windows-nt/Source/XPSP1/NT/enduser/stuff/itircl/property/rsimp.cpp

1482 lines
41 KiB
C++
Raw Normal View History

2020-09-26 03:20:57 -05:00
/*************************************************************************
* @doc SHROOM EXTERNAL API *
* *
* RSIMP.CPP *
* *
* Copyright (C) Microsoft Corporation 1997 *
* All Rights reserved. *
* *
* This file contains the implementation of the result set object *
* *
* *
**************************************************************************
* *
* Written By : Erin Foxford *
* Current Owner: erinfox *
* *
**************************************************************************/
#include <mvopsys.h>
#ifdef _DEBUG
static char s_aszModule[] = __FILE__; /* For error report */
#endif
#include <_mvutil.h>
#include <atlinc.h> // Includes for ATL
#include <itpropl.h>
#include <itrs.h>
#include <orkin.h>
#include "rsimp.h"
#include "prop.h"
#include "plist.h"
#define THEORETICAL_MAX_PROP 30 // This would be a lot of properties - Assert only
const unsigned int CHUNK_SIZE = 512; // How many "chunks" of rows that can be allocated
const unsigned int ROW_CHUNK = 1024; // Number of rows allocated per chunk
const unsigned int ROW_CHUNK_LESS1 = ROW_CHUNK - 1; // Number of rows minus 1
// optimization for Real % ROW_CHUNK
#define RealToLogical(Real, Logical) (Logical = Real & ROW_CHUNK_LESS1)
static unsigned int g_iAlloc = 0;
static unsigned int g_iFreed = 0;
CITResultSet::CITResultSet() : m_PageMap(NULL), m_hResultSet(NULL),
m_AppendRow(0), m_cProp(0), m_NumberOfPages(0),
m_fInit(FALSE), m_RowsReserved(0), m_Chunk(-1)
{
// Allocate memory for data pool - if allocation fails,
// how do we notify user?
m_pMemPool = BlockInitiate((DWORD)16384, 0, 0, 0);
ITASSERT (NULL != m_pMemPool);
// Allocate an array of DWORD pointers. These pointers will point to the virtual memory
// space of the result set
m_hResultSet = _GLOBALALLOC(GMEM_MOVEABLE | GMEM_ZEROINIT, CHUNK_SIZE*sizeof(DWORD*));
ITASSERT (NULL != m_hResultSet);
if (m_hResultSet)
m_ResultSet = (DWORD_PTR**)_GLOBALLOCK(m_hResultSet);
// The number of chunks allocated in one shot
m_NumChunks = CHUNK_SIZE;
}
CITResultSet::~CITResultSet()
{
// Free up virtual memory
Clear();
// Free memory pool
if (m_pMemPool)
{
BlockFree(m_pMemPool);
m_pMemPool = NULL;
}
// Free result set array
if (m_hResultSet)
{
_GLOBALUNLOCK(m_hResultSet);
_GLOBALFREE(m_hResultSet);
m_hResultSet = NULL;
}
}
// We wrap VirtualAlloc(.... MEM_RESERVE) so in the future we can
// make this nothing on the Mac
//
// NOTE: Uses #define PAGE_SIZE, which is defined in mvopsys.h.
// Currently it's 8k for Alpha, 4k for everything else
//
HRESULT WINAPI CITResultSet::Reserve()
{
#ifdef _DEBUG
SYSTEM_INFO si;
GetSystemInfo(&si);
ITASSERT(PAGE_SIZE == si.dwPageSize);
#endif
m_Chunk++;
// we have to realloc...
if (m_Chunk == m_NumChunks)
{
// We're allocating a BUNCH of memory here. If we hit this code often, we need
// to redesign our allocation scheme.
ITASSERT(0);
_GLOBALUNLOCK(m_hResultSet);
m_NumChunks += CHUNK_SIZE;
m_hResultSet = _GLOBALREALLOC(m_hResultSet, m_NumChunks*sizeof(DWORD*), GMEM_MOVEABLE | GMEM_ZEROINIT);
if (m_hResultSet)
{
m_ResultSet = (DWORD_PTR**) _GLOBALLOCK(m_hResultSet);
}
else
return SetErrReturn(E_OUTOFMEMORY);
}
// Calculate some info we'll need later on
if (!m_fInit)
{
// how many rows fit into a page of memory?
m_RowsPerPage = PAGE_SIZE/(m_cProp* sizeof(DWORD *));
// the number of pages in a chunk of rows
m_NumberOfPages = (m_cProp * ROW_CHUNK * sizeof(DWORD *))/PAGE_SIZE + 1;
// allocation size in bytes.
m_BytesReserved = m_NumberOfPages*PAGE_SIZE;
if (NULL == (m_PageMap = new BOOL[m_NumberOfPages]))
return SetErrReturn(E_OUTOFMEMORY);
m_fInit = TRUE;
}
// set page map to 0 - this keeps track of which pages are allocated
MEMSET(m_PageMap, 0, m_NumberOfPages*sizeof(BOOL));
// reserve memory for an entire chunk - this call shouldn't happen very frequently.
if (NULL == (m_ResultSet[m_Chunk] = (DWORD_PTR *) VirtualAlloc(NULL, m_BytesReserved,
MEM_RESERVE, PAGE_READWRITE)))
{
#ifdef _DEBUG
int Error = GetLastError( );
DPF1("CITResultSet::Reserve: MEM_RESERVE faild: %u", Error);
#endif // _DEBUG
return SetErrReturn(E_OUTOFMEMORY);
}
#ifdef _DEBUG
MEMORY_BASIC_INFORMATION MemInfo;
VirtualQuery(m_ResultSet[m_Chunk], &MemInfo, sizeof(MEMORY_BASIC_INFORMATION));
#endif // _DEBUG
m_RowsReserved += ROW_CHUNK;
return S_OK;
}
// Wrapper for VirtualAlloc(... MEM_COMMIT...)
// One possible optimization: replace w/ SEH instead
// of maintaining a page map
HRESULT WINAPI CITResultSet::Commit(LONG RowNum)
{
LONG LogicalRowNum;
LONG PageNum;
RealToLogical(RowNum, LogicalRowNum);
PageNum = LogicalRowNum/m_RowsPerPage;
// Only allocate if page hasn't been committed before
// Note that allocated memory gets zero'd
if (FALSE == m_PageMap[PageNum])
{
// this will allocate a full page
if (NULL == VirtualAlloc(&m_ResultSet[m_Chunk][ROW_CHUNK * PageNum], PAGE_SIZE,
MEM_COMMIT, PAGE_READWRITE) )
{
// TODO: map relevant Win32 error to HRESULT... for now I'm assuming
// we ran out of memory
#ifdef _DEBUG
int Error = GetLastError( );
DPF1("CITResultSet::Reserve: MEM_COMMIT faild: %u", Error);
#endif // _DEBUG
return SetErrReturn(E_OUTOFMEMORY);
}
g_iAlloc++;
m_PageMap[PageNum] = TRUE;
}
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Add |
* Adds columns to result set, given a header containing pairs
* of property ID followed by property type
*
* @parm LPVOID | lpvHdr | Buffer containing property ID/property pairs
*
* @rvalue S_OK | The columns were successfully added
*
* @xcomm
* The format of lpvHdr is identical to the output format of
* PorpertyList::SaveHeader. If PL:SaveHeader changes we
* may need to modify this code to maintain compatability.
*
********************************************************************/
STDMETHODIMP CITResultSet::Add(LPVOID lpvHdr)
{
LPBYTE pCurHdr = (LPBYTE) lpvHdr;
DWORD dwPropID;
DWORD dwType;
DWORD dwProps;
DWORD dwMax;
DWORD iProp; // loop index
m_cs.Lock();
// Number of properties in header
MEMCPY(&dwProps, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
dwMax = m_cProp + dwProps;
if (dwMax >= MAX_COLUMNS)
{
m_cs.Unlock();
return (SetErrReturn(E_TOOMANYCOLUMNS));
}
// For each property in header, create a column
for (iProp = m_cProp; iProp < dwMax; iProp++)
{
// clear header
MEMSET(&m_Header[iProp], NULL, sizeof(CHeader));
MEMCPY(&dwPropID, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
MEMCPY(&dwType, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
m_Header[iProp].dwPropID = dwPropID;
m_Header[iProp].lpvData = NULL;
m_Header[iProp].dwType = dwType;
m_Header[iProp].Priority = PRIORITY_NORMAL; // default priority
} // end for over properties in header
m_cProp += dwProps;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Add |
* Adds a column to the result set
*
* @parm PROPID | PropID | Property ID associated with column
* @parm DWORD | dwDefautData | Default data value
* @parm PRIORITY | Priority | Download priority of column (PRIORITY_LOW, PRIORITY_NORMAL,
* or PRIORITY_HIGH)
*
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The column was successfully added
*
* @comm This method is used to add a column for numerical properties.
********************************************************************/
STDMETHODIMP CITResultSet::Add(PROPID PropID, DWORD dwDefaultData, PRIORITY Priority)
{
if (m_cProp >= MAX_COLUMNS)
return (SetErrReturn(E_TOOMANYCOLUMNS));
m_cs.Lock();
// clear header
MEMSET(&m_Header[m_cProp], NULL, sizeof(CHeader));
// copy data
m_Header[m_cProp].dwPropID = PropID;
m_Header[m_cProp].dwValue = dwDefaultData;
m_Header[m_cProp].dwType = TYPE_VALUE;
m_Header[m_cProp].Priority = Priority; // default priority
m_cProp++;
// Wow! This is a lot of properties!
ITASSERT (m_cProp < THEORETICAL_MAX_PROP);
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Add |
* Adds a column to the result set
*
* @parm PROPID | PropID | Property ID associated with column
* @parm LPCWSTR | lpDefaultData | Default value of string
* @parm PRIORITY | Priority | Download priority of column (PRIORITY_LOW, PRIORITY_NORMAL,
* or PRIORITY_HIGH)
*
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The column was successfully added
*
* @comm This method is used to add a column for string properties.
********************************************************************/
STDMETHODIMP CITResultSet::Add(PROPID PropID, LPCWSTR lpszwDefault, PRIORITY Priority)
{
if (m_cProp >= MAX_COLUMNS)
return (SetErrReturn(E_TOOMANYCOLUMNS));
m_cs.Lock();
// clear header
MEMSET(&m_Header[m_cProp], NULL, sizeof(CHeader));
// copy data
if (lpszwDefault)
{
LPBYTE pBuffer;
DWORD cbData = (DWORD)(WSTRLEN(lpszwDefault) + 1) * sizeof(WCHAR);
if (NULL == (pBuffer = (LPBYTE) BlockCopy(m_pMemPool, NULL, cbData + 4, 0)))
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
MEMCPY(pBuffer + sizeof (DWORD), lpszwDefault, cbData);
*(LPDWORD)pBuffer = cbData;
m_Header[m_cProp].lpvData = pBuffer;
}
else
m_Header[m_cProp].lpvData = NULL;
m_Header[m_cProp].dwPropID = PropID;
m_Header[m_cProp].dwType = TYPE_STRING;
m_Header[m_cProp].Priority = Priority; // default priority
m_cProp++;
// Wow! This is a lot of properties!
ITASSERT (m_cProp < THEORETICAL_MAX_PROP);
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Add |
* Adds a column to the result set
*
* @parm PROPID | PropID | Property ID associated with column
* @parm LPVOID | lpvDefaultData | Buffer containing default value of data
* @parm DWORD | cbData | Size of buffer
* @parm PRIORITY | Priority | Download priority of column (PRIORITY_LOW, PRIORITY_NORMAL,
* or PRIORITY_HIGH)
*
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The column was successfully added
*
* @comm This method is used to add a column for pointer properties.
********************************************************************/
STDMETHODIMP CITResultSet::Add(PROPID PropID, LPVOID lpvDefaultData, DWORD cbData, PRIORITY Priority)
{
if (m_cProp >= MAX_COLUMNS)
return (SetErrReturn(E_TOOMANYCOLUMNS));
m_cs.Lock();
// clear header
MEMSET(&m_Header[m_cProp], NULL, sizeof(CHeader));
// copy data
if (lpvDefaultData)
{
LPBYTE pBuffer;
if (NULL == (pBuffer = (LPBYTE) BlockCopy(m_pMemPool, NULL, cbData + 4, 0)))
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
MEMCPY(pBuffer + 4, lpvDefaultData, cbData);
*(LPDWORD)pBuffer = cbData;
m_Header[m_cProp].lpvData = pBuffer;
}
else
m_Header[m_cProp].lpvData = NULL;
m_Header[m_cProp].dwPropID = PropID;
m_Header[m_cProp].dwType = TYPE_POINTER;
m_Header[m_cProp].Priority = Priority; // default priority
m_cProp++;
// Wow! This is a lot of properties!
ITASSERT (m_cProp < THEORETICAL_MAX_PROP);
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | SetColumnPriority |
* Sets the download priority for the given column in the result set
*
* @parm LONG | lColumnIndex | Index of column to set
* @parm PRIORITY | Priority | Priority, which can be one of the following:
*
* @flag PRIORITY_LOW | Low priority
* @flag PRIORITY_NORMAL | Normal priority
* @flag PRIORITY_HIGH | High priority
*
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue S_OK | The priority was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::SetColumnPriority(LONG lColumnIndex, PRIORITY ColumnPriority)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
m_cs.Lock();
m_Header[lColumnIndex].Priority = ColumnPriority;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | SetColumnHeap |
* Sets the heap which DWORD values in this column point into.
*
* @parm LONG | lColumnIndex | Index of column to set
* @parm LPVOID | lpvHeap | Pointer to the heap.
* @parm PFNCOLHEAPFREE | pfnColHeapFree |
* Pointer to a function which can be called to free the heap
* when the result set is cleared or freed.
*
*
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue S_OK | The heap was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::SetColumnHeap(LONG lColumnIndex, LPVOID lpvHeap,
PFNCOLHEAPFREE pfnColHeapFree)
{
HRESULT hr = S_OK;
if (lColumnIndex < 0)
return SetErrReturn(E_INVALIDARG);
m_cs.Lock();
if (lColumnIndex >= m_cProp)
hr = E_NOTEXIST;
else
if (m_Header[lColumnIndex].lpvHeap != NULL)
hr = E_ALREADYINIT;
if (SUCCEEDED(hr))
{
m_Header[lColumnIndex].lpvHeap = lpvHeap;
m_Header[lColumnIndex].pfnHeapFree = pfnColHeapFree;
}
m_cs.Unlock();
return (hr);
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetColumnPriority |
* Gets the download priority for the given column in the result set
*
* @parm LONG | lColumnIndex | Index of column to get
* @parm PRIORITY& | Priority | Priority, which can be one of the following:
*
* @flag PRIORITY_LOW | Low priority
* @flag PRIORITY_NORMAL | Normal priority
* @flag PRIORITY_HIGH | High priority
*
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue S_OK | The priority was successfully retrieved
********************************************************************/
STDMETHODIMP CITResultSet::GetColumnPriority(LONG lColumnIndex, PRIORITY& ColumnPriority)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
ColumnPriority = m_Header[lColumnIndex].Priority;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | SetKeyProp |
* Sets the property to be used as the key.
*
* @parm PROPID | KeyPropID | Property ID
*
* @rvalue S_OK | The key property was successfully set
*
********************************************************************/
inline STDMETHODIMP CITResultSet::SetKeyProp(PROPID KeyPropID)
{
m_cs.Lock();
m_dwKeyProp = KeyPropID;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetKeyProp |
* Retrieves the property used as the key
*
* @parm PROPID | KeyPropID | Property ID
*
* @rvalue S_OK | The key property was successfully retrieved
*
********************************************************************/
inline STDMETHODIMP CITResultSet::GetKeyProp(PROPID& KeyPropID)
{
KeyPropID = m_dwKeyProp;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Append |
* Given header (prop ID and type) and data, this function forms a
* row and appends it to the current result set
*
* @parm LPVOID | lpvHdr | Pointer to buffer containing header
* @parm LPVOID | lpvData | Pointer to buffer containing data
*
* @rvalue E_INVALIDARG | Exceeded maximum number of rows
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_FALSE | No columns were found to set, but no failure occurred
* @rvalue S_OK | The row was successfully filled in
*
* @xcomm
* The format of lpvHdr is identical to the output format of
* PorpertyList::SaveData. If PL:Savedata changes we may
* need to modify this code to maintain compatability.
*
********************************************************************/
STDMETHODIMP CITResultSet::Append(LPVOID lpvHdr, LPVOID lpvData)
{
return Set(m_AppendRow, lpvHdr, lpvData);
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Set |
* Given header (prop ID and type) and data, this function puts
* the information into the specified row in the result set
*
* @parm LONG | lRowIndex | Index of row in result set
* @parm LPVOID | lpvHdr | Pointer to buffer containing header
* @parm LPVOID | lpvData | Pointer to buffer containing data
*
* @rvalue E_INVALIDARG | Exceeded maximum number of rows
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_FALSE | No columns were found to set, but no failure occurred
* @rvalue S_OK | The row was successfully filled in
*
* @xcomm
* The format of lpvHdr is identical to the output format of
* PorpertyList::SaveData. If PL:Savedata changes we may
* need to modify this code to maintain compatability.
*
********************************************************************/
STDMETHODIMP CITResultSet::Set(LONG lRowIndex, LPVOID lpvHdr, LPVOID lpvData)
{
LPBYTE pCurHdr = (LPBYTE) lpvHdr;
LPBYTE pCurData = (LPBYTE) lpvData;
LPBYTE pBitField = (LPBYTE) lpvData;
DWORD dwProps;
DWORD dwPropID;
DWORD dwType;
LPBYTE pBuffer;
DWORD cbSize;
LONG lColumn;
HRESULT hr = S_FALSE;
m_cs.Lock();
// Reserve memory if necessary
if (lRowIndex >= m_RowsReserved)
{
if ( FAILED(Reserve()) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
}
// Commit memory
if ( FAILED(Commit(lRowIndex)) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
// Number of properties in header
MEMCPY(&dwProps, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
// Shift pCurData up to leave room for the property bit field
pCurData += (dwProps/8 + 1);
// For each property in header, look for matching result set column
for (DWORD iProp = 0; iProp < dwProps; iProp++)
{
MEMCPY(&dwPropID, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
MEMCPY(&dwType, pCurHdr, sizeof(DWORD));
pCurHdr += sizeof(DWORD);
// UNDONE: Figure out an optimization. If columns match up w/ header,
// then by the time we get to the end of the header, we'll have to loop
// over ALL the columns just to find which the column that matches the
// property ID. The problem is, it's not guaranteed that the columns
// match up exactly; the user could set them in any order.
// Is there data for this property?
BYTE WhichBit = (0x80 >> (iProp % 8));
int BitSet = pBitField[iProp/8] & WhichBit;
HRESULT hrFound = GetColumnFromPropID(dwPropID, lColumn);
if (SUCCEEDED(hrFound))
hr = S_OK; // found at least one column
if (BitSet)
{
if (TYPE_VALUE == dwType)
{
// copy data value
if (SUCCEEDED(hrFound))
m_ResultSet[nChunk][nRow + lColumn] = *(LPDWORD) pCurData;
// if user didn't specify column, we still need to skip the data for
// this property
pCurData += sizeof(DWORD);
}
else
{
// get size
MEMCPY(&cbSize, pCurData, sizeof(DWORD));
pCurData += sizeof(DWORD);
if (SUCCEEDED(hrFound))
{
if (NULL == (pBuffer = (LPBYTE) BlockCopy(m_pMemPool, NULL, cbSize + 4, 0)))
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
// append size to data
MEMCPY(pBuffer + 4, pCurData, cbSize);
*(LPDWORD)pBuffer = cbSize;
m_ResultSet[nChunk][nRow + lColumn] = (DWORD_PTR) pBuffer;
}
pCurData += cbSize;
}
}
else
{
// No, so use default
if (SUCCEEDED(hrFound))
m_ResultSet[nChunk][nRow + lColumn] = (DWORD_PTR) m_Header[lColumn].lpvData;
}
} // end for over properties in header
// Always maintain one past the last row as the append row
// unless we didn't put anything in the result set
if ( (lRowIndex >= m_AppendRow) && (S_OK == hr))
m_AppendRow = lRowIndex + 1;
m_cs.Unlock();
return hr;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Set |
* Sets the property in the specified row to the property value.
*
* @parm LONG | lRowIndex | Row in which property belongs
* @parm LONG | lColumnIndex | Column in which property belongs
* @parm DWORD | dwData | Data to set
*
* @rvalue E_INVALIDARG | Exceed maximum row count
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The row was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::Set(LONG lRowIndex, LONG lColumnIndex, DWORD dwData)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
m_cs.Lock();
// Reserve memory if necessary
if (lRowIndex >= m_RowsReserved)
{
if ( FAILED(Reserve()) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
}
// Commit memory
if ( FAILED(Commit(lRowIndex)) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
m_ResultSet[nChunk][nRow + lColumnIndex] = dwData;
// always maintain one past the last row as the append row
if (lRowIndex >= m_AppendRow)
m_AppendRow = lRowIndex + 1;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Set |
* Sets the property in the specified row to the property value.
*
* @parm LONG | lRowIndex | Row in which property belongs
* @parm LONG | lColumnIndex | Column in which property belongs
* @parm DWORD | dwData | Data to set
*
* @rvalue E_INVALIDARG | Exceed maximum row count
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The row was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::Set(LONG lRowIndex, LONG lColumnIndex, DWORD_PTR dwData)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
m_cs.Lock();
// Reserve memory if necessary
if (lRowIndex >= m_RowsReserved)
{
if ( FAILED(Reserve()) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
}
// Commit memory
if ( FAILED(Commit(lRowIndex)) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
m_ResultSet[nChunk][nRow + lColumnIndex] = dwData;
// always maintain one past the last row as the append row
if (lRowIndex >= m_AppendRow)
m_AppendRow = lRowIndex + 1;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Set |
* Sets the property in the specified row to the property value.
*
* @parm LONG | lRowIndex | Row in which property belongs
* @parm LPCWSTR | lpszString | Data to set
*
* @rvalue E_INVALIDARG | Exceed maximum row count
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The row was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::Set(LONG lRowIndex, LONG lColumnIndex, LPCWSTR lpszString)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
m_cs.Lock();
// Reserve memory if necessary
if (lRowIndex >= m_RowsReserved)
{
if ( FAILED(Reserve()) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
}
// Commit memory
if ( FAILED(Commit(lRowIndex)) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
DWORD cbData = 0;
if (lpszString)
cbData = (DWORD) (2*(WSTRLEN(lpszString) + 1));
LPBYTE pBuffer = (LPBYTE) BlockCopy(m_pMemPool, NULL, cbData + sizeof (DWORD), 0);
if (NULL == pBuffer)
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
MEMCPY(pBuffer + 4, lpszString, cbData);
*(LPDWORD)pBuffer = cbData;
m_ResultSet[nChunk][nRow + lColumnIndex] = (DWORD_PTR) pBuffer;
// always maintain one past the last row as the append row
if (lRowIndex >= m_AppendRow)
m_AppendRow = lRowIndex + 1;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Set |
* Sets the property in the specified row to the property value.
*
* @parm LONG | lRowIndex | Row in which property belongs
* @parm DWORD | dwData | Data to set
*
* @rvalue E_INVALIDARG | Exceed maximum row count
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue E_OUTOFMEMORY | Memory allocation failed
* @rvalue S_OK | The row was successfully set
********************************************************************/
STDMETHODIMP CITResultSet::Set(LONG lRowIndex, LONG lColumnIndex, LPVOID lpvData, DWORD cbData)
{
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
m_cs.Lock();
// Reserve memory if necessary
if (lRowIndex >= m_RowsReserved)
{
if ( FAILED(Reserve()) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
}
// Commit memory
if ( FAILED(Commit(lRowIndex)) )
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
LPBYTE pBuffer = (LPBYTE) BlockCopy(m_pMemPool, NULL, cbData + sizeof (DWORD), 0);
if (NULL == pBuffer)
{
m_cs.Unlock();
return SetErrReturn(E_OUTOFMEMORY);
}
*(LPDWORD)pBuffer = cbData;
MEMCPY(pBuffer + sizeof (DWORD), lpvData, cbData);
m_ResultSet[nChunk][nRow + lColumnIndex] = (DWORD_PTR) pBuffer;
// always maintain one past the last row as the append row
if (lRowIndex >= m_AppendRow)
m_AppendRow = lRowIndex + 1;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Get |
* Gets the property in the specified row and column and fills the given
* property object.
*
* @parm LONG | lRowIndex | Row in which property belongs
* @parm LONG | lColumnIndex | Column in which property belongs
* @parm CProperty& | Prop | Property object to fill
*
* @rvalue E_NOTEXIST | The row or column does not exist in the row set
* @rvalue S_OK | The row was successfully retrieved
*
********************************************************************/
STDMETHODIMP CITResultSet::Get(LONG lRowIndex, LONG lColumnIndex, CProperty& Prop)
{
if (lRowIndex >= m_AppendRow || lColumnIndex >= m_cProp)
return SetErrReturn(E_NOTEXIST);
LONG LogicalRow;
RealToLogical(lRowIndex, LogicalRow);
LONG nRow = LogicalRow * m_cProp;
LONG nChunk = lRowIndex/ROW_CHUNK;
Prop.dwPropID = m_Header[lColumnIndex].dwPropID;
if (TYPE_VALUE == (Prop.dwType = m_Header[lColumnIndex].dwType))
{
// For data types, we have no way of knowing how to return
// the default, so we just return 0 if this cell was never filled in
Prop.lpvData = (LPVOID) m_ResultSet[nChunk][nRow+lColumnIndex];
Prop.dwValue = (DWORD) m_ResultSet[nChunk][nRow+lColumnIndex];
Prop.cbData = sizeof(DWORD);
}
else
{
LPBYTE pBuffer = (LPBYTE) m_ResultSet[nChunk][nRow+lColumnIndex];
if (pBuffer)
{
Prop.cbData = *(LPDWORD)pBuffer;
Prop.lpvData = pBuffer + sizeof (DWORD);
}
else
{
// there's nothing there, so return default
if (m_Header[lColumnIndex].lpvData)
{
Prop.cbData = *(LPDWORD) m_Header[lColumnIndex].lpvData;
Prop.lpvData = (LPDWORD)(m_Header[lColumnIndex].lpvData) + 1;
}
else
{
// default was specified as NULL, so we make sure we
// return that
Prop.cbData = 0;
Prop.lpvData = NULL;
}
}
}
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetColumnCount |
* Gets number of columns in result set
*
* @parm LONG& | lNumberOfColumns | Number of columns
*
* @rvalue S_OK | The number of columns was successfully retrieved
*
********************************************************************/
inline STDMETHODIMP CITResultSet::GetColumnCount(LONG& lNumberOfColumns)
{
lNumberOfColumns = m_cProp;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetRowCount |
* Gets number of rows in result set
*
* @parm LONG& | lNumberOfRows | Number of rows
*
* @rvalue S_OK | The number of rows was successfully retrieved
*
********************************************************************/
inline STDMETHODIMP CITResultSet::GetRowCount(LONG& lNumberOfRows)
{
lNumberOfRows = m_AppendRow;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetColumn|
* Gets property ID and default value associated with a column.
*
* @parm LONG | lColumnIndex | Column number
* @parm PROPID | PropID | Property ID
* @parm DWORD | dwType | Property Type (TYPE_VALUE, TYPE_POINTER, TYPE_STRING)
* @parm LPVOID& | lpvDefaultValue | Default value
* @parm DWORD& | cbSize |Length of data (in bytes)
* @parm PRIORITY& | Priority | Column priority
*
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue S_OK | The column was successfully retrieved
*
********************************************************************/
STDMETHODIMP CITResultSet::GetColumn(LONG lColumnIndex, PROPID& PropID, DWORD& dwType, LPVOID& lpvDefaultValue,
DWORD& cbSize, PRIORITY& Priority)
{
// check against invalid lColumnIndex
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
PropID = m_Header[lColumnIndex].dwPropID;
dwType = m_Header[lColumnIndex].dwType;
// it could be NULL
if (m_Header[lColumnIndex].lpvData)
{
lpvDefaultValue = (LPBYTE) m_Header[lColumnIndex].lpvData + sizeof (DWORD);
if (TYPE_VALUE == dwType)
cbSize = sizeof(DWORD);
else
cbSize = *((LPDWORD)m_Header[lColumnIndex].lpvData);
}
else
{
lpvDefaultValue = NULL;
cbSize = 0;
}
Priority = m_Header[lColumnIndex].Priority;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetColumn|
* Gets property ID for a given column index.
*
* @parm LONG | lColumnIndex | Column number
* @parm PROPID | PropID | Property ID
*
* @rvalue E_NOTEXIST | Column does not exist
* @rvalue S_OK | The column was successfully retrieved
*
********************************************************************/
STDMETHODIMP CITResultSet::GetColumn(LONG lColumnIndex, PROPID& PropID)
{
// check against invalid lColumnIndex
if (lColumnIndex >= m_cProp || lColumnIndex < 0)
return SetErrReturn(E_NOTEXIST);
PropID = m_Header[lColumnIndex].dwPropID;
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | GetColumnFromPropID |
* Gets column index for which a property ID is associated
*
* @parm PROPID | PropID | Property ID
* @parm LONG& | lColumnIndex | Column index
*
* @rvalue E_NOTEXIST | The column does not exist
* @rvalue S_OK | The column index was successfully returned
*
********************************************************************/
STDMETHODIMP CITResultSet::GetColumnFromPropID(PROPID PropID, LONG& lColumnIndex)
{
// Loop over all columns, looking for match
for (LONG iIndex = 0; iIndex < m_cProp; iIndex++)
{
if (PropID == m_Header[iIndex].dwPropID)
{
// Found it
lColumnIndex = iIndex;
return S_OK;
}
}
return SetErrReturn(E_NOTEXIST);
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | AppendRows|
* Appends rows from the given source resultset
*
* @parm IITResultSet* | pResSrc | Source resultset
* @parm LONG | lRowSrcFirst | Source row number to start the copy
* @parm LONG | cSrcRows | Number of rows to append
* @parm LONG& | lRowFirstDest | First destination row number appended
*
* @rvalue E_OUTOFMEMORY | Not enough memory to append to the destination
* @rvalue S_OK | All rows were successfully appended
*
********************************************************************/
// UNDONE: this is terribly inefficient. We need to fix resultsets so that rows can
// be copied more easily
STDMETHODIMP CITResultSet::AppendRows(IITResultSet* pResSrc, LONG lRowSrcFirst, LONG cSrcRows,
LONG& lRowDestFirst)
{
LONG lColumn;
HRESULT hr = E_NOTEXIST;
PROPID PropID;
LONG lDestColumn;
CProperty Prop;
lRowDestFirst = m_AppendRow;
m_cs.Lock();
pResSrc->GetColumnCount(lColumn);
// Loop over columns in source result set
for (LONG iProp = 0; iProp < lColumn; iProp++)
{
// get column in dest result set
pResSrc->GetColumn(iProp, PropID);
if (FAILED(GetColumnFromPropID(PropID, lDestColumn)))
continue;
hr = S_OK; // there's at least one matching column
// Loop over rows in input result set
LONG iRow; // loop index
switch( m_Header[lDestColumn].dwType )
{
case TYPE_VALUE:
for (iRow = 0; iRow < cSrcRows; iRow++)
{
pResSrc->Get(lRowSrcFirst + iRow, iProp, Prop);
Set(lRowDestFirst + iRow, lDestColumn, Prop.dwValue);
}
break;
case TYPE_STRING:
for (iRow = 0; iRow < cSrcRows; iRow++)
{
pResSrc->Get(lRowSrcFirst + iRow, iProp, Prop);
Set(lRowDestFirst + iRow, lDestColumn, Prop.lpszwData);
}
break;
case TYPE_POINTER:
for (iRow = 0; iRow < cSrcRows; iRow++)
{
pResSrc->Get(lRowSrcFirst + iRow, iProp, Prop);
Set(lRowDestFirst + iRow, lDestColumn, Prop.lpvData, Prop.cbData);
}
break;
}
}
m_cs.Unlock();
return hr;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Copy |
* Copies the rows associated with the columns set in the
* given result set. This method can be used to take a larger result
* set and reduce it Passing an empty resultset will cause all columns
* and rows to be added and copied from the source resultset.
*
* @parm IITResultSet* | pRSCopy | Result set object containing copied
* rows.
*
* @rvalue E_NOTEXIST | No columns match the input result set
*
* @rvalue S_OK | The rows were successfully copied.
*
********************************************************************/
STDMETHODIMP CITResultSet::Copy(IITResultSet* pRSCopy)
{
LONG lColumn;
HRESULT hr;
pRSCopy->GetColumnCount(lColumn);
if (0L == lColumn)
{
LONG cCols;
LONG iCol;
// add all columns from the source to the dest
GetColumnCount(cCols);
for (iCol = 0; iCol < cCols; iCol++)
{
PROPID pid;
DWORD dwType;
LPVOID lpv;
DWORD cbSize;
PRIORITY pri;
// UNDONE: get rid of this dwType stuff
// UNDONE: simpler place to put column info (i.e. struct/class)
GetColumn(iCol, pid, dwType, lpv, cbSize, pri);
if (dwType == TYPE_VALUE)
{
if (FAILED(hr = pRSCopy->Add(pid, (DWORD)0, PRIORITY_NORMAL)))
return hr;
}
else
{
if (FAILED(hr= pRSCopy->Add(pid, (LPWSTR)0, PRIORITY_NORMAL)))
return hr;
}
}
lColumn = cCols;
}
PROPID PropID;
LONG lInputColumn;
CProperty Prop;
m_cs.Lock();
hr = E_NOTEXIST;
// Loop over columns in output result set
for (LONG iProp = 0; iProp < lColumn; iProp++)
{
// get column in input result set
pRSCopy->GetColumn(iProp, PropID);
if (FAILED(GetColumnFromPropID(PropID, lInputColumn)))
continue;
hr = S_OK; // there's at least one matching column
// Loop over rows in input result set
LONG iRow; // loop index
switch( m_Header[lInputColumn].dwType )
{
case TYPE_VALUE:
for (iRow = 0; iRow < m_AppendRow; iRow++)
{
Get(iRow, lInputColumn, Prop);
pRSCopy->Set(iRow, iProp, Prop.dwValue);
}
break;
case TYPE_STRING:
for (iRow = 0; iRow < m_AppendRow; iRow++)
{
Get(iRow, lInputColumn, Prop);
pRSCopy->Set(iRow, iProp, Prop.lpszwData);
}
break;
case TYPE_POINTER:
for (iRow = 0; iRow < m_AppendRow; iRow++)
{
Get(iRow, lInputColumn, Prop);
pRSCopy->Set(iRow, iProp, Prop.lpvData, Prop.cbData);
}
break;
}
}
m_cs.Unlock();
return hr;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | Clear |
* Frees all memory associated with a result set
*
* @rvalue S_OK | The result set was successfully cleared
*
* @comm This method can be called to clear a result set without
* requiring the set to be destroyed before being used again.
********************************************************************/
STDMETHODIMP CITResultSet::Clear()
{
LONG iProp;
m_cs.Lock();
// Free any column heaps which may have been set.
for (iProp = 0; iProp < m_cProp; iProp++)
{
LPVOID lpvHeap;
PFNCOLHEAPFREE pfnHeapFree;
if ((lpvHeap = m_Header[iProp].lpvHeap) != NULL &&
(pfnHeapFree = m_Header[iProp].pfnHeapFree) != NULL)
{
(*pfnHeapFree)(lpvHeap);
}
}
ClearRows();
m_cProp = 0;
m_cs.Unlock();
return S_OK;
}
/********************************************************************
* @method STDMETHODIMP | IITResultSet | ClearRows |
* Frees all memory associated with a result set, without
* resetting column information
*
* @rvalue S_OK | The result set was successfully cleared
*
* @comm This method can be called to clear a result set without
* requiring the set to be destroyed before being used again.
********************************************************************/
STDMETHODIMP CITResultSet::ClearRows()
{
m_cs.Lock();
// Free page map
if (m_PageMap)
{
delete m_PageMap;
m_PageMap = NULL;
}
// Decommit and release virtual memory
FreeMem();
// Reset memory pool
if (m_pMemPool)
BlockReset(m_pMemPool);
// Reset member data
m_NumberOfPages = 0;
m_fInit = FALSE;
m_RowsReserved = 0;
m_AppendRow = 0;
m_Chunk = -1;
m_cs.Unlock();
return S_OK;
}
// Wrapper for VirtualFree(MEM_DECOMMIT) and VirtualFree(MEM_RELEASE)
HRESULT WINAPI CITResultSet::FreeMem()
{
BOOL fRet;
int iLoop;
// Decommit result set - remember that it's an array of chunks (of rows)
// so we have to loop over all the chunks
// erinfox: bug fix, iLoop <= m_Chunk because m_Chunk is numbered from 0!
for (iLoop = 0; iLoop <= m_Chunk; iLoop++)
{
g_iFreed++;
fRet = VirtualFree(m_ResultSet[iLoop], m_BytesReserved, MEM_DECOMMIT);
ITASSERT(TRUE == fRet);
}
// Then release it
for (iLoop = 0; iLoop <= m_Chunk; iLoop++)
{
fRet = VirtualFree(m_ResultSet[iLoop], 0, MEM_RELEASE);
ITASSERT(TRUE == fRet);
}
return S_OK;
}
STDMETHODIMP CITResultSet::Free()
{
return E_NOTIMPL;
}
//////////////////////// Asynchronous //////////////////////////////////////
STDMETHODIMP CITResultSet::IsCompleted()
{
return E_NOTIMPL;
}
STDMETHODIMP CITResultSet::Cancel()
{
return E_NOTIMPL;
}
STDMETHODIMP CITResultSet::Pause(BOOL fPause)
{
return E_NOTIMPL;
}
STDMETHODIMP CITResultSet::GetRowStatus(LONG lRowFirst, LONG cRows, LPROWSTATUS lpRowStatus)
{
return E_NOTIMPL;
}
STDMETHODIMP CITResultSet::GetColumnStatus(LPCOLUMNSTATUS lpColStatus)
{
return E_NOTIMPL;
}