X-Git-Url: https://git.saurik.com/wxWidgets.git/blobdiff_plain/16a12a3d27d3d3e2557bcde15365245ef2675bb4..a7c26d107d0c6f191e55d30613f4816e9c2f9386:/src/html/helpdata.cpp diff --git a/src/html/helpdata.cpp b/src/html/helpdata.cpp index 150984d981..c0418b8d39 100644 --- a/src/html/helpdata.cpp +++ b/src/html/helpdata.cpp @@ -9,8 +9,8 @@ // Licence: wxWindows licence ///////////////////////////////////////////////////////////////////////////// -#ifdef __GNUG__ -#pragma implementation +#if defined(__GNUG__) && !defined(NO_GCC_PRAGMA) +#pragma implementation "helpdata.h" #endif // For compilers that support precompilation, includes "wx.h". @@ -22,18 +22,26 @@ #include "wx/defs.h" -#if wxUSE_HTML +#if wxUSE_HTML && wxUSE_STREAMS #ifndef WXPRECOMP -#include "wx/wx.h" + #include "wx/intl.h" + #include "wx/log.h" #endif +#include + #include "wx/html/helpdata.h" #include "wx/tokenzr.h" #include "wx/wfstream.h" #include "wx/busyinfo.h" +#include "wx/encconv.h" +#include "wx/fontmap.h" +#include "wx/log.h" #include "wx/html/htmlpars.h" #include "wx/html/htmldefs.h" +#include "wx/html/htmlfilt.h" +#include "wx/filename.h" #include "wx/arrimpl.cpp" WX_DEFINE_OBJARRAY(wxHtmlBookRecArray) @@ -43,34 +51,30 @@ WX_DEFINE_OBJARRAY(wxHtmlBookRecArray) //----------------------------------------------------------------------------- // Reads one line, stores it into buf and returns pointer to new line or NULL. -static char* ReadLine(char *line, char *buf) +static const wxChar* ReadLine(const wxChar *line, wxChar *buf, size_t bufsize) { - char *writeptr = buf, *readptr = line; + wxChar *writeptr = buf; + wxChar *endptr = buf + bufsize - 1; + const wxChar *readptr = line; - while (*readptr != 0 && *readptr != '\r' && *readptr != '\n') *(writeptr++) = *(readptr++); + while (*readptr != 0 && *readptr != _T('\r') && *readptr != _T('\n') && + writeptr != endptr) + *(writeptr++) = *(readptr++); *writeptr = 0; - while (*readptr == '\r' || *readptr == '\n') readptr++; - if (*readptr == 0) return NULL; - else return readptr; + while (*readptr == _T('\r') || *readptr == _T('\n')) + readptr++; + if (*readptr == 0) + return NULL; + else + return readptr; } -static wxString SafeFileName(const wxString& s) -{ - wxString res = s; - res.Replace(":", "_", TRUE); - res.Replace(" ", "_", TRUE); - res.Replace("/", "_", TRUE); - res.Replace("\\", "_", TRUE); - res.Replace("#", "_", TRUE); - res.Replace(".", "_", TRUE); - return res; -} - -static int LINKAGEMODE IndexCompareFunc(const void *a, const void *b) +extern "C" int LINKAGEMODE +wxHtmlHelpIndexCompareFunc(const void *a, const void *b) { - return strcmp(((wxHtmlContentsItem*)a) -> m_Name, ((wxHtmlContentsItem*)b) -> m_Name); + return wxStricmp(((wxHtmlContentsItem*)a)->m_Name, ((wxHtmlContentsItem*)b)->m_Name); } @@ -80,9 +84,15 @@ static int LINKAGEMODE IndexCompareFunc(const void *a, const void *b) class HP_Parser : public wxHtmlParser { - public: - void AddText(const char* WXUNUSED(text)) { } - wxObject* GetProduct() { return NULL; } +public: + HP_Parser() { } + + wxObject* GetProduct() { return NULL; } + +protected: + virtual void AddText(const wxChar* WXUNUSED(txt)) {} + + DECLARE_NO_COPY_CLASS(HP_Parser) }; @@ -95,52 +105,85 @@ class HP_TagHandler : public wxHtmlTagHandler private: wxString m_Name, m_Page; int m_Level; - int m_ID; + int m_ID; int m_Index; wxHtmlContentsItem *m_Items; int m_ItemsCnt; wxHtmlBookRecord *m_Book; public: - HP_TagHandler(wxHtmlBookRecord *b) : wxHtmlTagHandler() {m_Book = b; m_Items = NULL; m_ItemsCnt = 0; m_Name = m_Page = wxEmptyString; m_Level = 0;} - wxString GetSupportedTags() {return "UL,OBJECT,PARAM";} + HP_TagHandler(wxHtmlBookRecord *b) : wxHtmlTagHandler() + { m_Book = b; m_Items = NULL; m_ItemsCnt = 0; m_Name = m_Page = wxEmptyString; + m_Level = 0; m_ID = -1; } + wxString GetSupportedTags() { return wxT("UL,OBJECT,PARAM"); } bool HandleTag(const wxHtmlTag& tag); void WriteOut(wxHtmlContentsItem*& array, int& size); void ReadIn(wxHtmlContentsItem* array, int size); + + DECLARE_NO_COPY_CLASS(HP_TagHandler) }; bool HP_TagHandler::HandleTag(const wxHtmlTag& tag) { - if (tag.GetName() == "UL") { + if (tag.GetName() == wxT("UL")) + { m_Level++; ParseInner(tag); m_Level--; return TRUE; } - - else if (tag.GetName() == "OBJECT") { + else if (tag.GetName() == wxT("OBJECT")) + { m_Name = m_Page = wxEmptyString; ParseInner(tag); - if (m_Page != wxEmptyString) { + +#if 0 + if (!m_Page.IsEmpty()) + /* Valid HHW's file may contain only two object tags: + + + + + + or + + + + + + + We're interested in the latter. !m_Page.IsEmpty() is valid + condition because text/site properties does not contain Local param + */ +#endif + if (tag.GetParam(wxT("TYPE")) == wxT("text/sitemap")) + { if (m_ItemsCnt % wxHTML_REALLOC_STEP == 0) - m_Items = (wxHtmlContentsItem*) realloc(m_Items, (m_ItemsCnt + wxHTML_REALLOC_STEP) * sizeof(wxHtmlContentsItem)); + m_Items = (wxHtmlContentsItem*) realloc(m_Items, + (m_ItemsCnt + wxHTML_REALLOC_STEP) * + sizeof(wxHtmlContentsItem)); + m_Items[m_ItemsCnt].m_Level = m_Level; m_Items[m_ItemsCnt].m_ID = m_ID; - m_Items[m_ItemsCnt].m_Page = new char[m_Page.Length() + 1]; - strcpy(m_Items[m_ItemsCnt].m_Page, m_Page.c_str()); - m_Items[m_ItemsCnt].m_Name = new char [m_Name.Length() + 1]; - strcpy(m_Items[m_ItemsCnt].m_Name, m_Name.c_str()); + m_Items[m_ItemsCnt].m_Page = new wxChar[m_Page.Length() + 1]; + wxStrcpy(m_Items[m_ItemsCnt].m_Page, m_Page.c_str()); + m_Items[m_ItemsCnt].m_Name = new wxChar [m_Name.Length() + 1]; + wxStrcpy(m_Items[m_ItemsCnt].m_Name, m_Name.c_str()); m_Items[m_ItemsCnt].m_Book = m_Book; m_ItemsCnt++; } + return TRUE; } - - else { // "PARAM" - if (m_Name == wxEmptyString && tag.GetParam("NAME") == "Name") m_Name = tag.GetParam("VALUE"); - if (tag.GetParam("NAME") == "Local") m_Page = tag.GetParam("VALUE"); - if (tag.GetParam("NAME") == "ID") tag.ScanParam("VALUE", "%i", &m_ID); + else + { // "PARAM" + if (m_Name == wxEmptyString && tag.GetParam(wxT("NAME")) == wxT("Name")) + m_Name = tag.GetParam(wxT("VALUE")); + if (tag.GetParam(wxT("NAME")) == wxT("Local")) + m_Page = tag.GetParam(wxT("VALUE")); + if (tag.GetParam(wxT("NAME")) == wxT("ID")) + tag.GetParamAsInt(wxT("VALUE"), &m_ID); return FALSE; } } @@ -161,10 +204,23 @@ void HP_TagHandler::ReadIn(wxHtmlContentsItem* array, int size) m_ItemsCnt = size; } + + + //----------------------------------------------------------------------------- // wxHtmlHelpData //----------------------------------------------------------------------------- +wxString wxHtmlBookRecord::GetFullPath(const wxString &page) const +{ + if (wxIsAbsolutePath(page)) + return page; + else + return m_BasePath + page; +} + + + IMPLEMENT_DYNAMIC_CLASS(wxHtmlHelpData, wxObject) wxHtmlHelpData::wxHtmlHelpData() @@ -182,15 +238,19 @@ wxHtmlHelpData::~wxHtmlHelpData() int i; m_BookRecords.Empty(); - if (m_Contents) { - for (i = 0; i < m_ContentsCnt; i++) { + if (m_Contents) + { + for (i = 0; i < m_ContentsCnt; i++) + { delete[] m_Contents[i].m_Page; delete[] m_Contents[i].m_Name; } free(m_Contents); } - if (m_Index) { - for (i = 0; i < m_IndexCnt; i++) { + if (m_Index) + { + for (i = 0; i < m_IndexCnt; i++) + { delete[] m_Index[i].m_Page; delete[] m_Index[i].m_Name; } @@ -201,81 +261,140 @@ wxHtmlHelpData::~wxHtmlHelpData() bool wxHtmlHelpData::LoadMSProject(wxHtmlBookRecord *book, wxFileSystem& fsys, const wxString& indexfile, const wxString& contentsfile) { wxFSFile *f; - char *buf; - int sz; + wxHtmlFilterHTML filter; + wxString buf; wxString string; HP_Parser parser; HP_TagHandler *handler = new HP_TagHandler(book); parser.AddTagHandler(handler); - f = ( contentsfile.IsEmpty() ? NULL : fsys.OpenFile(contentsfile) ); - if (f) { - sz = f -> GetStream() -> GetSize(); - buf = new char[sz+1]; - buf[sz] = 0; - f -> GetStream() -> Read(buf, sz); + f = ( contentsfile.IsEmpty() ? (wxFSFile*) NULL : fsys.OpenFile(contentsfile) ); + if (f) + { + buf.clear(); + buf = filter.ReadFile(*f); delete f; - handler -> ReadIn(m_Contents, m_ContentsCnt); + handler->ReadIn(m_Contents, m_ContentsCnt); parser.Parse(buf); - handler -> WriteOut(m_Contents, m_ContentsCnt); - delete[] buf; + handler->WriteOut(m_Contents, m_ContentsCnt); + } + else + { + wxLogError(_("Cannot open contents file: %s"), contentsfile.c_str()); } - f = ( indexfile.IsEmpty() ? NULL : fsys.OpenFile(indexfile) ); - if (f) { - sz = f -> GetStream() -> GetSize(); - buf = new char[sz+1]; - buf[sz] = 0; - f -> GetStream() -> Read(buf, sz); + f = ( indexfile.IsEmpty() ? (wxFSFile*) NULL : fsys.OpenFile(indexfile) ); + if (f) + { + buf.clear(); + buf = filter.ReadFile(*f); delete f; - handler -> ReadIn(m_Index, m_IndexCnt); + handler->ReadIn(m_Index, m_IndexCnt); parser.Parse(buf); - handler -> WriteOut(m_Index, m_IndexCnt); - delete[] buf; + handler->WriteOut(m_Index, m_IndexCnt); + } + else if (!indexfile.IsEmpty()) + { + wxLogError(_("Cannot open index file: %s"), indexfile.c_str()); } return TRUE; } + +inline static void CacheWriteInt32(wxOutputStream *f, wxInt32 value) +{ + wxInt32 x = wxINT32_SWAP_ON_BE(value); + f->Write(&x, sizeof(x)); +} + +inline static wxInt32 CacheReadInt32(wxInputStream *f) +{ + wxInt32 x; + f->Read(&x, sizeof(x)); + return wxINT32_SWAP_ON_BE(x); +} + +inline static void CacheWriteString(wxOutputStream *f, const wxChar *str) +{ +#if wxUSE_UNICODE + wxWX2MBbuf mbstr(wxConvUTF8.cWX2MB(str)); +#else + const wxChar *mbstr = str; +#endif + size_t len = strlen(mbstr)+1; + CacheWriteInt32(f, len); + f->Write(mbstr, len); +} + +inline static wxChar* CacheReadString(wxInputStream *f) +{ + char *str; + size_t len = (size_t)CacheReadInt32(f); + str = new char[len]; + f->Read(str, len); +#if !wxUSE_UNICODE + return str; +#else + wxMB2WXbuf wxstr(wxConvUTF8.cMB2WX(str)); + wxChar *outstr = new wxChar[wxStrlen(wxstr)+1]; + wxStrcpy(outstr, wxstr); + return outstr; +#endif +} + +#define CURRENT_CACHED_BOOK_VERSION 4 + +// Additional flags to detect incompatibilities of the runtime environment: +#define CACHED_BOOK_FORMAT_FLAGS \ + (wxUSE_UNICODE << 0) + + bool wxHtmlHelpData::LoadCachedBook(wxHtmlBookRecord *book, wxInputStream *f) { int i, st; - int x; + wxInt32 version; + + /* load header - version info : */ + version = CacheReadInt32(f); + + if (version != CURRENT_CACHED_BOOK_VERSION) + { + // NB: We can just silently return FALSE here and don't worry about + // it anymore, because AddBookParam will load the MS project in + // absence of (properly versioned) .cached file and automatically + // create new .cached file immediately afterward. + return FALSE; + } - /* load contents : */ + if (CacheReadInt32(f) != CACHED_BOOK_FORMAT_FLAGS) + return FALSE; - f -> Read(&x, sizeof(x)); + /* load contents : */ st = m_ContentsCnt; - m_ContentsCnt += x; - m_Contents = (wxHtmlContentsItem*) realloc(m_Contents, (m_ContentsCnt / wxHTML_REALLOC_STEP + 1) * wxHTML_REALLOC_STEP * sizeof(wxHtmlContentsItem)); - for (i = st; i < m_ContentsCnt; i++) { - f -> Read(&x, sizeof(x)); - m_Contents[i].m_Level = x; - f -> Read(&x, sizeof(x)); - m_Contents[i].m_ID = x; - f -> Read(&x, sizeof(x)); - m_Contents[i].m_Name = new char[x]; - f -> Read(m_Contents[i].m_Name, x); - f -> Read(&x, sizeof(x)); - m_Contents[i].m_Page = new char[x]; - f -> Read(m_Contents[i].m_Page, x); + m_ContentsCnt += CacheReadInt32(f); + m_Contents = (wxHtmlContentsItem*) realloc(m_Contents, + (m_ContentsCnt / wxHTML_REALLOC_STEP + 1) * + wxHTML_REALLOC_STEP * sizeof(wxHtmlContentsItem)); + for (i = st; i < m_ContentsCnt; i++) + { + m_Contents[i].m_Level = CacheReadInt32(f); + m_Contents[i].m_ID = CacheReadInt32(f); + m_Contents[i].m_Name = CacheReadString(f); + m_Contents[i].m_Page = CacheReadString(f); m_Contents[i].m_Book = book; } /* load index : */ - - f -> Read(&x, sizeof(x)); st = m_IndexCnt; - m_IndexCnt += x; - m_Index = (wxHtmlContentsItem*) realloc(m_Index, (m_IndexCnt / wxHTML_REALLOC_STEP + 1) * wxHTML_REALLOC_STEP * sizeof(wxHtmlContentsItem)); - for (i = st; i < m_IndexCnt; i++) { - f -> Read(&x, sizeof(x)); - m_Index[i].m_Name = new char[x]; - f -> Read(m_Index[i].m_Name, x); - f -> Read(&x, sizeof(x)); - m_Index[i].m_Page = new char[x]; - f -> Read(m_Index[i].m_Page, x); + m_IndexCnt += CacheReadInt32(f); + m_Index = (wxHtmlContentsItem*) realloc(m_Index, (m_IndexCnt / wxHTML_REALLOC_STEP + 1) * + wxHTML_REALLOC_STEP * sizeof(wxHtmlContentsItem)); + for (i = st; i < m_IndexCnt; i++) + { + m_Index[i].m_Name = CacheReadString(f); + m_Index[i].m_Page = CacheReadString(f); m_Index[i].m_Book = book; } return TRUE; @@ -285,40 +404,40 @@ bool wxHtmlHelpData::LoadCachedBook(wxHtmlBookRecord *book, wxInputStream *f) bool wxHtmlHelpData::SaveCachedBook(wxHtmlBookRecord *book, wxOutputStream *f) { int i; - int x; + wxInt32 cnt; - /* save contents : */ + /* save header - version info : */ + CacheWriteInt32(f, CURRENT_CACHED_BOOK_VERSION); + CacheWriteInt32(f, CACHED_BOOK_FORMAT_FLAGS); - x = 0; - for (i = 0; i < m_ContentsCnt; i++) if (m_Contents[i].m_Book == book && m_Contents[i].m_Level > 0) x++; - f -> Write(&x, sizeof(x)); - for (i = 0; i < m_ContentsCnt; i++) { - if (m_Contents[i].m_Book != book || m_Contents[i].m_Level == 0) continue; - x = m_Contents[i].m_Level; - f -> Write(&x, sizeof(x)); - x = m_Contents[i].m_ID; - f -> Write(&x, sizeof(x)); - x = strlen(m_Contents[i].m_Name) + 1; - f -> Write(&x, sizeof(x)); - f -> Write(m_Contents[i].m_Name, x); - x = strlen(m_Contents[i].m_Page) + 1; - f -> Write(&x, sizeof(x)); - f -> Write(m_Contents[i].m_Page, x); + /* save contents : */ + for (cnt = 0, i = 0; i < m_ContentsCnt; i++) + if (m_Contents[i].m_Book == book && m_Contents[i].m_Level > 0) + cnt++; + CacheWriteInt32(f, cnt); + + for (i = 0; i < m_ContentsCnt; i++) + { + if (m_Contents[i].m_Book != book || m_Contents[i].m_Level == 0) + continue; + CacheWriteInt32(f, m_Contents[i].m_Level); + CacheWriteInt32(f, m_Contents[i].m_ID); + CacheWriteString(f, m_Contents[i].m_Name); + CacheWriteString(f, m_Contents[i].m_Page); } /* save index : */ - - x = 0; - for (i = 0; i < m_IndexCnt; i++) if (m_Index[i].m_Book == book && m_Index[i].m_Level > 0) x++; - f -> Write(&x, sizeof(x)); - for (i = 0; i < m_IndexCnt; i++) { - if (m_Index[i].m_Book != book || m_Index[i].m_Level == 0) continue; - x = strlen(m_Index[i].m_Name) + 1; - f -> Write(&x, sizeof(x)); - f -> Write(m_Index[i].m_Name, x); - x = strlen(m_Index[i].m_Page) + 1; - f -> Write(&x, sizeof(x)); - f -> Write(m_Index[i].m_Page, x); + for (cnt = 0, i = 0; i < m_IndexCnt; i++) + if (m_Index[i].m_Book == book && m_Index[i].m_Level > 0) + cnt++; + CacheWriteInt32(f, cnt); + + for (i = 0; i < m_IndexCnt; i++) + { + if (m_Index[i].m_Book != book || m_Index[i].m_Level == 0) + continue; + CacheWriteString(f, m_Index[i].m_Name); + CacheWriteString(f, m_Index[i].m_Page); } return TRUE; } @@ -327,67 +446,123 @@ bool wxHtmlHelpData::SaveCachedBook(wxHtmlBookRecord *book, wxOutputStream *f) void wxHtmlHelpData::SetTempDir(const wxString& path) { if (path == wxEmptyString) m_TempPath = path; - else { - if (wxIsAbsolutePath(path)) m_TempPath = path; - else m_TempPath = wxGetCwd() + "/" + path; + else + { + if (wxIsAbsolutePath(path)) m_TempPath = path; + else m_TempPath = wxGetCwd() + _T("/") + path; - if (m_TempPath[m_TempPath.Length() - 1] != '/') - m_TempPath << "/"; + if (m_TempPath[m_TempPath.Length() - 1] != _T('/')) + m_TempPath << _T('/'); } } -bool wxHtmlHelpData::AddBookParam(const wxString& title, const wxString& contfile, - const wxString& indexfile, const wxString& deftopic, - const wxString& path) + +static wxString SafeFileName(const wxString& s) +{ + wxString res(s); + res.Replace(wxT("#"), wxT("_")); + res.Replace(wxT(":"), wxT("_")); + res.Replace(wxT("\\"), wxT("_")); + res.Replace(wxT("/"), wxT("_")); + return res; +} + +bool wxHtmlHelpData::AddBookParam(const wxFSFile& bookfile, + wxFontEncoding encoding, + const wxString& title, const wxString& contfile, + const wxString& indexfile, const wxString& deftopic, + const wxString& path) { wxFileSystem fsys; wxFSFile *fi; wxHtmlBookRecord *bookr; - wxString safetitle; - if (! path.IsEmpty()) - // workaround for bug in ChangePathTo(name, TRUE) - fsys.ChangePathTo(path+"/gaga"); + int IndexOld = m_IndexCnt, + ContentsOld = m_ContentsCnt; + + if (!path.IsEmpty()) + fsys.ChangePathTo(path, TRUE); - bookr = new wxHtmlBookRecord(path+'/', title, deftopic); + size_t booksCnt = m_BookRecords.GetCount(); + for (size_t i = 0; i < booksCnt; i++) + { + if ( m_BookRecords[i].GetBookFile() == bookfile.GetLocation() ) + return TRUE; // book is (was) loaded + } + bookr = new wxHtmlBookRecord(bookfile.GetLocation(), fsys.GetPath(), title, deftopic); + if (m_ContentsCnt % wxHTML_REALLOC_STEP == 0) m_Contents = (wxHtmlContentsItem*) realloc(m_Contents, (m_ContentsCnt + wxHTML_REALLOC_STEP) * sizeof(wxHtmlContentsItem)); m_Contents[m_ContentsCnt].m_Level = 0; m_Contents[m_ContentsCnt].m_ID = 0; - m_Contents[m_ContentsCnt].m_Page = new char[deftopic.Length() + 1]; - strcpy(m_Contents[m_ContentsCnt].m_Page, deftopic.c_str()); - m_Contents[m_ContentsCnt].m_Name = new char [title.Length() + 1]; - strcpy(m_Contents[m_ContentsCnt].m_Name, title.c_str()); + m_Contents[m_ContentsCnt].m_Page = new wxChar[deftopic.Length() + 1]; + wxStrcpy(m_Contents[m_ContentsCnt].m_Page, deftopic.c_str()); + m_Contents[m_ContentsCnt].m_Name = new wxChar [title.Length() + 1]; + wxStrcpy(m_Contents[m_ContentsCnt].m_Name, title.c_str()); m_Contents[m_ContentsCnt].m_Book = bookr; // store the contents index for later int cont_start = m_ContentsCnt++; // Try to find cached binary versions: - safetitle = SafeFileName(title); - fi = fsys.OpenFile(safetitle + ".cached"); - if (fi == NULL) fi = fsys.OpenFile(m_TempPath + safetitle + ".cached"); - if ((fi == NULL) || (m_TempPath == wxEmptyString)) { - LoadMSProject(bookr, fsys, indexfile, contfile); - if (m_TempPath != wxEmptyString) { - wxFileOutputStream *outs = new wxFileOutputStream(m_TempPath + safetitle + ".cached"); - SaveCachedBook(bookr, outs); - delete outs; - } - } - else { - LoadCachedBook(bookr, fi -> GetStream()); - delete fi; + // 1. save file as book, but with .hhp.cached extension + // 2. same as 1. but in temp path + // 3. otherwise or if cache load failed, load it from MS. + + fi = fsys.OpenFile(bookfile.GetLocation() + wxT(".cached")); + + if (fi == NULL || +#if wxUSE_DATETIME + fi->GetModificationTime() < bookfile.GetModificationTime() || +#endif // wxUSE_DATETIME + !LoadCachedBook(bookr, fi->GetStream())) + { + if (fi != NULL) delete fi; + fi = fsys.OpenFile(m_TempPath + wxFileNameFromPath(bookfile.GetLocation()) + wxT(".cached")); + if (m_TempPath == wxEmptyString || fi == NULL || +#if wxUSE_DATETIME + fi->GetModificationTime() < bookfile.GetModificationTime() || +#endif // wxUSE_DATETIME + !LoadCachedBook(bookr, fi->GetStream())) + { + LoadMSProject(bookr, fsys, indexfile, contfile); + if (m_TempPath != wxEmptyString) + { + wxFileOutputStream *outs = new wxFileOutputStream(m_TempPath + + SafeFileName(wxFileNameFromPath(bookfile.GetLocation())) + wxT(".cached")); + SaveCachedBook(bookr, outs); + delete outs; + } + } } + if (fi != NULL) delete fi; + // Now store the contents range bookr->SetContentsRange(cont_start, m_ContentsCnt); + // Convert encoding, if neccessary: + if (encoding != wxFONTENCODING_SYSTEM) + { + wxFontEncodingArray a = wxEncodingConverter::GetPlatformEquivalents(encoding); + if (a.GetCount() != 0 && a[0] != encoding) + { + int i; + wxEncodingConverter conv; + conv.Init(encoding, a[0]); + + for (i = IndexOld; i < m_IndexCnt; i++) + conv.Convert(m_Index[i].m_Name); + for (i = ContentsOld; i < m_ContentsCnt; i++) + conv.Convert(m_Contents[i].m_Name); + } + } + m_BookRecords.Add(bookr); if (m_IndexCnt > 0) - qsort(m_Index, m_IndexCnt, sizeof(wxHtmlContentsItem), IndexCompareFunc); + qsort(m_Index, m_IndexCnt, sizeof(wxHtmlContentsItem), wxHtmlHelpIndexCompareFunc); return TRUE; } @@ -395,47 +570,84 @@ bool wxHtmlHelpData::AddBookParam(const wxString& title, const wxString& contfil bool wxHtmlHelpData::AddBook(const wxString& book) { + wxString extension(book.Right(4).Lower()); + if (extension == wxT(".zip") || +#if wxUSE_LIBMSPACK + extension == wxT(".chm") /*compressed html help book*/ || +#endif + extension == wxT(".htb") /*html book*/) + { + wxFileSystem fsys; + wxString s; + bool rt = FALSE; + +#if wxUSE_LIBMSPACK + if (extension == wxT(".chm")) + s = fsys.FindFirst(book + wxT("#chm:*.hhp"), wxFILE); + else +#endif + s = fsys.FindFirst(book + wxT("#zip:*.hhp"), wxFILE); + + while (!s.IsEmpty()) + { + if (AddBook(s)) rt = TRUE; + s = fsys.FindNext(); + } + + return rt; + } + wxFSFile *fi; wxFileSystem fsys; - wxInputStream *s; - wxString bookFull; - - int sz; - char *buff, *lineptr; - char linebuf[300]; wxString title = _("noname"), safetitle, start = wxEmptyString, - contents = wxEmptyString, index = wxEmptyString; - - if (wxIsAbsolutePath(book)) bookFull = book; - else bookFull = wxGetCwd() + "/" + book; - - fi = fsys.OpenFile(bookFull); - if (fi == NULL) return FALSE; - fsys.ChangePathTo(bookFull); - s = fi -> GetStream(); - sz = s -> GetSize(); - buff = new char[sz+1]; - buff[sz] = 0; - s -> Read(buff, sz); - lineptr = buff; - delete fi; - - while ((lineptr = ReadLine(lineptr, linebuf)) != NULL) { - if (strstr(linebuf, "Title=") == linebuf) - title = linebuf + strlen("Title="); - if (strstr(linebuf, "Default topic=") == linebuf) - start = linebuf + strlen("Default topic="); - if (strstr(linebuf, "Index file=") == linebuf) - index = linebuf + strlen("Index file="); - if (strstr(linebuf, "Contents file=") == linebuf) - contents = linebuf + strlen("Contents file="); + contents = wxEmptyString, + index = wxEmptyString, + charset = wxEmptyString; + + fi = fsys.OpenFile(book); + if (fi == NULL) + { + wxLogError(_("Cannot open HTML help book: %s"), book.c_str()); + return FALSE; } - delete[] buff; - - return AddBookParam(title, contents, index, start, fsys.GetPath()); + fsys.ChangePathTo(book); + + const wxChar *lineptr; + wxChar linebuf[300]; + wxString tmp; + wxHtmlFilterPlainText filter; + tmp = filter.ReadFile(*fi); + lineptr = tmp.c_str(); + + do + { + lineptr = ReadLine(lineptr, linebuf, 300); + + for (wxChar *ch = linebuf; *ch != wxT('\0') && *ch != wxT('='); ch++) + *ch = tolower(*ch); + + if (wxStrstr(linebuf, _T("title=")) == linebuf) + title = linebuf + wxStrlen(_T("title=")); + if (wxStrstr(linebuf, _T("default topic=")) == linebuf) + start = linebuf + wxStrlen(_T("default topic=")); + if (wxStrstr(linebuf, _T("index file=")) == linebuf) + index = linebuf + wxStrlen(_T("index file=")); + if (wxStrstr(linebuf, _T("contents file=")) == linebuf) + contents = linebuf + wxStrlen(_T("contents file=")); + if (wxStrstr(linebuf, _T("charset=")) == linebuf) + charset = linebuf + wxStrlen(_T("charset=")); + } while (lineptr != NULL); + + wxFontEncoding enc; + if (charset == wxEmptyString) enc = wxFONTENCODING_SYSTEM; + else enc = wxFontMapper::Get()->CharsetToEncoding(charset); + bool rtval = AddBookParam(*fi, enc, + title, contents, index, start, fsys.GetPath()); + delete fi; + return rtval; } wxString wxHtmlHelpData::FindPageByName(const wxString& x) @@ -449,10 +661,12 @@ wxString wxHtmlHelpData::FindPageByName(const wxString& x) /* 1. try to open given file: */ cnt = m_BookRecords.GetCount(); - for (i = 0; i < cnt; i++) { - f = fsys.OpenFile(m_BookRecords[i].GetBasePath() + x); - if (f) { - url = m_BookRecords[i].GetBasePath() + x; + for (i = 0; i < cnt; i++) + { + f = fsys.OpenFile(m_BookRecords[i].GetFullPath(x)); + if (f) + { + url = m_BookRecords[i].GetFullPath(x); delete f; return url; } @@ -461,9 +675,11 @@ wxString wxHtmlHelpData::FindPageByName(const wxString& x) /* 2. try to find a book: */ - for (i = 0; i < cnt; i++) { - if (m_BookRecords[i].GetTitle() == x) { - url = m_BookRecords[i].GetBasePath() + m_BookRecords[i].GetStart(); + for (i = 0; i < cnt; i++) + { + if (m_BookRecords[i].GetTitle() == x) + { + url = m_BookRecords[i].GetFullPath(m_BookRecords[i].GetStart()); return url; } } @@ -471,9 +687,11 @@ wxString wxHtmlHelpData::FindPageByName(const wxString& x) /* 3. try to find in contents: */ cnt = m_ContentsCnt; - for (i = 0; i < cnt; i++) { - if (strcmp(m_Contents[i].m_Name, x) == 0) { - url = m_Contents[i].m_Book -> GetBasePath() + m_Contents[i].m_Page; + for (i = 0; i < cnt; i++) + { + if (wxStrcmp(m_Contents[i].m_Name, x) == 0) + { + url = m_Contents[i].GetFullPath(); return url; } } @@ -482,9 +700,11 @@ wxString wxHtmlHelpData::FindPageByName(const wxString& x) /* 4. try to find in index: */ cnt = m_IndexCnt; - for (i = 0; i < cnt; i++) { - if (strcmp(m_Index[i].m_Name, x) == 0) { - url = m_Index[i].m_Book -> GetBasePath() + m_Index[i].m_Page; + for (i = 0; i < cnt; i++) + { + if (wxStrcmp(m_Index[i].m_Name, x) == 0) + { + url = m_Index[i].GetFullPath(); return url; } } @@ -497,9 +717,11 @@ wxString wxHtmlHelpData::FindPageById(int id) int i; wxString url(wxEmptyString); - for (i = 0; i < m_ContentsCnt; i++) { - if (m_Contents[i].m_ID == id) { - url = m_Contents[i].m_Book -> GetBasePath() + m_Contents[i].m_Page; + for (i = 0; i < m_ContentsCnt; i++) + { + if (m_Contents[i].m_ID == id) + { + url = m_Contents[i].GetFullPath(); return url; } } @@ -512,69 +734,161 @@ wxString wxHtmlHelpData::FindPageById(int id) //---------------------------------------------------------------------------------- wxHtmlSearchStatus::wxHtmlSearchStatus(wxHtmlHelpData* data, const wxString& keyword, - const wxString& book) + bool case_sensitive, bool whole_words_only, + const wxString& book) { m_Data = data; m_Keyword = keyword; wxHtmlBookRecord* bookr = NULL; - if (book != wxEmptyString) { - // we have to search in a specific book. Find it first - int i,cnt = data->m_BookRecords.GetCount(); - for (i=0; im_BookRecords[i].GetTitle() == book) { - bookr = &(data->m_BookRecords[i]); - m_CurIndex = bookr->GetContentsStart(); - m_MaxIndex = bookr->GetContentsEnd(); - break; - } - // check; we won't crash if the book doesn't exist, but it's Bad Anyway. - wxASSERT(bookr); - } - if (! bookr) { - // no book specified; search all books - m_CurIndex = 0; - m_MaxIndex = m_Data->m_ContentsCnt; - } - m_Engine.LookFor(keyword); + if (book != wxEmptyString) + { + // we have to search in a specific book. Find it first + int i, cnt = data->m_BookRecords.GetCount(); + for (i = 0; i < cnt; i++) + if (data->m_BookRecords[i].GetTitle() == book) + { + bookr = &(data->m_BookRecords[i]); + m_CurIndex = bookr->GetContentsStart(); + m_MaxIndex = bookr->GetContentsEnd(); + break; + } + // check; we won't crash if the book doesn't exist, but it's Bad Anyway. + wxASSERT(bookr); + } + if (! bookr) + { + // no book specified; search all books + m_CurIndex = 0; + m_MaxIndex = m_Data->m_ContentsCnt; + } + m_Engine.LookFor(keyword, case_sensitive, whole_words_only); m_Active = (m_CurIndex < m_MaxIndex); - m_LastPage = wxEmptyString; + m_LastPage = NULL; } bool wxHtmlSearchStatus::Search() { - wxFileSystem fsys; wxFSFile *file; - int i = m_CurIndex; // shortcut + int i = m_CurIndex; // shortcut bool found = FALSE; + wxChar *thepage; - if (! m_Active) { - // sanity check. Illegal use, but we'll try to prevent a crash anyway -#if !defined(__VISAGECPP__) -wxASSERT(0); -#else -wxASSERT(m_Active); -#endif - return FALSE; + if (!m_Active) + { + // sanity check. Illegal use, but we'll try to prevent a crash anyway + wxASSERT(m_Active); + return FALSE; } - m_ContentsItem = NULL; m_Name = wxEmptyString; + m_ContentsItem = NULL; + thepage = m_Data->m_Contents[i].m_Page; - file = fsys.OpenFile(m_Data->m_Contents[i].m_Book -> GetBasePath() + - m_Data->m_Contents[i].m_Page); - if (file) { - if (m_LastPage != file->GetLocation()) { - m_LastPage = file->GetLocation(); - if (m_Engine.Scan(file -> GetStream())) { - m_Name = m_Data->m_Contents[i].m_Name; - m_ContentsItem = m_Data->m_Contents + i; - found = TRUE; - } - } - delete file; - } m_Active = (++m_CurIndex < m_MaxIndex); + // check if it is same page with different anchor: + if (m_LastPage != NULL) + { + wxChar *p1, *p2; + for (p1 = thepage, p2 = m_LastPage; + *p1 != 0 && *p1 != _T('#') && *p1 == *p2; p1++, p2++) {} + + m_LastPage = thepage; + + if (*p1 == 0 || *p1 == _T('#')) + return FALSE; + } + else m_LastPage = thepage; + + wxFileSystem fsys; + file = fsys.OpenFile(m_Data->m_Contents[i].m_Book->GetFullPath(thepage)); + if (file) + { + if (m_Engine.Scan(*file)) + { + m_Name = m_Data->m_Contents[i].m_Name; + m_ContentsItem = m_Data->m_Contents + i; + found = TRUE; + } + delete file; + } return found; } + + + + + + + +//-------------------------------------------------------------------------------- +// wxHtmlSearchEngine +//-------------------------------------------------------------------------------- + +void wxHtmlSearchEngine::LookFor(const wxString& keyword, bool case_sensitive, bool whole_words_only) +{ + m_CaseSensitive = case_sensitive; + m_WholeWords = whole_words_only; + if (m_Keyword) delete[] m_Keyword; + m_Keyword = new wxChar[keyword.Length() + 1]; + wxStrcpy(m_Keyword, keyword.c_str()); + + if (!m_CaseSensitive) + { + for (int i = wxStrlen(m_Keyword) - 1; i >= 0; i--) + { + if ((m_Keyword[i] >= wxT('A')) && (m_Keyword[i] <= wxT('Z'))) + m_Keyword[i] += wxT('a') - wxT('A'); + } + } +} + + +static inline bool WHITESPACE(wxChar c) +{ + return c == _T(' ') || c == _T('\n') || c == _T('\r') || c == _T('\t'); +} + +bool wxHtmlSearchEngine::Scan(const wxFSFile& file) +{ + wxASSERT_MSG(m_Keyword != NULL, wxT("wxHtmlSearchEngine::LookFor must be called before scanning!")); + + int i, j; + int wrd = wxStrlen(m_Keyword); + bool found = FALSE; + wxHtmlFilterHTML filter; + wxString tmp = filter.ReadFile(file); + int lng = tmp.length(); + const wxChar *buf = tmp.c_str(); + + if (!m_CaseSensitive) + for (i = 0; i < lng; i++) + tmp[(size_t)i] = (wxChar)wxTolower(tmp[(size_t)i]); + + if (m_WholeWords) + { + for (i = 0; i < lng - wrd; i++) + { + if (WHITESPACE(buf[i])) continue; + j = 0; + while ((j < wrd) && (buf[i + j] == m_Keyword[j])) j++; + if (j == wrd && WHITESPACE(buf[i + j])) { found = TRUE; break; } + } + } + + else + { + for (i = 0; i < lng - wrd; i++) + { + j = 0; + while ((j < wrd) && (buf[i + j] == m_Keyword[j])) j++; + if (j == wrd) { found = TRUE; break; } + } + } + + return found; +} + + + #endif