|
|
//========= Copyright Valve Corporation, All rights reserved. ============//
//
// Purpose:
//
// $NoKeywords: $
//=============================================================================
#include "stdafx.h"
//#include "sqlaccess/sqlaccess.h"
// memdbgon must be the last include file in a .cpp file!!!
#include "tier0/memdbgon.h"
namespace GCSDK { #ifndef STEAM
bool isspace( char ch ) { return ch == ' ' || ch == '\t' || ch == '\n' || ch == '\r'; }
int Q_strnlen( const char *str, int count ) { // can't make more meaningful checks, because this routine is used itself
// to check the NUL-terminatedness of strings
if ( !str || count < 0 ) return -1;
for ( const char *pch = str; pch < str + count; pch++ ) { if ( *pch == '\0' ) return pch - str; }
return -1; }
#endif
//-----------------------------------------------------------------------------
// Purpose: convert an ESchemaCatalog into a string for diagnostics and logging.
// this can't be in enum_names because of data type dependencies.
//-----------------------------------------------------------------------------
const char* PchNameFromESchemaCatalog( ESchemaCatalog e ) { switch (e) { case k_ESchemaCatalogInvalid: return "k_ESchemaCatalogInvalid"; break;
case k_ESchemaCatalogMain: return "k_ESchemaCatalogMain"; break; }
AssertMsg1( false, "unknown ESchemaCatalog (%d)", e ); return "Unknown"; }
//-----------------------------------------------------------------------------
// Purpose: Constructor
//-----------------------------------------------------------------------------
CSchema::CSchema() { m_iTable = -1; m_rgchName[0] = 0; m_cubRecord = 0; m_bTestTable = false; m_cRecordMax = 0; m_bHasVarFields = false; m_nHasPrimaryKey = k_EPrimaryKeyTypeNone; m_iPKIndex = -1; m_pRecordInfo = NULL; m_wipePolicy = k_EWipePolicyPreserveAlways; m_bAllowWipeInProd = false; m_bPrepopulatedTable = false; m_nFullTextIndexCatalog = -1; m_eSchemaCatalog = k_ESchemaCatalogInvalid; }
//-----------------------------------------------------------------------------
// Purpose: Destructor
//-----------------------------------------------------------------------------
CSchema::~CSchema() { SAFE_RELEASE( m_pRecordInfo ); }
//-----------------------------------------------------------------------------
// Purpose: Calculates offset of each field within a record structure, and the
// maximum length of a record structure.
//-----------------------------------------------------------------------------
void CSchema::CalcOffsets() { int dubOffsetCur = 0;
for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { m_VecField[iField].m_dubOffset = dubOffsetCur; dubOffsetCur += m_VecField[iField].m_cubLength;
if ( m_VecField[iField].BIsVariableLength() ) m_bHasVarFields = true; }
m_cubRecord = dubOffsetCur;
if ( m_bHasVarFields ) m_cubRecord += sizeof( VarFieldBlockInfo_t ); }
//-----------------------------------------------------------------------------
// Purpose: called to make final calculations when all fields/indexes/etc have
// been added and the schema is ready to be used
//-----------------------------------------------------------------------------
void CSchema::PrepareForUse() { // Create a record description (new form of schema information, for SQL) that corresponds to this schema object.
// This contains essentially the information as the CSchema object, we keep both for the moment to bridge the DS and SQL worlds.
Assert( !m_pRecordInfo ); SAFE_RELEASE( m_pRecordInfo ); m_pRecordInfo = CRecordInfo::Alloc(); m_pRecordInfo->InitFromDSSchema( this ); }
//-----------------------------------------------------------------------------
// Purpose: For a record that has variable-length fields, gets the info block from
// the tail end
// Input : pvRecord - Record data
//-----------------------------------------------------------------------------
VarFieldBlockInfo_t* CSchema::PVarFieldBlockInfoFromRecord( const void *pvRecord ) const { if ( !m_bHasVarFields ) return NULL;
uint8 *pubRecord = ( uint8* )pvRecord; return ( VarFieldBlockInfo_t * )( pubRecord + m_cubRecord ) - 1; }
//-----------------------------------------------------------------------------
// Purpose: Return the total size of the variable-length block for this record
// For records that have no variable-length fields, it will return zero
// Input : pvRecord - Record data
//-----------------------------------------------------------------------------
uint32 CSchema::CubRecordVariable( const void *pvRecord ) const { VarFieldBlockInfo_t *pVarFieldsBlockInfo = PVarFieldBlockInfoFromRecord( pvRecord ); if ( pVarFieldsBlockInfo ) return pVarFieldsBlockInfo->m_cubBlock; else return 0; }
void CSchema::RenderField( uint8 *pubRecord, int iField, int cchBuffer, char *pchBuffer ) { Field_t *pField = &m_VecField[iField];
uint8 *pubData; uint32 cubData; char chEmpty = 0;
if ( pField->BIsVariableLength() ) { if ( !BGetVarField( pubRecord, ( VarField_t * )( pubRecord + pField->m_dubOffset ), &pubData, &cubData ) ) { // just render a single byte
pubData = ( uint8* )&chEmpty; cubData = 1; } } else { pubData = pubRecord + pField->m_dubOffset; cubData = m_VecField[iField].m_cubLength; }
ConvertFieldToText( pField->m_EType, pubData, cubData, pchBuffer, cchBuffer ); }
//-----------------------------------------------------------------------------
// Purpose: Renders a text version of a record to the console.
// Input : pubRecord - Location of the record data
//-----------------------------------------------------------------------------
void CSchema::RenderRecord( uint8 *pubRecord ) { char rgchT[k_cMedBuff];
// First the header
EmitInfo( SPEW_CONSOLE, 2, 2, "%d\t*** Record header: # of lines ***\n", 1 + m_VecField.Count() );
// Render each field in turn
for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { Field_t *pField = &m_VecField[iField];
RenderField( pubRecord, iField, Q_ARRAYSIZE(rgchT), rgchT );
EmitInfo( SPEW_CONSOLE, 2, 2, "\t%s\t\t// Field %d (%s)\n", rgchT, iField, pField->m_rgchName ); } }
//-----------------------------------------------------------------------------
// Purpose: Get data and size of a field, whether fixed or variable length
// Input: pvRecord - Fixed-length part of record
// iField - index of field to get
// ppubField - receives pointer to field data
// pcubField - receives size in bytes of that data
// Output: true if successful
//-----------------------------------------------------------------------------
bool CSchema::BGetFieldData( const void *pvRecord, int iField, uint8 **ppubField, uint32 *pcubField ) const { *ppubField = NULL; *pcubField = 0; const Field_t &field = m_VecField[iField]; if ( field.BIsVariableLength() ) { return BGetVarField( pvRecord, ( VarField_t * )( ( uint8 * ) pvRecord + field.m_dubOffset ), ppubField, pcubField ); } else { *ppubField = ( ( uint8 * ) pvRecord + field.m_dubOffset ); *pcubField = field.CubFieldUpdateSize(); }
return true; }
//-----------------------------------------------------------------------------
// Purpose: Set data and size of a field, whether fixed or variable length
// Input: pvRecord - Fixed-length part of record
// iField - index of field to set
// pubField - pointer to field data to copy from
// cubField - size in bytes of that data
// Output: true if successful
//-----------------------------------------------------------------------------
bool CSchema::BSetFieldData( void *pvRecord, int iField, uint8 *pubField, uint32 cubField, bool *pbVarBlockRealloced ) { *pbVarBlockRealloced = false;
Field_t &field = m_VecField[iField];
if ( field.BIsVariableLength() ) { return BSetVarField( pvRecord, ( VarField_t * )( ( uint8 * ) pvRecord + field.m_dubOffset ), pubField, cubField, pbVarBlockRealloced, /*bFreeOnRealloc*/ true ); } else // fixed length
{ uint8 *pubFieldWrite = ( ( uint8 * ) pvRecord + field.m_dubOffset );
// Must fit in field and last byte for strings MUST be NULL
if ( cubField > field.m_cubLength || ( k_EGCSQLType_String == field.m_EType && Q_strnlen( reinterpret_cast< char* >( pubField ), cubField ) == -1 ) ) { Assert( false ); return false; }
if ( k_EGCSQLType_Blob != field.m_EType && k_EGCSQLType_Image != field.m_EType ) { // Copy the data (string or binary, doesn't matter)
if ( cubField > 0 ) Q_memcpy( pubFieldWrite, pubField, cubField );
// Null termination and overwrite any old data
if ( cubField < field.m_cubLength ) Q_memset( pubFieldWrite + cubField, 0, field.m_cubLength - cubField ); } else { // Only support fixed char or binary
Assert( false ); return false; } }
return true; }
//-----------------------------------------------------------------------------
// Purpose: Get data from a variable-length field
// Input: pvRecord - Fixed-length part of record
// pVarField - fixed part of field in that record
// ppubField - receives pointer to field data
// pcubField - receives size in bytes of that data
//-----------------------------------------------------------------------------
bool CSchema::BGetVarField( const void *pvRecord, const VarField_t *pVarField, uint8 **ppubField, uint32 *pcubField ) const { Assert( m_bHasVarFields );
*ppubField = 0; *pcubField = 0;
VarFieldBlockInfo_t *pVarFieldBlockInfo = PVarFieldBlockInfoFromRecord( pvRecord );
if ( !pVarField->m_cubField ) { *pcubField = 0; *ppubField = NULL; return true; }
if ( pVarFieldBlockInfo->m_cubBlock ) { uint8 *pubVarBlock = pVarFieldBlockInfo->m_pubBlock;
*ppubField = pubVarBlock + pVarField->m_dubOffset; *pcubField = pVarField->m_cubField;
// Sanity check
Assert( *pcubField <= k_cubVarFieldMax );
return true; } else { // Should never happen
Assert( false ); return false; } }
//-----------------------------------------------------------------------------
// Purpose: Update a variable-length field in a record (may realloc the var block)
// Input: pvRecord - Fixed-length part of record
// pVarField - fixed part of field in that record
// pvData - data to place in the field
// cubData - size of that data
// pbRealloced - set to indicate if the var block was realloced
// bFreeOnRealloc - If we have to grow or shrink the var block, should we free the old memory?
// Usually true, unless it lives in a NetPacket or something like that
//-----------------------------------------------------------------------------
bool CSchema::BSetVarField( void *pvRecord, VarField_t *pVarField, const void *pvData, uint32 cubData, bool *pbRealloced, bool bFreeOnRealloc ) { VarFieldBlockInfo_t *pVarFieldBlockInfo = PVarFieldBlockInfoFromRecord( pvRecord ); *pbRealloced = false;
if ( cubData > k_cubVarFieldMax ) { // field size is too big
Assert( false ); return false; }
// if no block exists, allocate and copy into it
if ( pVarFieldBlockInfo->m_cubBlock == 0 ) { // Nothing to do?
if ( !cubData ) return true;
// create it
void *pvBlock = PvAlloc( cubData ); *pbRealloced = true; // copy the data
Q_memcpy( pvBlock, pvData, cubData );
// set the record's structure to point at our new data
pVarFieldBlockInfo->m_cubBlock = cubData; pVarFieldBlockInfo->m_pubBlock = ( uint8 * )pvBlock;
// set the field to point at its landing place
pVarField->m_cubField = cubData; pVarField->m_dubOffset = 0; } else { // there is some block available.
// is this field changing size?
if ( cubData != 0 && ( cubData == pVarField->m_cubField ) ) { // no size change - no need to reallocate anything
Q_memcpy( pVarFieldBlockInfo->m_pubBlock + pVarField->m_dubOffset, pvData, cubData ); } else { // size change - realloc
*pbRealloced = true; uint32 cubFieldOld = pVarField->m_cubField; uint32 dubOffsetOld = pVarField->m_dubOffset; uint32 cubBlockNew = pVarFieldBlockInfo->m_cubBlock - pVarField->m_cubField + cubData;
if ( ( cubBlockNew + m_cubRecord ) > k_cubRecordMax ) { // total record size is too big
Assert( false ); return false; }
void *pvBlockNew = NULL;
if ( cubBlockNew ) { // if this field has never been placed in the block (that is, it's not changing an old value)
// and we have enough space, fastest to put it at the end in the free space.
if ( pVarField->m_cubField == 0 && pVarField->m_dubOffset == 0 && cubData <= pVarFieldBlockInfo->m_cubBlockFree ) { uint8 *pubLastUsed = pVarFieldBlockInfo->m_pubBlock; for ( int iField = 0; iField < m_VecField.Count(); ++iField ) { Field_t &field = m_VecField[iField]; if ( field.BIsVariableLength() ) { // Needs updating
VarField_t *pVarFieldCur = ( VarField_t * )( ( uint8 * )pvRecord + field.m_dubOffset ); pubLastUsed += pVarFieldCur->m_cubField; } }
// copy it there
Q_memcpy( pubLastUsed, pvData, cubData );
// set up the field
pVarField->m_cubField = cubData; pVarField->m_dubOffset = static_cast<int>( (pubLastUsed - pVarFieldBlockInfo->m_pubBlock) );
// note that we used some up
pVarFieldBlockInfo->m_cubBlockFree -= cubData; } else { // yes ... rellocate
pvBlockNew = PvAlloc( cubBlockNew );
uint8 *pubBlockOldCursor = pVarFieldBlockInfo->m_pubBlock; uint8 *pubBlockNewCursor = ( uint8* )pvBlockNew;
// copy data, skipping over this field (will put at end)
while ( pubBlockOldCursor < ( pVarFieldBlockInfo->m_pubBlock + pVarFieldBlockInfo->m_cubBlock ) ) { if ( pVarField->m_cubField && ( (int)pVarField->m_dubOffset == ( pubBlockOldCursor - pVarFieldBlockInfo->m_pubBlock ) ) ) { pubBlockOldCursor += pVarField->m_cubField; } else { *pubBlockNewCursor++ = *pubBlockOldCursor++; } }
// put this field data at the end
Q_memcpy( pubBlockNewCursor, pvData, cubData );
// free the old block
if ( bFreeOnRealloc ) FreePv( pVarFieldBlockInfo->m_pubBlock );
// Update the block info
pVarFieldBlockInfo->m_cubBlock = cubBlockNew; pVarFieldBlockInfo->m_pubBlock = ( uint8 * )pvBlockNew; pVarFieldBlockInfo->m_cubBlockFree = 0;
// update this field
pVarField->m_cubField = cubData; if ( cubData > 0 ) pVarField->m_dubOffset = static_cast<int>( pubBlockNewCursor - pVarFieldBlockInfo->m_pubBlock ); else pVarField->m_dubOffset = 0;
// update other fields
for ( int iField = 0; iField < m_VecField.Count(); ++iField ) { Field_t &field = m_VecField[iField]; if ( field.BIsVariableLength() ) { // Needs updating
VarField_t *pVarFieldCur = ( VarField_t * )( ( uint8 * )pvRecord + field.m_dubOffset );
// Except the one we just changed
if ( pVarFieldCur == pVarField ) continue;
// And empty fields
if ( !pVarFieldCur->m_cubField ) continue;
if ( pVarFieldCur->m_dubOffset > dubOffsetOld ) pVarFieldCur->m_dubOffset -= cubFieldOld; } } } } else { // all of the variable data is gone, so
// free the old block
if ( bFreeOnRealloc ) FreePv( pVarFieldBlockInfo->m_pubBlock );
// ... update the block info
pVarFieldBlockInfo->m_cubBlock = 0; pVarFieldBlockInfo->m_pubBlock = NULL; pVarFieldBlockInfo->m_cubBlockFree = 0;
// ... and update this field
pVarField->m_dubOffset = 0; pVarField->m_cubField = cubData; } } }
return true; }
//-----------------------------------------------------------------------------
// Purpose: If this is a variable-length record, and we just read it from
// a stream, then the var block is after the fixed-length part of the
// record. So, we need to update the record's pointer to reflect that
// Input: pvRecord - Beginning of the serialized record
//-----------------------------------------------------------------------------
void CSchema::FixupDeserializedRecord( void *pvRecord ) { // Nothing to do if not a variable record
if ( !BHasVariableFields() ) return;
uint8 *pubRecord = ( uint8 * )pvRecord; VarFieldBlockInfo_t *pVarBlockInfo = PVarFieldBlockInfoFromRecord( pvRecord ); pVarBlockInfo->m_pubBlock = pubRecord + m_cubRecord; }
//-----------------------------------------------------------------------------
// Purpose: Initializes a new record with random data.
// Input: pubRecord - The record's in-memory data that we'll fill out
// unPrimaryIndex - Primary index of the record
// pbRealloced - Set to 'true' if the var-fields block for this record was reallocated
// (or allocated for the first time)
// bFreeOnRealloc - If true, we should Free() the var block after reallocating a new block
// (should be false if that block lives inside a NetPacket)
//-----------------------------------------------------------------------------
void CSchema::InitRecordRandom( uint8 *pubRecord, uint32 unPrimaryIndex, bool *pbVarBlockRealloced, bool bFreeVarBlockOnRealloc ) { // Fill out each field in turn
for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { bool bRealloced = false; SetFieldRandom( pubRecord, iField, &bRealloced, bFreeVarBlockOnRealloc ); if ( bRealloced ) { *pbVarBlockRealloced = true; // if we just allocated it, we can free it next time we need to
bFreeVarBlockOnRealloc = true; } } }
//-----------------------------------------------------------------------------
// Purpose: Sets a single field of a record to a random value.
// Input: pubRecord - Where the record lives in memory
// iField - Which field to set
// pbRealloced - Set to 'true' if the var-fields block for this record was reallocated
// (or allocated for the first time)
// bFreeOnRealloc - If true, we should Free() the var block after reallocating a new block
// (should be false if that block lives inside a NetPacket)
//-----------------------------------------------------------------------------
void CSchema::SetFieldRandom( uint8 *pubRecord, int iField, bool *pbVarBlockRealloced, bool bFreeVarBlockOnRealloc ) { Field_t &field = m_VecField[iField]; *pbVarBlockRealloced = false;
// Strings get random text
if ( k_EGCSQLType_String == field.m_EType ) { // Generate up to cubLength - 1 chars
uint32 cch = UNRandFast() % field.m_cubLength; uint32 ich = 0; for ( ; ich < cch; ich++ ) *( ( char * ) pubRecord + field.m_dubOffset + ich ) = CHRandFast();
// Null termination
for ( ; ich < field.m_cubLength; ich++ ) *( ( char * ) pubRecord + field.m_dubOffset + ich ) = 0; } else if ( field.BIsVariableLength() ) { // Need temp buffer to randomize before setting
// kept reasonably small to prevent spamming the console
uint8 rgubBuff[512]; uint32 cubData = UNRandFast() % sizeof(rgubBuff);
// For strings, put in (cubData-1) random characters (each randomly in the range [32,126])
// then a trailing NULL
if ( field.BIsStringType() ) { uint32 ich = 0; for ( ; ich < (cubData-1); ich++ ) rgubBuff[ich] = CHRandFast();
rgubBuff[ich] = 0; } else { // binary - just fill in random bytes
for ( uint32 iub = 0; iub < cubData; iub++ ) { rgubBuff[iub] = ( uint8 )( UNRandFast() % 256 ); } }
// Set the variable field in the record
BSetVarField( pubRecord, ( VarField_t * ) ( pubRecord + field.m_dubOffset ), rgubBuff, cubData, pbVarBlockRealloced, bFreeVarBlockOnRealloc ); } // Binaries are filled with completely random bytes
else { for ( uint32 iub = 0; iub < field.m_cubLength; iub++ ) { *( pubRecord + field.m_dubOffset + iub ) = ( uint8 ) ( UNRandFast() % 256 ); } } }
//-----------------------------------------------------------------------------
// Purpose: Returns checksum of our contents
// Output : checksum
//-----------------------------------------------------------------------------
uint32 CSchema::CalcChecksum() { CRC32_t crc32; CRC32_Init( &crc32 );
FOR_EACH_VEC( m_VecField, nField ) { CRC32_ProcessBuffer( &crc32, &m_VecField[nField], sizeof( m_VecField[nField] ) ); }
// keep checksum for entire record info
CRC32_Final( &crc32 );
return (uint32)crc32; }
void CSchema::AddIntField( char *pchName, char *pchSQLName, EGCSQLType eType, int cubSize ) { int nExpectedSize = -1; switch ( eType ) { case k_EGCSQLType_int8: nExpectedSize = 1; break;
case k_EGCSQLType_int16: nExpectedSize = 2; break;
case k_EGCSQLType_int32: case k_EGCSQLType_float: nExpectedSize = 4; break;
case k_EGCSQLType_int64: case k_EGCSQLType_double: nExpectedSize = 8; break; }
AssertMsg2( nExpectedSize != -1, "Unexpected EType in AddIntField: %d for column %s", eType, pchSQLName ); AssertMsg3( nExpectedSize == cubSize, "Unexpected size for in AddIntField for column %s: %d doesn't match %d ", pchSQLName, nExpectedSize, cubSize );
AddField( pchName, pchSQLName, eType, cubSize, 0 ); }
//-----------------------------------------------------------------------------
// Purpose: Adds a field from our intrinsic schema to this schema.
// Input: pchName - Name of the field
// pchSQLName - Name of the field in SQL database
// eType - Type of the field
// cubSize - Size of the field
// pfnCompare - Function used to compare fields (NULL if none specified)
//-----------------------------------------------------------------------------
void CSchema::AddField( char *pchName, char *pchSQLName, EGCSQLType eType, uint32 cubSize, int cchMaxLength ) { int iFieldNew = m_VecField.AddToTail(); Field_t &field = m_VecField[iFieldNew];
field.m_EType = eType; field.m_cubLength = cubSize; field.m_nColFlags = 0; Q_strncpy( field.m_rgchName, pchName, Q_ARRAYSIZE( field.m_rgchName ) ); Q_strncpy( field.m_rgchSQLName, pchSQLName, Q_ARRAYSIZE( field.m_rgchSQLName ) ); field.m_cchMaxLength = cchMaxLength; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as the primary index
// Input:
// bClustered - true to create a clustered index
// pchName - Name of the field to make primary index
//-----------------------------------------------------------------------------
int CSchema::PrimaryKey( bool bClustered, int nFillFactor, const char *pchName ) { int iField = FindIField( pchName ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchName, m_rgchName );
Assert( m_nHasPrimaryKey == k_EPrimaryKeyTypeNone ); // may not already have a primary key defined
Assert( m_iPKIndex == k_iFieldNil );
// must not have been indexed in any way before
Assert( 0 == ( m_VecField[ iField ].m_nColFlags & ( k_nColFlagPrimaryKey | k_nColFlagIndexed | k_nColFlagUnique | k_nColFlagClustered ) ) );
// note that we have a single-column PK
m_nHasPrimaryKey = k_EPrimaryKeyTypeSingle;
// set our flags
m_VecField[ iField ].m_nColFlags |= ( k_nColFlagPrimaryKey | k_nColFlagIndexed | k_nColFlagUnique ); if ( bClustered ) m_VecField[ iField ].m_nColFlags |= k_nColFlagClustered;
// add to list of primary keys and remember the indexID
CUtlVector<int> vecColumns; vecColumns.AddToTail( iField );
FieldSet_t vecIndex( true /* unique */ , bClustered, vecColumns, NULL ); vecIndex.SetFillFactor( nFillFactor ); m_iPKIndex = m_VecIndexes.AddToTail( vecIndex );
return m_iPKIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a set of fields as the primary index
// Input:
// bClustered - clustered index created if true
// nFillFactor - fill facto to use; 0 is database default
// pchName - Name of the field to make primary index
//-----------------------------------------------------------------------------
int CSchema::PrimaryKeys( bool bClustered, int nFillFactor, const char *pchNames ) { Assert( pchNames != NULL ); // no bogus parameters, please
Assert( m_nHasPrimaryKey == k_EPrimaryKeyTypeNone ); // may not already have a primary key defined
Assert( m_iPKIndex == k_iFieldNil ); // no primary key defined
int nFlags = k_nColFlagPrimaryKey | k_nColFlagIndexed | k_nColFlagUnique; if ( bClustered ) nFlags |= k_nColFlagClustered;
// go add all those fields as Indexed
int nNewIndex = AddIndexToFieldList( pchNames, NULL, nFlags, nFillFactor ); if ( nNewIndex != k_iFieldNil ) { m_nHasPrimaryKey = k_EPrimaryKeyTypeMulti; // remember that we have multiple keys
m_iPKIndex = nNewIndex; } return nNewIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a set of fields as the primary index
// Input:
// pchName - Names of the fields for the primary index; comma-separated if multiple
// pchIndexName - Name of the index object (not in SQL)
// nFlags - flags for CColumnInfo on this object
// nFillFactor - fill facto to use; 0 is database default
//-----------------------------------------------------------------------------
int CSchema::AddIndexToFieldList( const char *pchNames, const char *pchIndexName, int nFlags, int nFillFactor ) { // need a copy of string list since the argument is const and read-only
int cNamesLen = Q_strlen( pchNames ) + 1; char* pchNamesCopy = (char*) PvAlloc( cNamesLen ); Q_strncpy( pchNamesCopy, pchNames, cNamesLen );
if (pchNames == NULL) { // not enough memory!
AssertFatal( false ); return k_iFieldNil; }
CUtlVector<int> vecKeys;
char* pchCurrent = pchNamesCopy; do { // find next token
char* pchEnd = strchr(pchCurrent, ','); if ( pchEnd != NULL ) { *pchEnd++ = 0; }
// skip whitespace
while (isspace(*pchCurrent)) pchCurrent++;
// find the name; we expect a C++ name, not a SQL name
int iField = FindIField( pchCurrent ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchCurrent, m_rgchName );
// mark as a primary key
m_VecField[iField].m_nColFlags |= nFlags ;
// add it to our collection
vecKeys.AddToTail( iField );
// move past the end of this token in the string
pchCurrent = pchEnd; } while ( pchCurrent != NULL );
// release our copy
FreePv(pchNamesCopy);
// create a fieldset with our list of indexes
// and add it to our indexes collection
bool bUnique = 0 != (nFlags & k_nColFlagUnique); bool bClustered = 0 != (nFlags & k_nColFlagClustered); FieldSet_t vecIndex( bUnique, bClustered, vecKeys, pchIndexName ); vecIndex.SetFillFactor( nFillFactor ); int iReturn = m_VecIndexes.AddToTail( vecIndex ); return iReturn; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as indexed.
// Input: pchName - Name of the field to index
// pchIndexName - Name of the index object (not in SQL)
//-----------------------------------------------------------------------------
int CSchema::IndexField( const char *pchIndexName, const char *pchName ) { Assert( pchName != NULL ); Assert( pchIndexName != NULL );
// find the field by name
int iField = FindIField( pchName ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchName, m_rgchName );
// add an index to it
return AddIndexToFieldNumber( iField, pchIndexName, false /* Not clustered */ ); }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as indexed.
// Input: pchIndexName - Name of the index object (not in SQL)
// pchName - Name of the fields to index; comma separated if multiple
//-----------------------------------------------------------------------------
int CSchema::IndexFields( const char *pchIndexName, const char *pchNames ) { Assert( pchNames != NULL ); Assert( pchIndexName != NULL );
// go add all those fields as Indexed
int nNewIndex = AddIndexToFieldList( pchNames, pchIndexName, k_nColFlagIndexed, 0 ); return nNewIndex; }
//-----------------------------------------------------------------------------
// Purpose: Includes a column (or columns) in an index for the INCLUDE clause
// Input: pchIndexName - Name of the index object (not in SQL)
// pchName - Name of the fields to index; comma separated if multiple
//-----------------------------------------------------------------------------
void CSchema::AddIncludedFields( const char *pchIndexName, const char *pchNames ) { Assert( pchNames != NULL ); Assert( pchIndexName != NULL );
// find that index by name
int nIndexIndex = -1; FOR_EACH_VEC( m_VecIndexes, i ) { FieldSet_t &refSet = m_VecIndexes.Element(i); const char *pstrMatch = refSet.GetIndexName(); if ( Q_stricmp( pstrMatch, pchIndexName ) == 0 ) { nIndexIndex = i; break; } }
// must have found it
AssertFatalMsg1( nIndexIndex != -1, "Index \"%s\" not found", pchIndexName ); FieldSet_t &refSet = m_VecIndexes.Element( nIndexIndex );
// need a copy of string list since the argument is const and read-only
int cNamesLen = Q_strlen( pchNames ) + 1; char* pchNamesCopy = (char*) PvAlloc( cNamesLen ); Q_strncpy( pchNamesCopy, pchNames, cNamesLen );
char* pchCurrent = pchNamesCopy; do { // find next token
char* pchEnd = strchr(pchCurrent, ','); if ( pchEnd != NULL ) { *pchEnd++ = 0; }
// skip whitespace
while (isspace(*pchCurrent)) pchCurrent++;
// find the name; we expect a C++ name, not a SQL name
int iField = FindIField( pchCurrent ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchCurrent, m_rgchName );
// add it to our collection
refSet.AddIncludedColumn( iField );
// move past the end of this token in the string
pchCurrent = pchEnd; } while ( pchCurrent != NULL );
// release our copy
FreePv( pchNamesCopy );
return; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as indexed.
// Input: pchIndexName - Name of the index object (not in SQL)
// pchName - Name of the fields to index; comma separated if multiple
//-----------------------------------------------------------------------------
int CSchema::UniqueFields( const char *pchIndexName, const char *pchNames ) { Assert( pchNames != NULL ); Assert( pchIndexName != NULL );
// go add all those fields as Indexed
int nNewIndex = AddIndexToFieldList( pchNames, pchIndexName, k_nColFlagIndexed | k_nColFlagUnique, 0 ); return nNewIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field a having a clustered index.
// Input: pchName - Name of the field to index
// pchIndexName - Name of the index object (not in SQL)
//-----------------------------------------------------------------------------
int CSchema::ClusteredIndexField( int nFillFactor, const char *pchIndexName, const char *pchName ) { Assert( pchName != NULL ); Assert( pchIndexName != NULL );
// can't previously have some other clustered index
FOR_EACH_VEC( m_VecIndexes, iIndex ) { if ( m_VecIndexes[iIndex].IsClustered() ) { AssertFatalMsg3( false, "On table %s, can't make index %s clustered because %s is already the clustered index\n", m_rgchName, pchIndexName, m_VecIndexes[iIndex].GetIndexName() ); return -1; } }
// find the field by name
int iField = FindIField( pchName ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchName, m_rgchName );
// add an index to it
int iIndex = AddIndexToFieldNumber( iField, pchIndexName, true /* clustered */ ); m_VecIndexes[iIndex].SetFillFactor( nFillFactor ); return iIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given set of fields as having a clustered index.
// Input: pchIndexName - Name of the index object (not in SQL)
// pchName - Name of the fields to index; comma separated if multiple
//-----------------------------------------------------------------------------
int CSchema::ClusteredIndexFields( int nFillFactor, const char *pchIndexName, const char *pchNames ) { Assert( pchNames != NULL ); Assert( pchIndexName != NULL );
// can't previously have some other clustered index
FOR_EACH_VEC( m_VecIndexes, iIndex ) { if ( m_VecIndexes[iIndex].IsClustered() ) { AssertFatalMsg3( false, "On table %s, can't make index %s clustered because %s is already the clustered index\n", m_rgchName, pchIndexName, m_VecIndexes[iIndex].GetIndexName() ); return -1; } }
// go add all those fields as Indexed
int nNewIndex = AddIndexToFieldList( pchNames, pchIndexName, k_nColFlagClustered | k_nColFlagIndexed, nFillFactor ); return nNewIndex; }
//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
void CSchema::AddFullTextIndex( CSchemaFull *pSchemaFull, const char *pchCatalogName, const char *pchColumnName ) { Assert( pchCatalogName != NULL ); Assert( pchColumnName != NULL );
// need a copy of string list since the argument is const and read-only
int cNamesLen = Q_strlen( pchColumnName ) + 1; char* pchNamesCopy = (char*) PvAlloc( cNamesLen ); Q_strncpy( pchNamesCopy, pchColumnName, cNamesLen );
if ( pchNamesCopy == NULL ) { // not enough memory!
AssertFatal( false ); return; }
char* pchCurrent = pchNamesCopy; do { // find next token
char* pchEnd = strchr(pchCurrent, ','); if ( pchEnd != NULL ) { *pchEnd++ = 0; }
// skip whitespace
while (isspace(*pchCurrent)) pchCurrent++;
// find the name; we expect a C++ name, not a SQL name
int iField = FindIField( pchCurrent ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchCurrent, m_rgchName );
// add it to our collection
m_VecFullTextIndexes.AddToTail( iField );
// move past the end of this token in the string
pchCurrent = pchEnd; } while ( pchCurrent != NULL );
// release our copy
FreePv(pchNamesCopy);
// make a note of the catalog we want
int nCatalogID = pSchemaFull->GetFTSCatalogByName( m_eSchemaCatalog, pchCatalogName ); AssertFatalMsg2( nCatalogID != -1, "Could not find fulltext catalog \"%s\" on table \"%s\"", pchCatalogName, m_rgchName ); m_nFullTextIndexCatalog = nCatalogID;
return; }
//-----------------------------------------------------------------------------
// Purpose: Marks a field as indexed given its field number
//-----------------------------------------------------------------------------
int CSchema::AddIndexToFieldNumber( int iField, const char *pchIndexName, bool bClustered ) { // mark the field as indexed
m_VecField[iField].m_nColFlags |= k_nColFlagIndexed;
// meant to be clustered?
if ( bClustered ) m_VecField[iField].m_nColFlags |= k_nColFlagClustered;
// add to list of indexes, and remember the indexID
CUtlVector<int> vecColumns; vecColumns.AddToTail( iField );
// false: not unique
// false: not clustered
FieldSet_t vecIndex( false, bClustered, vecColumns, pchIndexName); int iIndex = m_VecIndexes.AddToTail( vecIndex );
return iIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as unique.
// Input: pchName - Name of the field to index
// pchIndexName - Name of the index object
//-----------------------------------------------------------------------------
int CSchema::UniqueField( const char *pchIndexName, const char *pchName ) { Assert( pchName != NULL ); Assert( pchIndexName != NULL );
int iField = FindIField( pchName ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchName, m_rgchName );
m_VecField[iField].m_nColFlags |= ( k_nColFlagUnique | k_nColFlagIndexed );
// add to list of indexes, and remember the indexID
CUtlVector<int> vecColumns; vecColumns.AddToTail( iField ); // true: unique
// false: not clustered
FieldSet_t vecIndex( true, false, vecColumns, pchIndexName ); int iIndex = m_VecIndexes.AddToTail( vecIndex );
return iIndex; }
//-----------------------------------------------------------------------------
// Purpose: Marks a given field as auto increment.
// Input: pchName - Name of the field
//-----------------------------------------------------------------------------
void CSchema::AutoIncrementField( char *pchName ) { int iField = FindIField( pchName ); AssertFatalMsg2( k_iFieldNil != iField, "Could not find index column \"%s\" on table \"%s\"", pchName, m_rgchName ); m_VecField[iField].m_nColFlags |= k_nColFlagAutoIncrement; }
//-----------------------------------------------------------------------------
// Purpose: Finds the field with a given name.
// Input: pchName - Name of the field to search for
// Output: Index of the matching field (k_iFieldNil if there isn't one)
//-----------------------------------------------------------------------------
int CSchema::FindIField( const char *pchName ) { for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { if ( 0 == Q_strncmp( pchName, m_VecField[iField].m_rgchName, k_cSQLObjectNameMax ) ) return iField; }
return k_iFieldNil; }
//-----------------------------------------------------------------------------
// Purpose: Finds the field with a given SQL name.
// Input: pchName - Name of the field to search for
// Output: Index of the matching field (k_iFieldNil if there isn't one)
//-----------------------------------------------------------------------------
int CSchema::FindIFieldSQL( const char *pchName ) { for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { if ( 0 == Q_strncmp( pchName, m_VecField[iField].m_rgchSQLName, k_cSQLObjectNameMax ) ) return iField; }
return k_iFieldNil; }
//-----------------------------------------------------------------------------
// Purpose: Adds a schema conversion instruction (for use in converting from
// a different Schema to this one).
//-----------------------------------------------------------------------------
void CSchema::AddDeleteField( const char *pchFieldName ) { DeleteField_t &deleteField = m_VecDeleteField[m_VecDeleteField.AddToTail()]; Q_strncpy( deleteField.m_rgchFieldName, pchFieldName, sizeof( deleteField.m_rgchFieldName ) ); }
//-----------------------------------------------------------------------------
// Purpose: Adds a schema conversion instruction (for use in converting from
// a different Schema to this one).
//-----------------------------------------------------------------------------
void CSchema::AddRenameField( const char *pchFieldNameOld, const char *pchFieldNameNew ) { RenameField_t &renameField = m_VecRenameField[m_VecRenameField.AddToTail()]; Q_strncpy( renameField.m_rgchFieldNameOld, pchFieldNameOld, sizeof( renameField.m_rgchFieldNameOld ) ); renameField.m_iFieldDst = FindIField( pchFieldNameNew ); Assert( k_iFieldNil != renameField.m_iFieldDst ); }
//-----------------------------------------------------------------------------
// Purpose: Adds a schema conversion instruction (for use in converting from
// a different Schema to this one).
// Input: pchFieldNameOld - Name of the field in the old schema
// pchFieldNameMew - Name of the field in the new schema
// pfnAlterField - Function to translate data from the old format to
// the new
//-----------------------------------------------------------------------------
void CSchema::AddAlterField( const char *pchFieldNameOld, const char *pchFieldNameNew, PfnAlterField_t pfnAlterField ) { Assert( pfnAlterField ); AlterField_t &alterField = m_VecAlterField[m_VecAlterField.AddToTail()]; Q_strncpy( alterField.m_rgchFieldNameOld, pchFieldNameOld, sizeof( alterField.m_rgchFieldNameOld ) ); alterField.m_iFieldDst = FindIField( pchFieldNameNew ); Assert( k_iFieldNil != alterField.m_iFieldDst ); alterField.m_pfnAlterFunc = pfnAlterField; }
//-----------------------------------------------------------------------------
// Purpose: Figures out how to map a field from another Schema into us.
// First we check our conversion instructions to see if any apply,
// and then we look for a straightforward match.
// Input: pchFieldName - Name of the field we're trying to map
// piFieldDst - [Return] Index of the field to map it to
// ppfnAlterField - [Return] Optional function to convert data
// Output: true if we know what to do with this field (if false, the conversion
// is undefined and dangerous).
//-----------------------------------------------------------------------------
bool CSchema::BCanConvertField( const char *pchFieldName, int *piFieldDst, PfnAlterField_t *ppfnAlterField ) { *ppfnAlterField = NULL;
// Should this field be deleted?
for ( int iDeleteField = 0; iDeleteField < m_VecDeleteField.Count(); iDeleteField++ ) { if ( 0 == Q_strcmp( pchFieldName, m_VecDeleteField[iDeleteField].m_rgchFieldName ) ) { *piFieldDst = k_iFieldNil; return true; } }
// Should this field be renamed?
for ( int iRenameField = 0; iRenameField < m_VecRenameField.Count(); iRenameField++ ) { if ( 0 == Q_strcmp( pchFieldName, m_VecRenameField[iRenameField].m_rgchFieldNameOld ) ) { *piFieldDst = m_VecRenameField[iRenameField].m_iFieldDst; return true; } }
// Was this field altered?
for ( int iAlterField = 0; iAlterField < m_VecAlterField.Count(); iAlterField++ ) { if ( 0 == Q_strcmp( pchFieldName, m_VecAlterField[iAlterField].m_rgchFieldNameOld ) ) { *piFieldDst = m_VecAlterField[iAlterField].m_iFieldDst; *ppfnAlterField = m_VecAlterField[iAlterField].m_pfnAlterFunc; return true; } }
// Find out which of our fields this field maps to (if it doesn't map
// to any of them, we don't know what to do with it).
*piFieldDst = FindIField( pchFieldName ); return ( k_iFieldNil != *piFieldDst ); }
//-----------------------------------------------------------------------------
// Purpose: Return the size to use when writing to a field.
//-----------------------------------------------------------------------------
int Field_t::CubFieldUpdateSize() const { switch ( m_EType ) { case k_EGCSQLType_String: case k_EGCSQLType_Blob: case k_EGCSQLType_Image: // Nobody should call this function for
// var-length fields
Assert( false ); return m_cubLength;
case k_EGCSQLType_int8: case k_EGCSQLType_int16: case k_EGCSQLType_int32: case k_EGCSQLType_int64: case k_EGCSQLType_float: case k_EGCSQLType_double: return m_cubLength;
default: Assert(false); return 0; } }
//-----------------------------------------------------------------------------
// Purpose: Tell whether or not a field is of string type.
//-----------------------------------------------------------------------------
bool Field_t::BIsStringType() const { return ( k_EGCSQLType_String == m_EType ); }
//-----------------------------------------------------------------------------
// Purpose: Tell whether or not the field is of variable-length type.
//-----------------------------------------------------------------------------
bool Field_t::BIsVariableLength() const { return ( k_EGCSQLType_String == m_EType ) || ( k_EGCSQLType_Blob == m_EType ) || ( k_EGCSQLType_Image == m_EType ); }
//-----------------------------------------------------------------------------
// Purpose: Add a foreign key
//-----------------------------------------------------------------------------
void CSchema::AddFK( const char* pchName, const char* pchColumn, const char* pchParentTable, const char* pchParentColumn, EForeignKeyAction eOnDeleteAction, EForeignKeyAction eOnUpdateAction ) { int iTail = m_VecFKData.AddToTail(); FKData_t &fkData = m_VecFKData[iTail];
Q_snprintf( fkData.m_rgchName, Q_ARRAYSIZE( fkData.m_rgchName ), "%s_%s", GetPchName(), pchName ); Q_strncpy( fkData.m_rgchParentTableName, pchParentTable, Q_ARRAYSIZE( fkData.m_rgchParentTableName ) ); fkData.m_eOnDeleteAction = eOnDeleteAction; fkData.m_eOnUpdateAction = eOnUpdateAction;
// Now we need to split up the column name strings and add their data
FKColumnRelation_t colRelation; Q_memset( &colRelation, 0, sizeof( FKColumnRelation_t ) ); uint iMyColumn = 0; uint iParentColumn = 0; uint iParentString = 0; for( uint i=0; i<(uint)Q_strlen( pchColumn )+1; ++i ) { if ( pchColumn[i] != ',' && pchColumn[i] != 0 ) { colRelation.m_rgchCol[ iMyColumn++ ] = pchColumn[i]; } else { Assert( Q_strlen( colRelation.m_rgchCol ) ); // Should have a matching column name in the parent string
while( iParentString < (uint)Q_strlen( pchParentColumn ) ) { if ( pchParentColumn[iParentString] != ',' ) { colRelation.m_rgchParentCol[ iParentColumn++ ] = pchParentColumn[iParentString]; ++iParentString; } else { ++iParentString; break; } }
AssertMsg( Q_strlen( colRelation.m_rgchParentCol ), "Column counts for FK do not match between child/parent\n" ); fkData.m_VecColumnRelations.AddToTail( colRelation ); Q_memset( &colRelation, 0, sizeof( FKColumnRelation_t ) ); // Reset positions
iMyColumn = 0; iParentColumn = 0; } } }
//-----------------------------------------------------------------------------
// Purpose: Caches the insert statement for each table so we don't need to
// generate it for each row we insert.
//-----------------------------------------------------------------------------
const char *CSchema::GetInsertStatementText() const { if ( m_sInsertStatementText.IsEmpty() ) { TSQLCmdStr sBuilder; BuildInsertStatementText( &sBuilder, GetRecordInfo() ); m_sInsertStatementText.Set( sBuilder ); }
return m_sInsertStatementText; }
//-----------------------------------------------------------------------------
// Purpose: Caches the insert via MERGE statement for each table so we don't need to
// generate it for each row we insert.
//-----------------------------------------------------------------------------
const char *CSchema::GetMergeStatementTextOnPKWhenMatchedUpdateWhenNotMatchedInsert() { if ( m_sMergeStatementTextOnPKWhenMatchedUpdateWhenNotMatchedInsert.IsEmpty() ) { TSQLCmdStr sBuilder; BuildMergeStatementTextOnPKWhenMatchedUpdateWhenNotMatchedInsert( &sBuilder, GetRecordInfo() ); m_sMergeStatementTextOnPKWhenMatchedUpdateWhenNotMatchedInsert.Set( sBuilder ); }
return m_sMergeStatementTextOnPKWhenMatchedUpdateWhenNotMatchedInsert; }
//-----------------------------------------------------------------------------
// Purpose: Caches the insert via MERGE statement for each table so we don't need to
// generate it for each row we insert.
//-----------------------------------------------------------------------------
const char *CSchema::GetMergeStatementTextOnPKWhenNotMatchedInsert() { if ( m_sMergeStatementTextOnPKWhenNotMatchedInsert.IsEmpty() ) { TSQLCmdStr sBuilder; BuildMergeStatementTextOnPKWhenNotMatchedInsert( &sBuilder, GetRecordInfo() ); m_sMergeStatementTextOnPKWhenNotMatchedInsert.Set( sBuilder ); }
return m_sMergeStatementTextOnPKWhenNotMatchedInsert; }
//-----------------------------------------------------------------------------
// Purpose: Get the number of foreign key constraints defined for the table
//-----------------------------------------------------------------------------
int CSchema::GetFKCount() { return m_VecFKData.Count(); }
//-----------------------------------------------------------------------------
// Purpose: Get data for a foreign key by index (valid for 0...GetFKCount()-1)
//-----------------------------------------------------------------------------
FKData_t &CSchema::GetFKData( int iIndex ) { return m_VecFKData[iIndex]; }
#ifdef DBGFLAG_VALIDATE
//-----------------------------------------------------------------------------
// Purpose: Run a global validation pass on all of our data structures and memory
// allocations.
// Input: validator - Our global validator object
// pchName - Our name (typically a member var in our container)
//-----------------------------------------------------------------------------
void CSchema::Validate( CValidator &validator, const char *pchName ) { // 1.
// Claim our memory
VALIDATE_SCOPE();
m_VecField.Validate( validator, "m_VecField" ); m_VecDeleteField.Validate( validator, "m_VecDeleteField" ); m_VecRenameField.Validate( validator, "m_VecRenameField" ); m_VecIndexes.Validate( validator, "m_VecIndexes" ); m_VecFullTextIndexes.Validate( validator, "m_VecFullTextIndexes" ); ValidateObj( m_VecFKData ); FOR_EACH_VEC( m_VecFKData, i ) { ValidateObj( m_VecFKData[i] ); }
for ( int iIndex = 0; iIndex < m_VecIndexes.Count(); iIndex++ ) { ValidateObj( m_VecIndexes[iIndex] ); }
ValidateObj( m_sInsertStatementText );
// 2.
// Validate that our fields make sense
#if defined(_DEBUG)
uint32 dubOffset = 0; for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { Assert( dubOffset == m_VecField[iField].m_dubOffset ); dubOffset += m_VecField[iField].m_cubLength; } #endif // defined(_DEBUG)
}
//-----------------------------------------------------------------------------
// Purpose: Validates that a given record from our table is in a good state.
// Input: pubRecord - Record to validate
//-----------------------------------------------------------------------------
void CSchema::ValidateRecord( uint8 *pubRecord ) { // Make sure each record is in a consistent state
for ( int iField = 0; iField < m_VecField.Count(); iField++ ) { Field_t &field = m_VecField[iField];
// Ensure that strings are null-terminated, and that everything after the terminator is 0
if ( k_EGCSQLType_String == field.m_EType ) { char *pchField = ( char * ) pubRecord + field.m_dubOffset; Assert( 0 == pchField[field.m_cubLength - 1] ); for ( uint32 ich = 0; ich < field.m_cubLength; ich++ ) { if ( 0 == pchField[ich] ) { while ( ich < field.m_cubLength ) { Assert( 0 == pchField[ich] ); ich++; } } } } } } #endif // DBGFLAG_VALIDATE
} // namespace GCSDK
|