SQLite

OVERVIEW

  The SQLite library is capable of parsing SQL foreign key constraints 
  supplied as part of CREATE TABLE statements, but it does not actually 
  implement them. However, most of the features of foreign keys may be
  implemented using SQL triggers, which SQLite does support. This program 
  extracts foreign key definitions from an existing SQLite database and
  outputs the set of CREATE TRIGGER statements required to implement
  the foreign key constraints.

CAPABILITIES

  An SQL foreign key is a constraint that requires that each row in
  the "child" table corresponds to a row in the "parent" table. For
  example, the following schema:

    CREATE TABLE parent(a, b, c, PRIMARY KEY(a, b));
    CREATE TABLE child(d, e, f, FOREIGN KEY(d, e) REFERENCES parent(a, b));

  implies that for each row in table "child", there must be a row in
  "parent" for which the expression (child.d==parent.a AND child.e==parent.b) 
  is true. The columns in the parent table are required to be either the
  primary key columns or subject to a UNIQUE constraint. There is no such
  requirement for the columns of the child table.

  At this time, all foreign keys are implemented as if they were 
  "MATCH NONE", even if the declaration specified "MATCH PARTIAL" or 
  "MATCH FULL". "MATCH NONE" means that if any of the key columns in
  the child table are NULL, then there is no requirement for a corresponding
  row in the parent table. So, taking this into account, the expression that
  must be true for every row of the child table in the above example is
  actually:

      (child.d IS NULL) OR 
      (child.e IS NULL) OR 
      (child.d==parent.a AND child.e==parent.b)

  Attempting to insert or update a row in the child table so that the 
  affected row violates this constraint results in an exception being 
  thrown.

  The effect of attempting to delete or update a row in the parent table 
  so that the constraint becomes untrue for one or more rows in the child
  table depends on the "ON DELETE" or "ON UPDATE" actions specified as
  part of the foreign key definition, respectively. Three different actions
  are supported: "RESTRICT" (the default), "CASCADE" and "SET NULL". SQLite
  will also parse the "SET DEFAULT" action, but this is not implemented
  and "RESTRICT" is used instead.

    RESTRICT: Attempting to update or delete a row in the parent table so
              that the constraint becomes untrue for one or more rows in
              the child table is not allowed. An exception is thrown.

    CASCADE:  Instead of throwing an exception, all corresponding child table
              rows are either deleted (if the parent row is being deleted)
              or updated to match the new parent key values (if the parent 
              row is being updated).

    SET NULL: Instead of throwing an exception, the foreign key fields of
              all corresponding child table rows are set to NULL.

LIMITATIONS

  Apart from those limitiations described above:

    * Implicit mapping to composite primary keys is not supported. If
      a parent table has a composite primary key, then any child table
      that refers to it must explicitly map each column. For example, given
      the following definition of table "parent":

        CREATE TABLE parent(a, b, c, PRIMARY KEY(a, b));

      only the first of the following two definitions of table "child"
      is supported:

        CREATE TABLE child(d, e, f, FOREIGN KEY(d, e) REFERENCES parent(a, b));
        CREATE TABLE child(d, e, f, FOREIGN KEY(d, e) REFERENCES parent);

      An implicit reference to a composite primary key is detected as an
      error when the program is run (see below).

    * SQLite does not support recursive triggers, and therefore this program
      does not support recursive CASCADE or SET NULL foreign key 
      relationships. If the parent and the child tables of a CASCADE or
      SET NULL foreign key are the same table, the generated triggers will
      malfunction. This is also true if the recursive foreign key constraint
      is indirect (for example if table A references table B which references
      table A with a CASCADE or SET NULL foreign key constraint).

      Recursive CASCADE or SET NULL foreign key relationships are *not*
      detected as errors when the program is run. Buyer beware.
      
COMPILATION

  The source code for this program consists of a single C file - genfkey.c.
  The only dependency is sqlite itself. Using gcc and the sqlite amalgamation
  source code, it may be compiled using the following command:

    gcc genfkey.c sqlite3.c -o genfkey

  If compiled/linked against an SQLite version earlier than 3.6.4, then
  all foreign key constraints are assumed to be "ON UPDATE RESTRICT" and
  "ON DELETE RESTRICT". If linked against 3.6.4 or newer, "CASCADE" and
  "SET NULL" are supported as well as "RESTRICT". All 3.x versions of SQLite
  may use the created triggers definitions.

USAGE

    genfkey ?--no-drop? ?--ignore-errors? <sqlite database>

  When this program is run, it first checks the schema of the supplied SQLite
  database for foreign key related errors or inconsistencies. For example,
  a foreign key that refers to a parent table that does not exist, or
  a foreign key that refers to columns in a parent table that are not
  guaranteed to be unique. If such errors are found, a message for each
  one is printed to stderr.

  If errors are found and the --ignore-errors option was not passed, the
  program exits. Otherwise, a series of SQL trigger definitions (CREATE 
  TRIGGER statements) that implement the foreign key constraints found
  in the database schema are written to stdout. If any errors were 
  found in the schema, no triggers for the problematic constraints are
  output. The output CREATE TRIGGER statements should be run against the
  database to enable enforcement of the foreign key constraints. For
  example, for a database named "test.db" in the current working directory:

    $ genfkey ./test.db | sqlite3 ./test.db

  All triggers generated by this program have names that match the pattern
  "genfkey*". Unless the --no-drop option is specified, then the program
  also outputs a "DROP TRIGGER" statement for each trigger that exists
  in the database with a name that matches this pattern. This allows the
  program to be used to upgrade a database schema for which foreign key
  triggers have already been installed (i.e. after new tables are created
  or existing tables dropped).
  

/*
** 2008 October 10
**
** 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 C code for 'genfkey', a program to generate trigger
** definitions that emulate foreign keys. See genfkey.README for details.
**
** $Id: genfkey.c,v 1.1 2008/10/10 17:58:27 danielk1977 Exp $
*/

#include "sqlite3.h"
#include <stdio.h>
#include <assert.h>
#include <stdlib.h>
#include <string.h>

/**************************************************************************
***************************************************************************
** Start of virtual table implementations.
**************************************************************************/

/* The code in this file defines a sqlite3 virtual-table module that
** provides a read-only view of the current database schema. There is one
** row in the schema table for each column in the database schema.
*/
#define SCHEMA \
"CREATE TABLE x("                                                            \
  "database,"          /* Name of database (i.e. main, temp etc.) */         \
  "tablename,"         /* Name of table */                                   \
  "cid,"               /* Column number (from left-to-right, 0 upward) */    \
  "name,"              /* Column name */                                     \
  "type,"              /* Specified type (i.e. VARCHAR(32)) */               \
  "not_null,"          /* Boolean. True if NOT NULL was specified */         \
  "dflt_value,"        /* Default value for this column */                   \
  "pk"                 /* True if this column is part of the primary key */  \
")"

#define SCHEMA2 \
"CREATE TABLE x("                                                            \
  "database,"          /* Name of database (i.e. main, temp etc.) */         \
  "from_tbl,"          /* Name of table */                                   \
  "fkid,"                                                                    \
  "seq,"                                                                     \
  "to_tbl,"                                                                  \
  "from_col,"                                                                \
  "to_col,"                                                                  \
  "on_update,"                                                               \
  "on_delete,"                                                               \
  "match"                                                                    \
")"

#define SCHEMA3 \
"CREATE TABLE x("                                                            \
  "database,"          /* Name of database (i.e. main, temp etc.) */         \
  "tablename,"         /* Name of table */                                   \
  "seq,"                                                                     \
  "name,"                                                                    \
  "isunique"                                                                 \
")"

#define SCHEMA4 \
"CREATE TABLE x("                                                            \
  "database,"          /* Name of database (i.e. main, temp etc.) */         \
  "indexname,"         /* Name of table */                                   \
  "seqno,"                                                                   \
  "cid,"                                                                     \
  "name"                                                                     \
")"

typedef struct SchemaTable SchemaTable;
struct SchemaTable {
  const char *zName;
  const char *zObject;
  const char *zPragma;
  const char *zSchema;
} aSchemaTable[] = {
  { "table_info",       "table", "PRAGMA %Q.table_info(%Q)",       SCHEMA },
  { "foreign_key_list", "table", "PRAGMA %Q.foreign_key_list(%Q)", SCHEMA2 },
  { "index_list",       "table", "PRAGMA %Q.index_list(%Q)",       SCHEMA3 },
  { "index_info",       "index", "PRAGMA %Q.index_info(%Q)",       SCHEMA4 },
  { 0, 0, 0, 0 }
};

typedef struct schema_vtab schema_vtab;
typedef struct schema_cursor schema_cursor;

/* A schema table object */
struct schema_vtab {
  sqlite3_vtab base;
  sqlite3 *db;
  SchemaTable *pType;
};

/* A schema table cursor object */
struct schema_cursor {
  sqlite3_vtab_cursor base;
  sqlite3_stmt *pDbList;
  sqlite3_stmt *pTableList;
  sqlite3_stmt *pColumnList;
  int rowid;
};

/*
** Table destructor for the schema module.
*/
static int schemaDestroy(sqlite3_vtab *pVtab){
  sqlite3_free(pVtab);
  return 0;
}

/*
** Table constructor for the schema module.
*/
static int schemaCreate(
  sqlite3 *db,
  void *pAux,
  int argc, const char *const*argv,
  sqlite3_vtab **ppVtab,
  char **pzErr
){
  int rc = SQLITE_NOMEM;
  schema_vtab *pVtab;
  SchemaTable *pType = &aSchemaTable[0];

  if( argc>3 ){
    int i;
    pType = 0;
    for(i=0; aSchemaTable[i].zName; i++){ 
      if( 0==strcmp(argv[3], aSchemaTable[i].zName) ){
        pType = &aSchemaTable[i];
      }
    }
    if( !pType ){
      return SQLITE_ERROR;
    }
  }

  pVtab = sqlite3_malloc(sizeof(schema_vtab));
  if( pVtab ){
    memset(pVtab, 0, sizeof(schema_vtab));
    pVtab->db = db;
    pVtab->pType = pType;
    rc = sqlite3_declare_vtab(db, pType->zSchema);
  }
  *ppVtab = (sqlite3_vtab *)pVtab;
  return rc;
}

/*
** Open a new cursor on the schema table.
*/
static int schemaOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
  int rc = SQLITE_NOMEM;
  schema_cursor *pCur;
  pCur = sqlite3_malloc(sizeof(schema_cursor));
  if( pCur ){
    memset(pCur, 0, sizeof(schema_cursor));
    *ppCursor = (sqlite3_vtab_cursor *)pCur;
    rc = SQLITE_OK;
  }
  return rc;
}

/*
** Close a schema table cursor.
*/
static int schemaClose(sqlite3_vtab_cursor *cur){
  schema_cursor *pCur = (schema_cursor *)cur;
  sqlite3_finalize(pCur->pDbList);
  sqlite3_finalize(pCur->pTableList);
  sqlite3_finalize(pCur->pColumnList);
  sqlite3_free(pCur);
  return SQLITE_OK;
}

/*
** Retrieve a column of data.
*/
static int schemaColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){
  schema_cursor *pCur = (schema_cursor *)cur;
  switch( i ){
    case 0:
      sqlite3_result_value(ctx, sqlite3_column_value(pCur->pDbList, 1));
      break;
    case 1:
      sqlite3_result_value(ctx, sqlite3_column_value(pCur->pTableList, 0));
      break;
    default:
      sqlite3_result_value(ctx, sqlite3_column_value(pCur->pColumnList, i-2));
      break;
  }
  return SQLITE_OK;
}

/*
** Retrieve the current rowid.
*/
static int schemaRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
  schema_cursor *pCur = (schema_cursor *)cur;
  *pRowid = pCur->rowid;
  return SQLITE_OK;
}

static int finalize(sqlite3_stmt **ppStmt){
  int rc = sqlite3_finalize(*ppStmt);
  *ppStmt = 0;
  return rc;
}

static int schemaEof(sqlite3_vtab_cursor *cur){
  schema_cursor *pCur = (schema_cursor *)cur;
  return (pCur->pDbList ? 0 : 1);
}

/*
** Advance the cursor to the next row.
*/
static int schemaNext(sqlite3_vtab_cursor *cur){
  int rc = SQLITE_OK;
  schema_cursor *pCur = (schema_cursor *)cur;
  schema_vtab *pVtab = (schema_vtab *)(cur->pVtab);
  char *zSql = 0;

  while( !pCur->pColumnList || SQLITE_ROW!=sqlite3_step(pCur->pColumnList) ){
    if( SQLITE_OK!=(rc = finalize(&pCur->pColumnList)) ) goto next_exit;

    while( !pCur->pTableList || SQLITE_ROW!=sqlite3_step(pCur->pTableList) ){
      if( SQLITE_OK!=(rc = finalize(&pCur->pTableList)) ) goto next_exit;

      assert(pCur->pDbList);
      while( SQLITE_ROW!=sqlite3_step(pCur->pDbList) ){
        rc = finalize(&pCur->pDbList);
        goto next_exit;
      }

      /* Set zSql to the SQL to pull the list of tables from the 
      ** sqlite_master (or sqlite_temp_master) table of the database
      ** identfied by the row pointed to by the SQL statement pCur->pDbList
      ** (iterating through a "PRAGMA database_list;" statement).
      */
      if( sqlite3_column_int(pCur->pDbList, 0)==1 ){
        zSql = sqlite3_mprintf(
            "SELECT name FROM sqlite_temp_master WHERE type=%Q",
            pVtab->pType->zObject
        );
      }else{
        sqlite3_stmt *pDbList = pCur->pDbList;
        zSql = sqlite3_mprintf(
            "SELECT name FROM %Q.sqlite_master WHERE type=%Q",
             sqlite3_column_text(pDbList, 1), pVtab->pType->zObject
        );
      }
      if( !zSql ){
        rc = SQLITE_NOMEM;
        goto next_exit;
      }

      rc = sqlite3_prepare(pVtab->db, zSql, -1, &pCur->pTableList, 0);
      sqlite3_free(zSql);
      if( rc!=SQLITE_OK ) goto next_exit;
    }

    /* Set zSql to the SQL to the table_info pragma for the table currently
    ** identified by the rows pointed to by statements pCur->pDbList and
    ** pCur->pTableList.
    */
    zSql = sqlite3_mprintf(pVtab->pType->zPragma,
        sqlite3_column_text(pCur->pDbList, 1),
        sqlite3_column_text(pCur->pTableList, 0)
    );

    if( !zSql ){
      rc = SQLITE_NOMEM;
      goto next_exit;
    }
    rc = sqlite3_prepare(pVtab->db, zSql, -1, &pCur->pColumnList, 0);
    sqlite3_free(zSql);
    if( rc!=SQLITE_OK ) goto next_exit;
  }
  pCur->rowid++;

next_exit:
  /* TODO: Handle rc */
  return rc;
}

/*
** Reset a schema table cursor.
*/
static int schemaFilter(
  sqlite3_vtab_cursor *pVtabCursor, 
  int idxNum, const char *idxStr,
  int argc, sqlite3_value **argv
){
  int rc;
  schema_vtab *pVtab = (schema_vtab *)(pVtabCursor->pVtab);
  schema_cursor *pCur = (schema_cursor *)pVtabCursor;
  pCur->rowid = 0;
  finalize(&pCur->pTableList);
  finalize(&pCur->pColumnList);
  finalize(&pCur->pDbList);
  rc = sqlite3_prepare(pVtab->db,"SELECT 0, 'main'", -1, &pCur->pDbList, 0);
  return (rc==SQLITE_OK ? schemaNext(pVtabCursor) : rc);
}

/*
** Analyse the WHERE condition.
*/
static int schemaBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){
  return SQLITE_OK;
}

/*
** A virtual table module that merely echos method calls into TCL
** variables.
*/
static sqlite3_module schemaModule = {
  0,                           /* iVersion */
  schemaCreate,
  schemaCreate,
  schemaBestIndex,
  schemaDestroy,
  schemaDestroy,
  schemaOpen,                  /* xOpen - open a cursor */
  schemaClose,                 /* xClose - close a cursor */
  schemaFilter,                /* xFilter - configure scan constraints */
  schemaNext,                  /* xNext - advance a cursor */
  schemaEof,                   /* xEof */
  schemaColumn,                /* xColumn - read data */
  schemaRowid,                 /* xRowid - read data */
  0,                           /* xUpdate */
  0,                           /* xBegin */
  0,                           /* xSync */
  0,                           /* xCommit */
  0,                           /* xRollback */
  0,                           /* xFindMethod */
  0,                           /* xRename */
};

/*
** Extension load function.
*/
static int installSchemaModule(sqlite3 *db){
  sqlite3_create_module(db, "schema", &schemaModule, 0);
  return 0;
}

/**************************************************************************
***************************************************************************
** End of virtual table implementations.
** Start of SQL user function implementations.
*/

/*
**   sj(zValue, zJoin)
**
** The following block contains the implementation of an aggregate 
** function that returns a string. Each time the function is stepped, 
** it appends data to an internal buffer. When the aggregate is finalized,
** the contents of the buffer are returned.
**
** The first time the aggregate is stepped the buffer is set to a copy
** of the first argument. The second time and subsequent times it is
** stepped a copy of the second argument is appended to the buffer, then
** a copy of the first.
**
** Example:
**
**   INSERT INTO t1(a) VALUES('1');
**   INSERT INTO t1(a) VALUES('2');
**   INSERT INTO t1(a) VALUES('3');
**   SELECT sj(a, ', ') FROM t1;
**
**     =>  "1, 2, 3"
**
*/
struct StrBuffer {
  char *zBuf;
};
typedef struct StrBuffer StrBuffer;
static void joinFinalize(sqlite3_context *context){
  StrBuffer *p;
  p = (StrBuffer *)sqlite3_aggregate_context(context, sizeof(StrBuffer));
  sqlite3_result_text(context, p->zBuf, -1, SQLITE_TRANSIENT);
  sqlite3_free(p->zBuf);
}
static void joinStep(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  StrBuffer *p;
  p = (StrBuffer *)sqlite3_aggregate_context(context, sizeof(StrBuffer));
  if( p->zBuf==0 ){
    p->zBuf = sqlite3_mprintf("%s", sqlite3_value_text(argv[0]));
  }else{
    char *zTmp = p->zBuf;
    p->zBuf = sqlite3_mprintf("%s%s%s", 
        zTmp, sqlite3_value_text(argv[1]), sqlite3_value_text(argv[0])
    );
    sqlite3_free(zTmp);
  }
}

/*
**   dq(zString)
**
** This scalar function accepts a single argument and interprets it as
** a text value. The return value is the argument enclosed in double
** quotes. If any double quote characters are present in the argument, 
** these are escaped.
**
**   dq('the raven "Nevermore."') == '"the raven ""Nevermore."""'
*/
static void doublequote(
  sqlite3_context *context, 
  int argc, 
  sqlite3_value **argv
){
  int ii;
  char *zOut;
  char *zCsr;
  const char *zIn = (const char *)sqlite3_value_text(argv[0]);
  int nIn = sqlite3_value_bytes(argv[0]);

  zOut = sqlite3_malloc(nIn*2+3);
  zCsr = zOut;
  *zCsr++ = '"';
  for(ii=0; ii<nIn; ii++){
    *zCsr++ = zIn[ii];
    if( zIn[ii]=='"' ){
      *zCsr++ = '"';
    }
  }
  *zCsr++ = '"';
  *zCsr++ = '\0';

  sqlite3_result_text(context, zOut, -1, SQLITE_TRANSIENT);
  sqlite3_free(zOut);
}

/*
**   multireplace(zString, zSearch1, zReplace1, ...)
*/
static void multireplace(
  sqlite3_context *context, 
  int argc, 
  sqlite3_value **argv
){
  int i = 0;
  char *zOut = 0;
  int nOut = 0;
  int nMalloc = 0;
  const char *zIn = (const char *)sqlite3_value_text(argv[0]);
  int nIn = sqlite3_value_bytes(argv[0]);

  while( i<nIn ){
    const char *zCopy = &zIn[i];
    int nCopy = 1;
    int nReplace = 1;
    int j;
    for(j=1; j<(argc-1); j+=2){
      const char *z = (const char *)sqlite3_value_text(argv[j]);
      int n = sqlite3_value_bytes(argv[j]);
      if( n<=(nIn-i) && 0==strncmp(z, zCopy, n) ){
        zCopy = (const char *)sqlite3_value_text(argv[j+1]);
        nCopy = sqlite3_value_bytes(argv[j+1]);
        nReplace = n;
        break;
      }
    }
    if( (nOut+nCopy)>nMalloc ){
      nMalloc += (nMalloc + 16);
      zOut = (char *)sqlite3_realloc(zOut, nMalloc);
    }
    memcpy(&zOut[nOut], zCopy, nCopy);
    i += nReplace;
    nOut += nCopy;
  }

  sqlite3_result_text(context, zOut, nOut, SQLITE_TRANSIENT);
  sqlite3_free(zOut);
}

/**************************************************************************
***************************************************************************
** End of SQL user function implementations.
** Start of application implementation.
*/

typedef struct Options Options;
struct Options {
  char *zDb;
  int ignoreErrors;
  int noDrop;
};

/*
** Print out a usage message for the command line and exit. This is
** called from processCmdLine() if the program is invoked incorrectly.
*/
static int usage(char *zProgram){
  fprintf(stderr, 
      "Usage: %s ?--ignore-errors? ?--no-drop? <database file>\n", zProgram
  );
  exit(-1);
}

static void processCmdLine(int nArg, char **azArg, Options *p){
  int i;
  assert( nArg>0 );
  if( nArg<2 ){
    usage(azArg[0]);
  }
  for(i=1; i<(nArg-1); i++){
    char *z = azArg[i];
    if( 0==strcmp(z, "--ignore-errors") ){
      p->ignoreErrors = 1;
    }
    else if( 0==strcmp(z, "--no-drop") ){
      p->noDrop = 1;
    }
    else usage(azArg[0]);
  }
  p->zDb = azArg[nArg-1];
}

/*
** A callback for sqlite3_exec() that prints its first argument to
** stdout followed by a newline.
*/
static int printString(void *p, int nArg, char **azArg, char **azCol){
  printf("%s\n", azArg[0]);
  return SQLITE_OK;
}

int detectSchemaProblem(
  sqlite3 *db,                   /* Database connection */
  const char *zMessage,          /* English language error message */
  const char *zSql,              /* SQL statement to run */
  int *pHasErrors                /* Set *pHasErrors==1 if errors found */
){
  sqlite3_stmt *pStmt;
  int rc;
  rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  while( SQLITE_ROW==sqlite3_step(pStmt) ){
    char *zDel;
    int iFk = sqlite3_column_int(pStmt, 0);
    const char *zTab = (const char *)sqlite3_column_text(pStmt, 1);
    fprintf(stderr, "Error in table %s: %s\n", zTab, zMessage);
    zDel = sqlite3_mprintf(
        "DELETE FROM temp.fkey WHERE from_tbl = %Q AND fkid = %d"
        , zTab, iFk
    );
    sqlite3_exec(db, zDel, 0, 0, 0);
    sqlite3_free(zDel);
    *pHasErrors = 1;
  }
  sqlite3_finalize(pStmt);
  return SQLITE_OK;
}

/*
** Create and populate temporary table "fkey".
*/
static int populateTempTable(sqlite3 *db, char **pzErr, int *pHasErrors){
  int rc;

  rc = sqlite3_exec(db, 
      "CREATE VIRTUAL TABLE temp.v_fkey USING schema(foreign_key_list);"
      "CREATE VIRTUAL TABLE temp.v_col USING schema(table_info);"
      "CREATE VIRTUAL TABLE temp.v_idxlist USING schema(index_list);"
      "CREATE VIRTUAL TABLE temp.v_idxinfo USING schema(index_info);"

      "CREATE TABLE temp.fkey AS "
        "SELECT from_tbl, to_tbl, fkid, from_col, to_col, on_update, on_delete "
        "FROM temp.v_fkey WHERE database = 'main';"
      "CREATE TABLE temp.col AS "
        "SELECT * FROM temp.v_col WHERE database = 'main';"

      , 0, 0, pzErr
  );
  if( rc!=SQLITE_OK ) return rc;

  rc = detectSchemaProblem(db, "foreign key columns do not exist",
    "SELECT fkid, from_tbl "
    "FROM temp.fkey "
    "WHERE to_col IS NOT NULL AND NOT EXISTS (SELECT 1 "
        "FROM temp.col WHERE tablename=to_tbl AND name==to_col"
    ")", pHasErrors
  );
  if( rc!=SQLITE_OK ) return rc;

  /* At this point the temp.fkey table is mostly populated. If any foreign
  ** keys were specified so that they implicitly refer to they primary
  ** key of the parent table, the "to_col" values of the temp.fkey rows
  ** are still set to NULL.
  **
  ** This is easily fixed for single column primary keys, but not for
  ** composites. With a composite primary key, there is no way to reliably
  ** query sqlite for the order in which the columns that make up the
  ** composite key were declared i.e. there is no way to tell if the
  ** schema actually contains "PRIMARY KEY(a, b)" or "PRIMARY KEY(b, a)".
  ** Therefore, this case is not handled. The following function call
  ** detects instances of this case.
  */
  rc = detectSchemaProblem(db, "implicit mapping to composite primary key",
    "SELECT fkid, from_tbl "
    "FROM temp.fkey "
    "WHERE to_col IS NULL "
    "GROUP BY fkid, from_tbl HAVING count(*) > 1", pHasErrors
  );
  if( rc!=SQLITE_OK ) return rc;

  /* Detect attempts to implicitly map to the primary key of a table 
  ** that has no primary key column.
  */
  rc = detectSchemaProblem(db, "implicit mapping to non-existant primary key",
    "SELECT fkid, from_tbl "
    "FROM temp.fkey "
    "WHERE to_col IS NULL AND NOT EXISTS "
      "(SELECT 1 FROM temp.col WHERE pk AND tablename = temp.fkey.to_tbl)"
    , pHasErrors
  );
  if( rc!=SQLITE_OK ) return rc;

  /* Fix all the implicit primary key mappings in the temp.fkey table. */
  rc = sqlite3_exec(db, 
    "UPDATE temp.fkey SET to_col = "
      "(SELECT name FROM temp.col WHERE pk AND tablename=temp.fkey.to_tbl)"
    " WHERE to_col IS NULL;"
    , 0, 0, pzErr
  );
  if( rc!=SQLITE_OK ) return rc;

  /* Now check that all all parent keys are either primary keys or 
  ** subject to a unique constraint.
  */
  rc = sqlite3_exec(db, 
    "CREATE TABLE temp.idx2 AS SELECT "
      "il.tablename AS tablename,"
      "ii.indexname AS indexname,"
      "ii.name AS col "
      "FROM temp.v_idxlist AS il, temp.v_idxinfo AS ii "
      "WHERE il.isunique AND il.database='main' AND ii.indexname = il.name;"
    "INSERT INTO temp.idx2 SELECT tablename, 'pk', name FROM temp.col WHERE pk;"

    "CREATE TABLE temp.idx AS SELECT "
      "tablename, indexname, sj(dq(col),',') AS cols "
      "FROM (SELECT * FROM temp.idx2 ORDER BY col) " 
      "GROUP BY tablename, indexname;"

    "CREATE TABLE temp.fkey2 AS SELECT "
        "fkid, from_tbl, to_tbl, sj(dq(to_col),',') AS cols "
        "FROM (SELECT * FROM temp.fkey ORDER BY to_col) " 
        "GROUP BY fkid, from_tbl;"
    , 0, 0, pzErr
  );
  if( rc!=SQLITE_OK ) return rc;
  rc = detectSchemaProblem(db, "foreign key is not unique",
    "SELECT fkid, from_tbl "
    "FROM temp.fkey2 "
    "WHERE NOT EXISTS (SELECT 1 "
        "FROM temp.idx WHERE tablename=to_tbl AND fkey2.cols==idx.cols"
    ")", pHasErrors
  );
  if( rc!=SQLITE_OK ) return rc;

  return rc;
}

int main(int argc, char **argv){
  sqlite3 *db;
  Options opt = {0, 0, 0};
  int rc;
  int hasErrors = 0;
  char *zErr = 0;
  const int enc = SQLITE_UTF8;

  const char *zSql =
    "SELECT multireplace('"

      "-- Triggers for foreign key mapping:\n"
      "--\n"
      "--     /from_readable/ REFERENCES /to_readable/\n"
      "--     on delete /on_delete/\n"
      "--     on update /on_update/\n"
      "--\n"

      /* The "BEFORE INSERT ON <referencing>" trigger. This trigger's job is to
      ** throw an exception if the user tries to insert a row into the
      ** referencing table for which there is no corresponding row in
      ** the referenced table.
      */
      "CREATE TRIGGER /name/_insert_referencing BEFORE INSERT ON /tbl/ WHEN \n"
      "    /key_notnull/ AND NOT EXISTS (SELECT 1 FROM /ref/ WHERE /cond1/)\n" 
      "BEGIN\n"
        "  SELECT RAISE(ABORT, ''constraint failed'');\n"
      "END;\n"

      /* The "BEFORE UPDATE ON <referencing>" trigger. This trigger's job 
      ** is to throw an exception if the user tries to update a row in the
      ** referencing table causing it to correspond to no row in the
      ** referenced table.
      */
      "CREATE TRIGGER /name/_update_referencing BEFORE\n"
      "    UPDATE OF /rkey_list/ ON /tbl/ WHEN \n"
      "    /key_notnull/ AND \n"
      "    NOT EXISTS (SELECT 1 FROM /ref/ WHERE /cond1/)\n" 
      "BEGIN\n"
        "  SELECT RAISE(ABORT, ''constraint failed'');\n"
      "END;\n"


      /* The "BEFORE DELETE ON <referenced>" trigger. This trigger's job 
      ** is to detect when a row is deleted from the referenced table to 
      ** which rows in the referencing table correspond. The action taken
      ** depends on the value of the 'ON DELETE' clause.
      */
      "CREATE TRIGGER /name/_delete_referenced BEFORE DELETE ON /ref/ WHEN\n"
      "    EXISTS (SELECT 1 FROM /tbl/ WHERE /cond2/)\n"
      "BEGIN\n"
      "  /delete_action/\n"
      "END;\n"

      /* The "BEFORE DELETE ON <referenced>" trigger. This trigger's job 
      ** is to detect when the key columns of a row in the referenced table 
      ** to which one or more rows in the referencing table correspond are
      ** updated. The action taken depends on the value of the 'ON UPDATE' 
      ** clause.
      */
      "CREATE TRIGGER /name/_update_referenced AFTER\n"
      "    UPDATE OF /fkey_list/ ON /ref/ WHEN \n"
      "    EXISTS (SELECT 1 FROM /tbl/ WHERE /cond2/)\n"
      "BEGIN\n"
      "  /update_action/\n"
      "END;\n"
    "'"

    /* These are used in the SQL comment written above each set of triggers */
    ", '/from_readable/',  from_tbl || '(' || sj(from_col, ', ') || ')'"
    ", '/to_readable/',    to_tbl || '(' || sj(to_col, ', ') || ')'"
    ", '/on_delete/', on_delete"
    ", '/on_update/', on_update"

    ", '/name/',   'genfkey' || min(rowid)"
    ", '/tbl/',    dq(from_tbl)"
    ", '/ref/',    dq(to_tbl)"
    ", '/key_notnull/', sj('new.' || dq(from_col) || ' IS NOT NULL', ' AND ')"

    ", '/fkey_list/', sj(to_col, ', ')"
    ", '/rkey_list/', sj(from_col, ', ')"

    ", '/cond1/',  sj(multireplace('new./from/ == /to/'"
                   ", '/from/', dq(from_col)"
                   ", '/to/',   dq(to_col)"
                   "), ' AND ')"
    ", '/cond2/',  sj(multireplace('old./to/ == /from/'"
                   ", '/from/', dq(from_col)"
                   ", '/to/',   dq(to_col)"
                   "), ' AND ')"

    ", '/update_action/', CASE on_update "
      "WHEN 'SET NULL' THEN "
        "multireplace('UPDATE /tbl/ SET /setlist/ WHERE /where/;' "
        ", '/setlist/', sj(from_col||' = NULL',', ')"
        ", '/tbl/',     dq(from_tbl)"
        ", '/where/',   sj(from_col||' = old.'||dq(to_col),' AND ')"
        ")"
      "WHEN 'CASCADE' THEN "
        "multireplace('UPDATE /tbl/ SET /setlist/ WHERE /where/;' "
        ", '/setlist/', sj(dq(from_col)||' = new.'||dq(to_col),', ')"
        ", '/tbl/',     dq(from_tbl)"
        ", '/where/',   sj(dq(from_col)||' = old.'||dq(to_col),' AND ')"
        ")"
      "ELSE "
      "  'SELECT RAISE(ABORT, ''constraint failed'');'"
      "END "

    ", '/delete_action/', CASE on_delete "
      "WHEN 'SET NULL' THEN "
        "multireplace('UPDATE /tbl/ SET /setlist/ WHERE /where/;' "
        ", '/setlist/', sj(from_col||' = NULL',', ')"
        ", '/tbl/',     dq(from_tbl)"
        ", '/where/',   sj(from_col||' = old.'||dq(to_col),' AND ')"
        ")"
      "WHEN 'CASCADE' THEN "
        "multireplace('DELETE FROM /tbl/ WHERE /where/;' "
        ", '/tbl/',     dq(from_tbl)"
        ", '/where/',   sj(dq(from_col)||' = old.'||dq(to_col),' AND ')"
        ")"
      "ELSE "
      "  'SELECT RAISE(ABORT, ''constraint failed'');'"
      "END "

    ") FROM temp.fkey "
    "GROUP BY from_tbl, fkid"
  ;

  processCmdLine(argc, argv, &opt);

  /* Open the database handle. */
  rc = sqlite3_open_v2(opt.zDb, &db, SQLITE_OPEN_READONLY, 0);
  if( rc!=SQLITE_OK ){
    fprintf(stderr, "Error opening database file: %s\n", sqlite3_errmsg(db));
    return -1;
  }

  /* Create the special scalar and aggregate functions used by this program. */
  sqlite3_create_function(db, "dq", 1, enc, 0, doublequote, 0, 0);
  sqlite3_create_function(db, "multireplace", -1, enc, db, multireplace, 0, 0);
  sqlite3_create_function(db, "sj", 2, enc, 0, 0, joinStep, joinFinalize);

  /* Install the "schema" virtual table module */
  installSchemaModule(db);

  /* Create and populate a temp table with the information required to
  ** build the foreign key triggers. See function populateTempTable()
  ** for details.
  */
  rc = populateTempTable(db, &zErr, &hasErrors);
  if( rc!=SQLITE_OK ){
    fprintf(stderr, "Error reading database: %s\n", zErr);
    return -1;
  }
  if( hasErrors && opt.ignoreErrors==0 ){
    return -1;
  }

  printf("BEGIN;\n");

  /* Unless the --no-drop option was specified, generate DROP TRIGGER
  ** statements to drop any triggers in the database generated by a
  ** previous run of this program.
  */
  if( opt.noDrop==0 ){
    rc = sqlite3_exec(db, 
      "SELECT 'DROP TRIGGER' || ' ' || dq(name) || ';'"
      "FROM sqlite_master "
      "WHERE type='trigger' AND substr(name, 0, 7) == 'genfkey'"
      , printString, 0, 0
    );
    if( rc!=SQLITE_OK ){
      const char *zMsg = sqlite3_errmsg(db);
      fprintf(stderr, "Generating drop triggers failed: %s\n", zMsg);
      return -1;
    }
  }

  /* Run the main query to create the trigger definitions. */
  rc = sqlite3_exec(db, zSql, printString, 0, 0);
  if( rc!=SQLITE_OK ){
    fprintf(stderr, "Generating triggers failed: %s\n", sqlite3_errmsg(db));
    return -1;
  }

  printf("COMMIT;\n");
  return 0;
}

package require sqlite3

proc do_test {name cmd expected} {
  puts -nonewline "$name ..."
  set res [uplevel $cmd]
  if {$res eq $expected} {
    puts Ok
  } else {
    puts Error
    puts "  Got: $res"
    puts "  Expected: $expected"
    exit
  }
}

proc execsql {sql} {
  uplevel [list db eval $sql]
}

proc catchsql {sql} {
  set rc [catch {uplevel [list db eval $sql]} msg]
  list $rc $msg
}

file delete -force test.db test.db.journal
sqlite3 db test.db

# The following tests - genfkey-1.* - test RESTRICT foreign keys.
#
do_test genfkey-1.1 {
  execsql {
    CREATE TABLE t1(a INTEGER PRIMARY KEY, b, c, UNIQUE(b, c));
    CREATE TABLE t2(e REFERENCES t1, f);
    CREATE TABLE t3(g, h, i, FOREIGN KEY (h, i) REFERENCES t1(b, c));
  }
} {}
do_test genfkey-1.2 {
  execsql [exec ./genfkey test.db]
} {}
do_test genfkey-1.3 {
  catchsql { INSERT INTO t2 VALUES(1, 2) }
} {1 {constraint failed}}
do_test genfkey-1.4 {
  execsql {
    INSERT INTO t1 VALUES(1, 2, 3);
    INSERT INTO t2 VALUES(1, 2);
  }
} {}
do_test genfkey-1.5 {
  execsql { INSERT INTO t2 VALUES(NULL, 3) }
} {}
do_test genfkey-1.6 {
  catchsql { UPDATE t2 SET e = 5 WHERE e IS NULL }
} {1 {constraint failed}}
do_test genfkey-1.7 {
  execsql { UPDATE t2 SET e = 1 WHERE e IS NULL }
} {}
do_test genfkey-1.8 {
  execsql { UPDATE t2 SET e = NULL WHERE f = 3 }
} {}
do_test genfkey-1.9 {
  catchsql { UPDATE t1 SET a = 10 }
} {1 {constraint failed}}
do_test genfkey-1.9a {
  catchsql { UPDATE t1 SET a = NULL }
} {1 {datatype mismatch}}
do_test genfkey-1.10 {
  catchsql { DELETE FROM t1 }
} {1 {constraint failed}}
do_test genfkey-1.11 {
  execsql { UPDATE t2 SET e = NULL }
} {}
do_test genfkey-1.12 {
  execsql { 
    UPDATE t1 SET a = 10 ;
    DELETE FROM t1;
    DELETE FROM t2;
  }
} {}

do_test genfkey-1.13 {
  execsql {
    INSERT INTO t3 VALUES(1, NULL, NULL);
    INSERT INTO t3 VALUES(1, 2, NULL);
    INSERT INTO t3 VALUES(1, NULL, 3);
  }
} {}
do_test genfkey-1.14 {
  catchsql { INSERT INTO t3 VALUES(3, 1, 4) }
} {1 {constraint failed}}
do_test genfkey-1.15 {
  execsql { 
    INSERT INTO t1 VALUES(1, 1, 4);
    INSERT INTO t3 VALUES(3, 1, 4);
  }
} {}
do_test genfkey-1.16 {
  catchsql { DELETE FROM t1 }
} {1 {constraint failed}}
do_test genfkey-1.17 {
  catchsql { UPDATE t1 SET b = 10}
} {1 {constraint failed}}
do_test genfkey-1.18 {
  execsql { UPDATE t1 SET a = 10}
} {}
do_test genfkey-1.19 {
  catchsql { UPDATE t3 SET h = 'hello' WHERE i = 3}
} {1 {constraint failed}}

do_test genfkey-1.X {
  execsql {
    DROP TABLE t1;
    DROP TABLE t2;
    DROP TABLE t3;
  }
} {}

# The following tests - genfkey-2.* - test CASCADE foreign keys.
#
do_test genfkey-2.1 {
  execsql {
    CREATE TABLE t1(a INTEGER PRIMARY KEY, b, c, UNIQUE(b, c));
    CREATE TABLE t2(e REFERENCES t1 ON UPDATE CASCADE ON DELETE CASCADE, f);
    CREATE TABLE t3(g, h, i, 
        FOREIGN KEY (h, i) 
        REFERENCES t1(b, c) ON UPDATE CASCADE ON DELETE CASCADE
    );
  }
} {}
do_test genfkey-2.2 {
  execsql [exec ./genfkey test.db]
} {}
do_test genfkey-2.3 {
  execsql {
    INSERT INTO t1 VALUES(1, 2, 3);
    INSERT INTO t1 VALUES(4, 5, 6);
    INSERT INTO t2 VALUES(1, 'one');
    INSERT INTO t2 VALUES(4, 'four');
  }
} {}
do_test genfkey-2.4 {
  execsql {
    UPDATE t1 SET a = 2 WHERE a = 1;
    SELECT * FROM t2;
  }
} {2 one 4 four}
do_test genfkey-2.5 {
  execsql {
    DELETE FROM t1 WHERE a = 4;
    SELECT * FROM t2;
  }
} {2 one}
do_test genfkey-2.6 {
  execsql {
    INSERT INTO t3 VALUES('hello', 2, 3);
    UPDATE t1 SET c = 2;
    SELECT * FROM t3;
  }
} {hello 2 2}
do_test genfkey-2.7 {
  execsql {
    DELETE FROM t1;
    SELECT * FROM t3;
  }
} {}
do_test genfkey-2.X {
  execsql {
    DROP TABLE t1;
    DROP TABLE t2;
    DROP TABLE t3;
  }
} {}


# The following tests - genfkey-3.* - test SET NULL foreign keys.
#
do_test genfkey-3.1 {
  execsql {
    CREATE TABLE t1(a INTEGER PRIMARY KEY, b, c, UNIQUE(c, b));
    CREATE TABLE t2(e REFERENCES t1 ON UPDATE SET NULL ON DELETE SET NULL, f);
    CREATE TABLE t3(g, h, i, 
        FOREIGN KEY (h, i) 
        REFERENCES t1(b, c) ON UPDATE SET NULL ON DELETE SET NULL
    );
  }
} {}
do_test genfkey-3.2 {
  execsql [exec ./genfkey test.db]
} {}
do_test genfkey-3.3 {
  execsql {
    INSERT INTO t1 VALUES(1, 2, 3);
    INSERT INTO t1 VALUES(4, 5, 6);
    INSERT INTO t2 VALUES(1, 'one');
    INSERT INTO t2 VALUES(4, 'four');
  }
} {}
do_test genfkey-3.4 {
  execsql {
    UPDATE t1 SET a = 2 WHERE a = 1;
    SELECT * FROM t2;
  }
} {{} one 4 four}
do_test genfkey-3.5 {
  execsql {
    DELETE FROM t1 WHERE a = 4;
    SELECT * FROM t2;
  }
} {{} one {} four}
do_test genfkey-3.6 {
  execsql {
    INSERT INTO t3 VALUES('hello', 2, 3);
    UPDATE t1 SET c = 2;
    SELECT * FROM t3;
  }
} {hello {} {}}
do_test genfkey-2.7 {
  execsql {
    UPDATE t3 SET h = 2, i = 2;
    DELETE FROM t1;
    SELECT * FROM t3;
  }
} {hello {} {}}
do_test genfkey-3.X {
  execsql {
    DROP TABLE t1;
    DROP TABLE t2;
    DROP TABLE t3;
  }
} {}

# The following tests - genfkey-4.* - test that errors in the schema
# are detected correctly.
#
do_test genfkey-4.1 {
  execsql {
    CREATE TABLE t1(a REFERENCES nosuchtable, b);
    CREATE TABLE t2(a REFERENCES t1, b);

    CREATE TABLE t3(a, b, c, PRIMARY KEY(a, b));
    CREATE TABLE t4(a, b, c, FOREIGN KEY(c, b) REFERENCES t3);

    CREATE TABLE t5(a REFERENCES t4(d), b, c);
    CREATE TABLE t6(a REFERENCES t4(a), b, c);
    CREATE TABLE t7(a REFERENCES t3(a), b, c);
    CREATE TABLE t8(a REFERENCES nosuchtable(a), b, c);
  }
} {}

do_test genfkey-4.X {
  set rc [catch {exec ./genfkey test.db} msg]
  list $rc $msg
} "1 {[string trim {
Error in table t5: foreign key columns do not exist
Error in table t8: foreign key columns do not exist
Error in table t4: implicit mapping to composite primary key
Error in table t1: implicit mapping to non-existant primary key
Error in table t2: implicit mapping to non-existant primary key
Error in table t6: foreign key is not unique
Error in table t7: foreign key is not unique
}]}"