зеркало из https://github.com/mozilla/gecko-dev.git
377 строки
12 KiB
C++
377 строки
12 KiB
C++
/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
|
|
*
|
|
* The contents of this file are subject to the Netscape Public License
|
|
* Version 1.0 (the "NPL"); you may not use this file except in
|
|
* compliance with the NPL. You may obtain a copy of the NPL at
|
|
* http://www.mozilla.org/NPL/
|
|
*
|
|
* Software distributed under the NPL is distributed on an "AS IS" basis,
|
|
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the NPL
|
|
* for the specific language governing rights and limitations under the
|
|
* NPL.
|
|
*
|
|
* The Initial Developer of this code under the NPL is Netscape
|
|
* Communications Corporation. Portions created by Netscape are
|
|
* Copyright (C) 1999 Netscape Communications Corporation. All Rights
|
|
* Reserved.
|
|
*/
|
|
|
|
#include "msgCore.h"
|
|
#include "prlog.h"
|
|
#include "nsMsgLineBuffer.h"
|
|
|
|
#include "nsIInputStream.h" // used by nsMsgLineStreamBuffer
|
|
|
|
nsByteArray::nsByteArray()
|
|
{
|
|
m_buffer = NULL;
|
|
m_bufferSize = 0;
|
|
m_bufferPos = 0;
|
|
}
|
|
|
|
nsByteArray::~nsByteArray()
|
|
{
|
|
PR_FREEIF(m_buffer);
|
|
}
|
|
|
|
nsresult nsByteArray::GrowBuffer(PRUint32 desired_size, PRUint32 quantum)
|
|
{
|
|
if (m_bufferSize < desired_size)
|
|
{
|
|
char *new_buf;
|
|
PRUint32 increment = desired_size - m_bufferSize;
|
|
if (increment < quantum) /* always grow by a minimum of N bytes */
|
|
increment = quantum;
|
|
|
|
|
|
new_buf = (m_buffer
|
|
? (char *) PR_REALLOC (m_buffer, (m_bufferSize + increment))
|
|
: (char *) PR_MALLOC (m_bufferSize + increment));
|
|
if (! new_buf)
|
|
return NS_ERROR_OUT_OF_MEMORY;
|
|
m_buffer = new_buf;
|
|
m_bufferSize += increment;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
nsresult nsByteArray::AppendString(const char *string)
|
|
{
|
|
PRUint32 strLength = (string) ? PL_strlen(string) : 0;
|
|
return AppendBuffer(string, strLength);
|
|
|
|
}
|
|
|
|
nsresult nsByteArray::AppendBuffer(const char *buffer, PRUint32 length)
|
|
{
|
|
nsresult ret = NS_OK;
|
|
if (m_bufferPos + length > m_bufferSize)
|
|
ret = GrowBuffer(m_bufferPos + length, 1024);
|
|
if (ret == NS_OK)
|
|
{
|
|
memcpy(m_buffer + m_bufferPos, buffer, length);
|
|
m_bufferPos += length;
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
nsMsgLineBuffer::nsMsgLineBuffer(nsMsgLineBufferHandler *handler, PRBool convertNewlinesP)
|
|
{
|
|
m_handler = handler;
|
|
m_convertNewlinesP = convertNewlinesP;
|
|
m_lookingForCRLF = PR_TRUE;
|
|
}
|
|
|
|
nsMsgLineBuffer::~nsMsgLineBuffer()
|
|
{
|
|
}
|
|
|
|
void
|
|
nsMsgLineBuffer::SetLookingForCRLF(PRBool b)
|
|
{
|
|
m_lookingForCRLF = b;
|
|
}
|
|
|
|
PRInt32 nsMsgLineBuffer::BufferInput(const char *net_buffer, PRInt32 net_buffer_size)
|
|
{
|
|
int status = 0;
|
|
if (m_bufferPos > 0 && m_buffer && m_buffer[m_bufferPos - 1] == CR &&
|
|
net_buffer_size > 0 && net_buffer[0] != LF) {
|
|
/* The last buffer ended with a CR. The new buffer does not start
|
|
with a LF. This old buffer should be shipped out and discarded. */
|
|
PR_ASSERT(m_bufferSize > m_bufferPos);
|
|
if (m_bufferSize <= m_bufferPos) return -1;
|
|
status = ConvertAndSendBuffer();
|
|
if (status < 0)
|
|
return status;
|
|
m_bufferPos = 0;
|
|
}
|
|
while (net_buffer_size > 0)
|
|
{
|
|
const char *net_buffer_end = net_buffer + net_buffer_size;
|
|
const char *newline = 0;
|
|
const char *s;
|
|
|
|
|
|
for (s = net_buffer; s < net_buffer_end; s++)
|
|
{
|
|
if (m_lookingForCRLF) {
|
|
/* Move forward in the buffer until the first newline.
|
|
Stop when we see CRLF, CR, or LF, or the end of the buffer.
|
|
*But*, if we see a lone CR at the *very end* of the buffer,
|
|
treat this as if we had reached the end of the buffer without
|
|
seeing a line terminator. This is to catch the case of the
|
|
buffers splitting a CRLF pair, as in "FOO\r\nBAR\r" "\nBAZ\r\n".
|
|
*/
|
|
if (*s == CR || *s == LF) {
|
|
newline = s;
|
|
if (newline[0] == CR) {
|
|
if (s == net_buffer_end - 1) {
|
|
/* CR at end - wait for the next character. */
|
|
newline = 0;
|
|
break;
|
|
}
|
|
else if (newline[1] == LF) {
|
|
/* CRLF seen; swallow both. */
|
|
newline++;
|
|
}
|
|
}
|
|
newline++;
|
|
break;
|
|
}
|
|
}
|
|
else {
|
|
/* if not looking for a CRLF, stop at CR or LF. (for example, when parsing the newsrc file). this fixes #9896, where we'd lose the last line of anything we'd parse that used CR as the line break. */
|
|
if (*s == CR || *s == LF) {
|
|
newline = s;
|
|
newline++;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* Ensure room in the net_buffer and append some or all of the current
|
|
chunk of data to it. */
|
|
{
|
|
const char *end = (newline ? newline : net_buffer_end);
|
|
PRUint32 desired_size = (end - net_buffer) + m_bufferPos + 1;
|
|
|
|
if (desired_size >= m_bufferSize)
|
|
{
|
|
status = GrowBuffer (desired_size, 1024);
|
|
if (status < 0)
|
|
return status;
|
|
}
|
|
memcpy (m_buffer + m_bufferPos, net_buffer, (end - net_buffer));
|
|
m_bufferPos += (end - net_buffer);
|
|
}
|
|
|
|
/* Now m_buffer contains either a complete line, or as complete
|
|
a line as we have read so far.
|
|
|
|
If we have a line, process it, and then remove it from `m_buffer'.
|
|
Then go around the loop again, until we drain the incoming data.
|
|
*/
|
|
if (!newline)
|
|
return 0;
|
|
|
|
status = ConvertAndSendBuffer();
|
|
if (status < 0) return status;
|
|
|
|
net_buffer_size -= (newline - net_buffer);
|
|
net_buffer = newline;
|
|
m_bufferPos = 0;
|
|
}
|
|
#ifdef DEBUG_bienvenu
|
|
printf("returning from buffer input m_bufferPos = %ld\n", m_bufferPos);
|
|
#endif
|
|
return 0;
|
|
}
|
|
|
|
PRInt32 nsMsgLineBuffer::HandleLine(char *line, PRUint32 line_length)
|
|
{
|
|
NS_ASSERTION(PR_FALSE, "must override this method if you don't provide a handler");
|
|
return 0;
|
|
}
|
|
|
|
PRInt32 nsMsgLineBuffer::ConvertAndSendBuffer()
|
|
{
|
|
/* Convert the line terminator to the native form.
|
|
*/
|
|
|
|
char *buf = m_buffer;
|
|
PRInt32 length = m_bufferPos;
|
|
|
|
char* newline;
|
|
|
|
PR_ASSERT(buf && length > 0);
|
|
if (!buf || length <= 0)
|
|
return -1;
|
|
newline = buf + length;
|
|
|
|
PR_ASSERT(newline[-1] == CR || newline[-1] == LF);
|
|
if (newline[-1] != CR && newline[-1] != LF)
|
|
return -1;
|
|
|
|
if (!m_convertNewlinesP)
|
|
{
|
|
}
|
|
#if (MSG_LINEBREAK_LEN == 1)
|
|
else if ((newline - buf) >= 2 &&
|
|
newline[-2] == CR &&
|
|
newline[-1] == LF)
|
|
{
|
|
/* CRLF -> CR or LF */
|
|
buf [length - 2] = MSG_LINEBREAK[0];
|
|
length--;
|
|
}
|
|
else if (newline > buf + 1 &&
|
|
newline[-1] != MSG_LINEBREAK[0])
|
|
{
|
|
/* CR -> LF or LF -> CR */
|
|
buf [length - 1] = MSG_LINEBREAK[0];
|
|
}
|
|
#else
|
|
else if (((newline - buf) >= 2 && newline[-2] != CR) ||
|
|
((newline - buf) >= 1 && newline[-1] != LF))
|
|
{
|
|
/* LF -> CRLF or CR -> CRLF */
|
|
length++;
|
|
buf[length - 2] = MSG_LINEBREAK[0];
|
|
buf[length - 1] = MSG_LINEBREAK[1];
|
|
}
|
|
#endif
|
|
|
|
return (m_handler) ? m_handler->HandleLine(buf, length) : HandleLine(buf, length);
|
|
}
|
|
|
|
// If there's still some data (non CRLF terminated) flush it out
|
|
PRInt32 nsMsgLineBuffer::FlushLastLine()
|
|
{
|
|
char *buf = m_buffer + m_bufferPos;
|
|
PRInt32 length = m_bufferPos - 1;
|
|
if (length > 0)
|
|
return (m_handler) ? m_handler->HandleLine(buf, length) : HandleLine(buf, length);
|
|
else
|
|
return 0;
|
|
}
|
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
// This is a utility class used to efficiently extract lines from an input stream by buffering
|
|
// read but unprocessed stream data in a buffer.
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
nsMsgLineStreamBuffer::nsMsgLineStreamBuffer(PRUint32 aBufferSize, const char * aEndOfLineToken,
|
|
PRBool aAllocateNewLines, PRBool aEatCRLFs)
|
|
: m_eatCRLFs(aEatCRLFs), m_allocateNewLines(aAllocateNewLines),
|
|
m_endOfLineToken(aEndOfLineToken)
|
|
{
|
|
NS_PRECONDITION(aBufferSize > 0, "invalid buffer size!!!");
|
|
m_dataBuffer = nsnull;
|
|
m_startPos = nsnull;
|
|
|
|
// used to buffer incoming data by ReadNextLineFromInput
|
|
if (aBufferSize > 0)
|
|
{
|
|
m_dataBuffer = (char *) PR_CALLOC(sizeof(char) * aBufferSize);
|
|
m_startPos = m_dataBuffer;
|
|
}
|
|
|
|
m_dataBufferSize = aBufferSize;
|
|
}
|
|
|
|
nsMsgLineStreamBuffer::~nsMsgLineStreamBuffer()
|
|
{
|
|
PR_FREEIF(m_dataBuffer); // release our buffer...
|
|
}
|
|
|
|
// the design for this method has an inherit bug: if the length of the line is greater than the size of m_dataBufferSize,
|
|
// then we'll never find the next line because we can't hold the whole line in memory.
|
|
// aInputStream - the input stream we want to read a line from
|
|
// aPauseForMoreData is returned as PR_TRUE if the stream does not yet contain a line and we must wait for more
|
|
// data to come into the stream.
|
|
|
|
// Note to people wishing to modify this function: Be *VERY CAREFUL* this is a critical function used by all of
|
|
// our mail protocols including imap, nntp, and pop. If you screw it up, you could break a lot of stuff.....
|
|
|
|
char * nsMsgLineStreamBuffer::ReadNextLine(nsIInputStream * aInputStream, PRUint32 &aNumBytesInLine, PRBool &aPauseForMoreData)
|
|
{
|
|
// try to extract a line from m_inputBuffer. If we don't have an entire line,
|
|
// then read more bytes out from the stream. If the stream is empty then wait
|
|
// on the monitor for more data to come in.
|
|
|
|
NS_PRECONDITION(m_startPos && m_dataBufferSize > 0, "invalid input arguments for read next line from input");
|
|
|
|
// initialize out values
|
|
aPauseForMoreData = PR_FALSE;
|
|
aNumBytesInLine = 0;
|
|
char * endOfLine = nsnull;
|
|
|
|
PRUint32 numBytesInBuffer = PL_strlen(m_startPos);
|
|
|
|
if (numBytesInBuffer > 0) // any data in our internal buffer?
|
|
endOfLine = PL_strstr(m_startPos, m_endOfLineToken); // see if we already have a line ending...
|
|
|
|
// it's possible that we got here before the first time we receive data from the server
|
|
// so aInputStream will be nsnull...
|
|
if (!endOfLine && aInputStream) // get some more data from the server
|
|
{
|
|
PRUint32 numBytesInStream = 0;
|
|
PRUint32 numBytesCopied = 0;
|
|
aInputStream->Available(&numBytesInStream);
|
|
// if the number of bytes we want to read from the stream, is greater than the number
|
|
// of bytes left in our buffer, then we need to shift the start pos and its contents
|
|
// down to the beginning of m_dataBuffer...
|
|
PRUint32 numFreeBytesInBuffer = (m_dataBuffer + m_dataBufferSize) - (m_startPos + numBytesInBuffer);
|
|
if (numBytesInStream > numFreeBytesInBuffer)
|
|
{
|
|
nsCRT::memcpy(m_dataBuffer, m_startPos, numBytesInBuffer);
|
|
m_dataBuffer[numBytesInBuffer] = '\0'; // make sure the end of the buffer is terminated
|
|
m_startPos = m_dataBuffer; // update the new start position
|
|
// update the number of free bytes in the buffer
|
|
numFreeBytesInBuffer = m_dataBufferSize - numBytesInBuffer;
|
|
}
|
|
|
|
PRUint32 numBytesToCopy = PR_MIN(numFreeBytesInBuffer - 1 /* leave one for a null terminator */, numBytesInStream);
|
|
// read the data into the end of our data buffer
|
|
if (numBytesToCopy > 0)
|
|
{
|
|
aInputStream->Read(m_startPos + numBytesInBuffer, numBytesToCopy, &numBytesCopied);
|
|
m_startPos[numBytesInBuffer + numBytesCopied] = '\0';
|
|
}
|
|
|
|
// okay, now that we've tried to read in more data from the stream, look for another end of line
|
|
// character
|
|
endOfLine = PL_strstr(m_startPos, m_endOfLineToken);
|
|
|
|
}
|
|
|
|
// okay, now check again for endOfLine.
|
|
if (endOfLine)
|
|
{
|
|
if (!m_eatCRLFs)
|
|
endOfLine += PL_strlen(m_endOfLineToken); // count for CRLF
|
|
|
|
// PR_CALLOC zeros out the allocated line
|
|
char* newLine = (char*) PR_CALLOC(endOfLine-m_startPos+1);
|
|
if (!newLine)
|
|
return nsnull;
|
|
|
|
nsCRT::memcpy(newLine, m_startPos, endOfLine-m_startPos); // copy the string into the new line buffer
|
|
aNumBytesInLine = endOfLine - m_startPos;
|
|
|
|
if (m_eatCRLFs)
|
|
endOfLine += PL_strlen(m_endOfLineToken); // advance past CRLF if we haven't already done so...
|
|
|
|
// now we need to update the data buffer to go past the line we just read out.
|
|
if (PL_strlen(endOfLine) <= 0) // if no more data in the buffer, then just zero out the buffer...
|
|
m_startPos[0] = '\0';
|
|
else // advance
|
|
m_startPos = endOfLine; // move us up to the end of the line
|
|
return newLine;
|
|
}
|
|
|
|
aPauseForMoreData = PR_TRUE;
|
|
return nsnull; // if we somehow got here. we don't have another line in the buffer yet...need to wait for more data...
|
|
}
|