mirror of
https://bitbucket.org/smil3y/kdelibs.git
synced 2025-02-24 10:52:49 +00:00

read once, match until match is found or otherwise. that means less disk I/O Signed-off-by: Ivailo Monev <xakepa10@gmail.com>
121 lines
4.7 KiB
C++
121 lines
4.7 KiB
C++
/* This file is part of the KDE libraries
|
|
* Copyright 2007 David Faure <faure@kde.org>
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Library General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2 of the License, or (at your option) any later version.
|
|
*
|
|
* This library is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* Library General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Library General Public License
|
|
* along with this library; see the file COPYING.LIB. If not, write to
|
|
* the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
|
|
* Boston, MA 02110-1301, USA.
|
|
*/
|
|
|
|
#include "kmimemagicrule_p.h"
|
|
#include <QIODevice>
|
|
#include <kdebug.h>
|
|
|
|
/*
|
|
* Historical note:
|
|
* The notion of indents is used differently from the old file(1) magic file.
|
|
* It is not enough that a top-level rule matches for the search to be over;
|
|
* in file(1) subrules were used as refinement (and in KMimeMagic they were
|
|
* mandatory if the toplevel rule didn't have a mimetype associated with it).
|
|
* Here they are mandatory.
|
|
* We need at least one continuation at every level to match, and then the match is valid:
|
|
[50:application/x-kformula]
|
|
>0=^B^_<8B>
|
|
1>10=^GKOffice
|
|
2>18=^Xapplication/x-kformula^D^F
|
|
>0=^DPK^C^D
|
|
1>30=^Hmimetype
|
|
2>38=^Vapplication/x-kformula
|
|
* Either it's an old (tar) file and the first hierarchy (0,1,2 levels) matches,
|
|
* or it's a newer file (zip) file and the second hierarchy (0,1,2 levels) has to match.
|
|
*
|
|
*/
|
|
|
|
static bool testMatches(const QByteArray &availableData, const QList<KMimeMagicMatch>& matches, const QString& mimeType)
|
|
{
|
|
for ( QList<KMimeMagicMatch>::const_iterator it = matches.begin(), end = matches.end() ;
|
|
it != end ; ++it ) {
|
|
const KMimeMagicMatch& match = *it;
|
|
if (match.match(availableData, mimeType)) {
|
|
// One of the hierarchies matched -> mimetype recognized.
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
bool KMimeMagicRule::match(const QByteArray &availableData) const
|
|
{
|
|
return testMatches(availableData, m_matches, m_mimetype);
|
|
}
|
|
|
|
bool KMimeMagicMatch::match(const QByteArray &availableData, const QString& mimeType) const
|
|
{
|
|
// First, check that "this" matches, then we'll dive into subMatches if any.
|
|
|
|
const qint64 mDataSize = m_data.size();
|
|
if (m_rangeStart + mDataSize > availableData.size())
|
|
return false; // file is too small
|
|
|
|
// Read in one block all the data we'll need
|
|
// Example: m_data="ABC", m_rangeLength=3 -> we need 3+3-1=5 bytes (ABCxx,xABCx,xxABC would match)
|
|
const int dataNeeded = qMin(mDataSize + m_rangeLength - 1, availableData.size() - m_rangeStart);
|
|
QByteArray readData = QByteArray::fromRawData(
|
|
availableData.constData() + m_rangeStart,
|
|
dataNeeded
|
|
);
|
|
// Warning, readData isn't null-terminated so this kDebug
|
|
// gives valgrind warnings (when printing as char* data).
|
|
//kDebug() << "readData (from availableData) at pos " << m_rangeStart << ":" << readData;
|
|
|
|
// All we need to do now, is to look for m_data in readData (whose size is dataNeeded).
|
|
// Either as a simple indexOf search, or applying the mask.
|
|
|
|
bool found = false;
|
|
if (m_mask.isEmpty()) {
|
|
//kDebug() << "m_data=" << m_data;
|
|
found = readData.indexOf(m_data) != -1;
|
|
//if (found)
|
|
// kDebug() << "Matched readData=" << readData << "with m_data=" << m_data << "so this is" << mimeType;
|
|
} else {
|
|
const char* mask = m_mask.constData();
|
|
const char* refData = m_data.constData();
|
|
const char* readDataBase = readData.constData();
|
|
// Example (continued from above):
|
|
// deviceSize is 4, so dataNeeded was max'ed to 4.
|
|
// maxStartPos = 4 - 3 + 1 = 2, and indeed
|
|
// we need to check for a match a positions 0 and 1 (ABCx and xABC).
|
|
const qint64 maxStartPos = dataNeeded - mDataSize + 1;
|
|
for (int i = 0; i < maxStartPos; ++i) {
|
|
const char* d = readDataBase + i;
|
|
bool valid = true;
|
|
for (int off = 0; off < mDataSize; ++off ) {
|
|
if ( ((*d++) & mask[off]) != ((refData[off] & mask[off])) ) {
|
|
valid = false;
|
|
break;
|
|
}
|
|
}
|
|
if (valid)
|
|
found = true;
|
|
}
|
|
}
|
|
if (!found)
|
|
return false;
|
|
|
|
// No submatch? Then we are done.
|
|
if (m_subMatches.isEmpty())
|
|
return true;
|
|
|
|
// Check that one of the submatches matches too
|
|
return testMatches(availableData, m_subMatches, mimeType);
|
|
}
|