCursor *pCur = (CsvCursor*)cur; + *pRowid = pCur->iRowid; + return SQLITE_OK; +} + +/* +** Return TRUE if the cursor has been moved off of the last +** row of output. +*/ +static int csvtabEof(sqlite3_vtab_cursor *cur){ + CsvCursor *pCur = (CsvCursor*)cur; + return pCur->iRowid<0; +} + +/* +** Only a full table scan is supported. So xFilter simply rewinds to +** the beginning. +*/ +static int csvtabFilter( + sqlite3_vtab_cursor *pVtabCursor, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + CsvCursor *pCur = (CsvCursor*)pVtabCursor; + CsvTable *pTab = (CsvTable*)pVtabCursor->pVtab; + pCur->iRowid = 0; + if( pCur->rdr.in==0 ){ + assert( pCur->rdr.zIn==pTab->zData ); + assert( pTab->iStart>=0 ); + assert( (size_t)pTab->iStart<=pCur->rdr.nIn ); + pCur->rdr.iIn = pTab->iStart; + }else{ + fseek(pCur->rdr.in, pTab->iStart, SEEK_SET); + pCur->rdr.iIn = 0; + pCur->rdr.nIn = 0; + } + return csvtabNext(pVtabCursor); +} + +/* +** Only a forward full table scan is supported. xBestIndex is mostly +** a no-op. If CSVTEST_FIDX is set, then the presence of equality +** constraints lowers the estimated cost, which is fiction, but is useful +** for testing certain kinds of virtual table behavior. +*/ +static int csvtabBestIndex( + sqlite3_vtab *tab, + sqlite3_index_info *pIdxInfo +){ + pIdxInfo->estimatedCost = 1000000; +#ifdef SQLITE_TEST + if( (((CsvTable*)tab)->tstFlags & CSVTEST_FIDX)!=0 ){ + /* The usual (and sensible) case is to always do a full table scan. + ** The code in this branch only runs when testflags=1. This code + ** generates an artifical and unrealistic plan which is useful + ** for testing virtual table logic but is not helpful to real applications. + ** + ** Any ==, LIKE, or GLOB constraint is marked as usable by the virtual + ** table (even though it is not) and the cost of running the virtual table + ** is reduced from 1 million to just 10. The constraints are *not* marked + ** as omittable, however, so the query planner should still generate a + ** plan that gives a correct answer, even if they plan is not optimal. + */ + int i; + int nConst = 0; + for(i=0; inConstraint; i++){ + unsigned char op; + if( pIdxInfo->aConstraint[i].usable==0 ) continue; + op = pIdxInfo->aConstraint[i].op; + if( op==SQLITE_INDEX_CONSTRAINT_EQ + || op==SQLITE_INDEX_CONSTRAINT_LIKE + || op==SQLITE_INDEX_CONSTRAINT_GLOB + ){ + pIdxInfo->estimatedCost = 10; + pIdxInfo->aConstraintUsage[nConst].argvIndex = nConst+1; + nConst++; + } + } + } +#endif + return SQLITE_OK; +} + + +static sqlite3_module CsvModule = { + 0, /* iVersion */ + csvtabCreate, /* xCreate */ + csvtabConnect, /* xConnect */ + csvtabBestIndex, /* xBestIndex */ + csvtabDisconnect, /* xDisconnect */ + csvtabDisconnect, /* xDestroy */ + csvtabOpen, /* xOpen - open a cursor */ + csvtabClose, /* xClose - close a cursor */ + csvtabFilter, /* xFilter - configure scan constraints */ + csvtabNext, /* xNext - advance a cursor */ + csvtabEof, /* xEof - check for end of scan */ + csvtabColumn, /* xColumn - read data */ + csvtabRowid, /* xRowid - read data */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ +}; + +#ifdef SQLITE_TEST +/* +** For virtual table testing, make a version of the CSV virtual table +** available that has an xUpdate function. But the xUpdate always returns +** SQLITE_READONLY since the CSV file is not really writable. +*/ +static int csvtabUpdate(sqlite3_vtab *p,int n,sqlite3_value**v,sqlite3_int64*x){ + return SQLITE_READONLY; +} +static sqlite3_module CsvModuleFauxWrite = { + 0, /* iVersion */ + csvtabCreate, /* xCreate */ + csvtabConnect, /* xConnect */ + csvtabBestIndex, /* xBestIndex */ + csvtabDisconnect, /* xDisconnect */ + csvtabDisconnect, /* xDestroy */ + csvtabOpen, /* xOpen - open a cursor */ + csvtabClose, /* xClose - close a cursor */ + csvtabFilter, /* xFilter - configure scan constraints */ + csvtabNext, /* xNext - advance a cursor */ + csvtabEof, /* xEof - check for end of scan */ + csvtabColumn, /* xColumn - read data */ + csvtabRowid, /* xRowid - read data */ + csvtabUpdate, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ +}; +#endif /* SQLITE_TEST */ + +#endif /* !defined(SQLITE_OMIT_VIRTUALTABLE) */ + + +#ifdef _WIN32 +__declspec(dllexport) +#endif +/* +** This routine is called when the extension is loaded. The new +** CSV virtual table module is registered with the calling database +** connection. +*/ +extern "C" { + int sqlite3_parquet_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi + ){ +#ifndef SQLITE_OMIT_VIRTUALTABLE + int rc; + SQLITE_EXTENSION_INIT2(pApi); + printf("foo!!!\n"); + gogo(); + rc = sqlite3_create_module(db, "parquet", &CsvModule, 0); +#ifdef SQLITE_TEST + if( rc==SQLITE_OK ){ + rc = sqlite3_create_module(db, "csv_wr", &CsvModuleFauxWrite, 0); + } +#endif + return rc; +#else + return SQLITE_OK; +#endif + } +} + +