X-Git-Url: https://git.creatis.insa-lyon.fr/pubgit/?a=blobdiff_plain;f=src%2FgdcmDocument.cxx;h=1561e8f9e25767d59e521e936384cbc847b2abf7;hb=b0f62020f3423bf7663fdf856000dc245e417d9a;hp=0b907069e0465b3684389aeb0dd4747898ee550d;hpb=84376760a7969280ff8f6152e56bc09e21e332d9;p=gdcm.git diff --git a/src/gdcmDocument.cxx b/src/gdcmDocument.cxx index 0b907069..1561e8f9 100644 --- a/src/gdcmDocument.cxx +++ b/src/gdcmDocument.cxx @@ -3,8 +3,8 @@ Program: gdcm Module: $RCSfile: gdcmDocument.cxx,v $ Language: C++ - Date: $Date: 2005/02/14 10:45:04 $ - Version: $Revision: 1.230 $ + Date: $Date: 2005/08/24 12:09:13 $ + Version: $Revision: 1.267 $ Copyright (c) CREATIS (Centre de Recherche et d'Applications en Traitement de l'Image). All rights reserved. See Doc/License.txt or @@ -31,35 +31,138 @@ #include #include #include +#include // for isdigit +#include // for atoi namespace gdcm { //----------------------------------------------------------------------------- -// Refer to Document::CheckSwap() -//const unsigned int Document::HEADER_LENGTH_TO_READ = 256; // Refer to Document::SetMaxSizeLoadEntry() const unsigned int Document::MAX_SIZE_LOAD_ELEMENT_VALUE = 0xfff; // 4096 -const unsigned int Document::MAX_SIZE_PRINT_ELEMENT_VALUE = 0x7fffffff; //----------------------------------------------------------------------------- // Constructor / Destructor // Constructors and destructors are protected to avoid user to invoke directly + /** - * \brief constructor - * @param filename 'Document' (File or DicomDir) to be opened for parsing + * \brief This default constructor neither loads nor parses the file. + * You should then invoke \ref Document::Load. + * */ -Document::Document( std::string const &filename ) +Document::Document() :ElementSet(-1) { - SetMaxSizeLoadEntry(MAX_SIZE_LOAD_ELEMENT_VALUE); - Filename = filename; + Fp = 0; + + SetMaxSizeLoadEntry(MAX_SIZE_LOAD_ELEMENT_VALUE); Initialize(); + SwapCode = 1234; + Filetype = ExplicitVR; + // Load will set it to true if sucessfull + Group0002Parsed = false; + IsDocumentAlreadyLoaded = false; + IsDocumentModified = true; + LoadMode = 0x00000000; // default : load everything, later + SetFileName(""); +} + +/** + * \brief Constructor (DEPRECATED : not to break the API) + * @param fileName 'Document' (File or DicomDir) to be open for parsing + */ +Document::Document( std::string const &fileName ) + :ElementSet(-1) +{ + Fp = 0; + + SetMaxSizeLoadEntry(MAX_SIZE_LOAD_ELEMENT_VALUE); + Initialize(); + SwapCode = 1234; + Filetype = ExplicitVR; + Group0002Parsed = false; + LoadMode = 0x00000000; // Load everything, later + + // Load will set it to true if sucessfull + IsDocumentAlreadyLoaded = false; + IsDocumentModified = true; + + SetFileName(fileName); + Load( ); +} +/** + * \brief Canonical destructor. + */ +Document::~Document () +{ + RefPubDict = NULL; + RefShaDict = NULL; +} + +//----------------------------------------------------------------------------- +// Public + +/** + * \brief Loader. use SetLoadMode(), SetFileName() before ! + * @return false if file cannot be open or no swap info was found, + * or no tag was found. + */ +bool Document::Load( ) +{ + if ( GetFileName() == "" ) + { + gdcmWarningMacro( "Use SetFileName, before !" ); + return false; + } + return DoTheLoadingDocumentJob( ); +} +/** + * \brief Loader. (DEPRECATED : not to break the API) + * @param fileName 'Document' (File or DicomDir) to be open for parsing + * @return false if file cannot be open or no swap info was found, + * or no tag was found. + */ +bool Document::Load( std::string const &fileName ) +{ + Filename = fileName; + return DoTheLoadingDocumentJob( ); +} + +/** + * \brief Performs the Loading Job (internal use only) + * @return false if file cannot be open or no swap info was found, + * or no tag was found. + */ +bool Document::DoTheLoadingDocumentJob( ) +{ + if ( ! IsDocumentModified ) // Nothing to do ! + return true; + + // if ( Filename == fileName ) + // { + // gdcmWarningMacro( "The file was already parsed inside this " + // << "gdcm::Document (its name is: " + // << Filename.c_str() ); + // return true; + // } + + //gdcmWarningMacro( "A file was already parsed inside this " + // << "gdcm::Document (previous name was: " + // << Filename.c_str() << ". New name is :" + // << fileName ); + // clean out the Entries, if already parsed + // (probabely a mistake from the user) + + ClearEntry(); Fp = 0; if ( !OpenFile() ) { - return; + // warning already performed in OpenFile() + //gdcmWarningMacro( "Unable to open as an ACR/DICOM file: " + // << Filename.c_str() ); + Filetype = Unknown; + return false; } Group0002Parsed = false; @@ -67,25 +170,50 @@ Document::Document( std::string const &filename ) gdcmWarningMacro( "Starting parsing of file: " << Filename.c_str()); Fp->seekg(0, std::ios::end); - long lgt = Fp->tellg(); - + long lgt = Fp->tellg(); // total length of the file + Fp->seekg(0, std::ios::beg); - CheckSwap(); - long beg = Fp->tellg(); - lgt -= beg; - - ParseDES( this, beg, lgt, false); // Loading is done during parsing + // CheckSwap returns a boolean + // (false if no swap info of any kind was found) + if (! CheckSwap() ) + { + gdcmWarningMacro( "Neither a DICOM V3 nor an ACR-NEMA file: " + << Filename.c_str()); + CloseFile(); + return false; + } + + long beg = Fp->tellg(); // just after DICOM preamble (if any) + + lgt -= beg; // remaining length to parse + + // Recursive call. + // Loading is done during parsing + ParseDES( this, beg, lgt, false); // delim_mode is first defaulted to false + + if ( IsEmpty() ) + { + gdcmWarningMacro( "No tag in internal hash table for: " + << Filename.c_str()); + CloseFile(); + return false; + } + IsDocumentAlreadyLoaded = true; - Fp->seekg( 0, std::ios::beg); + Fp->seekg( 0, std::ios::beg); // Load 'non string' values std::string PhotometricInterpretation = GetEntryValue(0x0028,0x0004); - if( PhotometricInterpretation == "PALETTE COLOR " ) + if ( PhotometricInterpretation == "PALETTE COLOR " ) { - LoadEntryBinArea(0x0028,0x1200); // gray LUT - /// FIXME FIXME FIXME + // FIXME + // Probabely this line should be outside the 'if' + // Try to find an image sample holding a 'gray LUT' + LoadEntryBinArea(0x0028,0x1200); // gray LUT + + /// FIXME /// The tags refered by the three following lines used to be CORRECTLY /// defined as having an US Value Representation in the public /// dictionary. BUT the semantics implied by the three following @@ -95,7 +223,7 @@ Document::Document( std::string const &filename ) /// altered on PURPOSE but now contains a WRONG value. /// In order to fix things and restore the dictionary to its /// correct value, one needs to decided of the semantics by deciding - /// whether the following tags are either: + /// whether the following tags are either : /// - multivaluated US, and hence loaded as ValEntry, but afterwards /// also used as BinEntry, which requires the proper conversion, /// - OW, and hence loaded as BinEntry, but afterwards also used @@ -110,18 +238,34 @@ Document::Document( std::string const &filename ) LoadEntryBinArea(0x0028,0x1222); // Segmented Blue Palette Color LUT Data LoadEntryBinArea(0x0028,0x1223); - } + } + //FIXME later : how to use it? - LoadEntryBinArea(0x0028,0x3006); //LUT Data (CTX dependent) + SeqEntry *modLutSeq = GetSeqEntry(0x0028,0x3000); + if ( modLutSeq !=0 ) + { + SQItem *sqi= modLutSeq->GetFirstSQItem(); + if ( sqi != 0 ) + { + BinEntry *b = sqi->GetBinEntry(0x0028,0x3006); + if ( b != 0 ) + { + if ( b->GetLength() != 0 ) + { + LoadEntryBinArea(b); //LUT Data (CTX dependent) + } + } + } + } CloseFile(); // ---------------------------- // Specific code to allow gdcm to read ACR-LibIDO formated images // Note: ACR-LibIDO is an extension of the ACR standard that was - // used at CREATIS. For the time being (say a couple years) - // we keep this kludge to allow a smooth move to gdcm for - // CREATIS developpers (sorry folks). + // used at CREATIS. For the time being (say a couple of years) + // we keep this kludge to allow CREATIS users + // reading their old images. // // if recognition code tells us we deal with a LibIDO image // we switch lineNumber and columnNumber @@ -139,35 +283,10 @@ Document::Document( std::string const &filename ) SetValEntry(rows , 0x0028, 0x0011); } // --- End of ACR-LibIDO kludge --- -} -/** - * \brief This default constructor doesn't parse the file. You should - * then invoke \ref Document::SetFileName and then the parsing. - */ -Document::Document() - :ElementSet(-1) -{ - Fp = 0; - - SetMaxSizeLoadEntry(MAX_SIZE_LOAD_ELEMENT_VALUE); - Initialize(); - SwapCode = 1234; - Filetype = ExplicitVR; - Group0002Parsed = false; -} - -/** - * \brief Canonical destructor. - */ -Document::~Document () -{ - RefPubDict = NULL; - RefShaDict = NULL; + return true; } -//----------------------------------------------------------------------------- -// Public /** * \brief Get the public dictionary used */ @@ -205,16 +324,15 @@ bool Document::SetShaDict(DictKey const &dictName) } /** - * \brief This predicate, based on hopefully reasonable heuristics, - * decides whether or not the current Document was properly parsed - * and contains the mandatory information for being considered as - * a well formed and usable Dicom/Acr File. - * @return true when Document is the one of a reasonable Dicom/Acr file, - * false otherwise. + * \brief This predicate tells us whether or not the current Document + * was properly parsed and contains at least *one* Dicom Element + * (and nothing more, sorry). + * @return false when we're 150 % sure it's NOT a Dicom/Acr file, + * true otherwise. */ bool Document::IsReadable() { - if( Filetype == Unknown) + if ( Filetype == Unknown ) { gdcmWarningMacro( "Wrong filetype"); return false; @@ -236,7 +354,7 @@ bool Document::IsReadable() bool Document::IsDicomV3() { // Checking if Transfer Syntax exists is enough - // Anyway, it's to late check if the 'Preamble' was found ... + // Anyway, it's too late check if the 'Preamble' was found ... // And ... would it be a rich idea to check ? // (some 'no Preamble' DICOM images exist !) return GetDocEntry(0x0002, 0x0010) != NULL; @@ -397,60 +515,67 @@ std::ifstream *Document::OpenFile() return 0; } - if(Fp) + if ( Fp ) { gdcmWarningMacro( "File already open: " << Filename.c_str()); CloseFile(); } Fp = new std::ifstream(Filename.c_str(), std::ios::in | std::ios::binary); - if( ! *Fp ) + if ( ! *Fp ) { - gdcmDebugMacro( "Cannot open file: " << Filename.c_str()); + // Don't user gdcmErrorMacro : + // a spurious message will appear when you use, for instance + // gdcm::FileHelper *fh = new gdcm::FileHelper( outputFileName ); + // to create outputFileName. + gdcmWarningMacro( "Cannot open file: " << Filename.c_str()); delete Fp; Fp = 0; return 0; + //exit(1); // No function is allowed to leave the application instead + // of warning the caller } uint16_t zero = 0; Fp->read((char*)&zero, (size_t)2); - if( Fp->eof() ) + if ( Fp->eof() ) { CloseFile(); return 0; } - //ACR -- or DICOM with no Preamble; may start with a Shadow Group -- - if( + //-- ACR or DICOM with no Preamble; may start with a Shadow Group -- + if ( zero == 0x0001 || zero == 0x0100 || zero == 0x0002 || zero == 0x0200 || zero == 0x0003 || zero == 0x0300 || zero == 0x0004 || zero == 0x0400 || zero == 0x0005 || zero == 0x0500 || zero == 0x0006 || zero == 0x0600 || zero == 0x0007 || zero == 0x0700 || zero == 0x0008 || zero == 0x0800 ) { - std::string msg - = Util::Format("ACR/DICOM with no preamble: (%04x)\n", zero); + std::string msg = Util::Format( + "ACR/DICOM starting at the begining of the file:(%04x)\n", zero); gdcmWarningMacro( msg.c_str() ); return Fp; } - //DICOM + //-- DICOM -- Fp->seekg(126L, std::ios::cur); char dicm[4] = {' ',' ',' ',' '}; Fp->read(dicm, (size_t)4); - if( Fp->eof() ) + if ( Fp->eof() ) { CloseFile(); return 0; } - if( memcmp(dicm, "DICM", 4) == 0 ) + if ( memcmp(dicm, "DICM", 4) == 0 ) { HasDCMPreamble = true; return Fp; } - + + // -- Neither ACR/No Preamble Dicom nor DICOMV3 file CloseFile(); - gdcmWarningMacro( "Not DICOM/ACR (missing preamble)" << Filename.c_str()); - + gdcmWarningMacro( "Neither ACR/No Preamble Dicom nor DICOMV3 file: " + << Filename.c_str()); return 0; } @@ -460,7 +585,7 @@ std::ifstream *Document::OpenFile() */ bool Document::CloseFile() { - if( Fp ) + if ( Fp ) { Fp->close(); delete Fp; @@ -524,7 +649,7 @@ void Document::LoadEntryBinArea(uint16_t group, uint16_t elem) return; BinEntry *binElement = dynamic_cast(docElement); - if( !binElement ) + if ( !binElement ) return; LoadEntryBinArea(binElement); @@ -537,11 +662,11 @@ void Document::LoadEntryBinArea(uint16_t group, uint16_t elem) */ void Document::LoadEntryBinArea(BinEntry *elem) { - if(elem->GetBinArea()) + if (elem->GetBinArea() ) return; bool openFile = !Fp; - if(openFile) + if ( openFile ) OpenFile(); size_t o =(size_t)elem->GetOffset(); @@ -549,14 +674,14 @@ void Document::LoadEntryBinArea(BinEntry *elem) size_t l = elem->GetLength(); uint8_t *a = new uint8_t[l]; - if( !a ) + if ( !a ) { gdcmWarningMacro( "Cannot allocate BinEntry content"); return; } Fp->read((char*)a, l); - if( Fp->fail() || Fp->eof()) + if ( Fp->fail() || Fp->eof() ) { delete[] a; return; @@ -564,7 +689,7 @@ void Document::LoadEntryBinArea(BinEntry *elem) elem->SetBinArea(a); - if(openFile) + if ( openFile ) CloseFile(); } @@ -576,7 +701,7 @@ void Document::LoadEntryBinArea(BinEntry *elem) */ void Document::LoadDocEntrySafe(DocEntry *entry) { - if(Fp) + if ( Fp ) { long PositionOnEntry = Fp->tellg(); LoadDocEntry(entry); @@ -596,11 +721,11 @@ bool Document::operator<(Document &document) // Patient Name std::string s1 = GetEntryValue(0x0010,0x0010); std::string s2 = document.GetEntryValue(0x0010,0x0010); - if(s1 < s2) + if (s1 < s2) { return true; } - else if( s1 > s2 ) + else if ( s1 > s2 ) { return false; } @@ -626,7 +751,7 @@ bool Document::operator<(Document &document) { return true; } - else if( s1 > s2 ) + else if ( s1 > s2 ) { return false; } @@ -639,7 +764,7 @@ bool Document::operator<(Document &document) { return true; } - else if( s1 > s2 ) + else if ( s1 > s2 ) { return false; } @@ -665,7 +790,7 @@ uint16_t Document::ReadInt16() { throw FormatError( "Document::ReadInt16()", " file error." ); } - if( Fp->eof() ) + if ( Fp->eof() ) { throw FormatError( "Document::ReadInt16()", "EOF." ); } @@ -687,7 +812,7 @@ uint32_t Document::ReadInt32() { throw FormatError( "Document::ReadInt32()", " file error." ); } - if( Fp->eof() ) + if ( Fp->eof() ) { throw FormatError( "Document::ReadInt32()", "EOF." ); } @@ -724,19 +849,19 @@ int Document::ComputeGroup0002Length( FileType filetype ) { gr = entry->GetGroup(); - if( gr == 0x0002 ) + if ( gr == 0x0002 ) { found0002 = true; - if( entry->GetElement() != 0x0000 ) + if ( entry->GetElement() != 0x0000 ) { vr = entry->GetVR(); - if( filetype == ExplicitVR ) + if ( filetype == ExplicitVR ) { - if ( (vr == "OB") || (vr == "OW") || (vr == "SQ") ) + if ( (vr == "OB") || (vr == "OW") || (vr == "SQ") || (vr == "UT") ) { - // explicit VR AND OB, OW, SQ : 4 more bytes + // explicit VR AND OB, OW, SQ, UT : 4 more bytes groupLength += 4; } } @@ -766,18 +891,22 @@ void Document::Initialize() /** * \brief Parses a DocEntrySet (Zero-level DocEntries or SQ Item DocEntries) - * @return length of the parsed set. + * @param set DocEntrySet we are going to parse ('zero level' or a SQItem) + * @param offset start of parsing + * @param l_max length to parse (meaningless when we are in 'delimitor mode') + * @param delim_mode : whether we are in 'delimitor mode' (l=0xffffff) or not */ void Document::ParseDES(DocEntrySet *set, long offset, long l_max, bool delim_mode) { - DocEntry *newDocEntry = 0; + DocEntry *newDocEntry; ValEntry *newValEntry; BinEntry *newBinEntry; SeqEntry *newSeqEntry; VRKey vr; - bool used = false; - + bool used; + bool delim_mode_intern = delim_mode; + bool first = true; while (true) { if ( !delim_mode && ((long)(Fp->tellg())-offset) >= l_max) @@ -785,23 +914,43 @@ void Document::ParseDES(DocEntrySet *set, long offset, break; } - used = true; newDocEntry = ReadNextDocEntry( ); + // Uncoment this printf line to be able to 'follow' the DocEntries + // when something *very* strange happens + + printf( "%04x|%04x %s\n",newDocEntry->GetGroup(), + newDocEntry->GetElement(), + newDocEntry->GetVR().c_str() ); + if ( !newDocEntry ) { break; } - vr = newDocEntry->GetVR(); + // an Item Starter found elsewhere but the first postition + // of a SeqEntry means previous entry was a Sequence + // but we didn't get it (private Sequence + Implicit VR) + // we have to backtrack. + if ( !first && newDocEntry->IsItemStarter() ) + { + newDocEntry = Backtrack(newDocEntry); + } + else + { + PreviousDocEntry = newDocEntry; + } + + used = true; newValEntry = dynamic_cast(newDocEntry); newBinEntry = dynamic_cast(newDocEntry); - newSeqEntry = dynamic_cast(newDocEntry); - if ( newValEntry || newBinEntry ) + if ( newValEntry || newBinEntry ) { + //////////////////////////// ContentEntry if ( newBinEntry ) { + vr = newDocEntry->GetVR(); if ( Filetype == ExplicitVR && !Global::GetVR()->IsVROfBinaryRepresentable(vr) ) { @@ -813,6 +962,7 @@ void Document::ParseDES(DocEntrySet *set, long offset, } //////////////////// BinEntry or UNKOWN VR: + // When "this" is a Document the Key is simply of the // form ( group, elem )... if ( dynamic_cast< Document* > ( set ) ) @@ -822,23 +972,32 @@ void Document::ParseDES(DocEntrySet *set, long offset, // but when "this" is a SQItem, we are inserting this new // valEntry in a sequence item, and the key has the // generalized form (refer to \ref BaseTagKey): - if (SQItem *parentSQItem = dynamic_cast< SQItem* > ( set ) ) + + // time waste hunting + //if (SQItem *parentSQItem = dynamic_cast< SQItem* > ( set ) ) + //{ + // newBinEntry->SetKey( parentSQItem->GetBaseTagKey() + // + newBinEntry->GetKey() ); + //} + + if ( !set->AddEntry( newBinEntry ) ) { - newBinEntry->SetKey( parentSQItem->GetBaseTagKey() - + newBinEntry->GetKey() ); + gdcmWarningMacro( "in ParseDES : cannot add a BinEntry " + << newBinEntry->GetKey() + << " (at offset : " + << newBinEntry->GetOffset() << " )" ); + used=false; } - - LoadDocEntry( newBinEntry ); - if( !set->AddEntry( newBinEntry ) ) + else { - //Expect big troubles if here - //delete newBinEntry; - used=false; + // Load only if we can add (not a duplicate key) + LoadDocEntry( newBinEntry ); } - } + } // end BinEntry else { /////////////////////// ValEntry + // When "set" is a Document, then we are at the top of the // hierarchy and the Key is simply of the form ( group, elem )... if ( dynamic_cast< Document* > ( set ) ) @@ -848,30 +1007,55 @@ void Document::ParseDES(DocEntrySet *set, long offset, // ...but when "set" is a SQItem, we are inserting this new // valEntry in a sequence item. Hence the key has the // generalized form (refer to \ref BaseTagKey): - if (SQItem *parentSQItem = dynamic_cast< SQItem* > ( set ) ) + + // time waste hunting + //if (SQItem *parentSQItem = dynamic_cast< SQItem* > ( set ) ) + //{ + // newValEntry->SetKey( parentSQItem->GetBaseTagKey() + // + newValEntry->GetKey() ); + //} + + if ( !set->AddEntry( newValEntry ) ) { - newValEntry->SetKey( parentSQItem->GetBaseTagKey() - + newValEntry->GetKey() ); + gdcmWarningMacro( "in ParseDES : cannot add a ValEntry " + << newValEntry->GetKey() + << " (at offset : " + << newValEntry->GetOffset() << " )" ); + used=false; } - - LoadDocEntry( newValEntry ); - bool delimitor=newValEntry->IsItemDelimitor(); - if( !set->AddEntry( newValEntry ) ) + else { - // If here expect big troubles - //delete newValEntry; //otherwise mem leak - used=false; + // Load only if we can add (not a duplicate key) + LoadDocEntry( newValEntry ); } - if (delimitor) + if ( newValEntry->GetElement() == 0x0000 ) // if on group length { - if(!used) - delete newDocEntry; - break; - } - if ( !delim_mode && ((long)(Fp->tellg())-offset) >= l_max) + if ( newValEntry->GetGroup()%2 != 0 ) // if Shadow Group + { + if ( LoadMode & NO_SHADOW ) // if user asked to skip shad.gr + { + std::string strLgrGroup = newValEntry->GetValue(); + int lgrGroup; + if ( strLgrGroup != GDCM_UNFOUND) + { + lgrGroup = atoi(strLgrGroup.c_str()); + Fp->seekg(lgrGroup, std::ios::cur); + used = false; + RemoveEntry( newDocEntry ); + newDocEntry = 0; + continue; + } + } + } + } + + bool delimitor=newValEntry->IsItemDelimitor(); + + if ( (delimitor) || + (!delim_mode && ((long)(Fp->tellg())-offset) >= l_max) ) { - if(!used) + if ( !used ) delete newDocEntry; break; } @@ -882,63 +1066,95 @@ void Document::ParseDES(DocEntrySet *set, long offset, } else { - // VR = "SQ" - unsigned long l = newDocEntry->GetReadLength(); - if ( l != 0 ) // don't mess the delim_mode for zero-length sequence + /////////////////////// SeqEntry : VR = "SQ" + + unsigned long l = newDocEntry->GetReadLength(); + if ( l != 0 ) // don't mess the delim_mode for 'zero-length sequence' { if ( l == 0xffffffff ) { - delim_mode = true; + delim_mode_intern = true; } else { - delim_mode = false; + delim_mode_intern = false; } } - // no other way to create it ... - newSeqEntry->SetDelimitorMode( delim_mode ); + + if ( (LoadMode & NO_SHADOWSEQ) && ! delim_mode_intern ) + { + // User asked to skip SeQuences *only* if they belong to Shadow Group + if ( newDocEntry->GetGroup()%2 != 0 ) + { + Fp->seekg( l, std::ios::cur); + used = false; + continue; + } + } + if ( (LoadMode & NO_SEQ) && ! delim_mode_intern ) + { + // User asked to skip *any* SeQuence + Fp->seekg( l, std::ios::cur); + used = false; + continue; + } + // delay the dynamic cast as late as possible + newSeqEntry = dynamic_cast(newDocEntry); + + // no other way to create the Delimitor ... + newSeqEntry->SetDelimitorMode( delim_mode_intern ); // At the top of the hierarchy, stands a Document. When "set" // is a Document, then we are building the first depth level. // Hence the SeqEntry we are building simply has a depth // level of one: - if (/*Document *dummy =*/ dynamic_cast< Document* > ( set ) ) + if ( dynamic_cast< Document* > ( set ) ) { - //(void)dummy; newSeqEntry->SetDepthLevel( 1 ); newSeqEntry->SetKey( newSeqEntry->GetKey() ); } // But when "set" is already a SQItem, we are building a nested // sequence, and hence the depth level of the new SeqEntry // we are building, is one level deeper: + + // time waste hunting if (SQItem *parentSQItem = dynamic_cast< SQItem* > ( set ) ) { newSeqEntry->SetDepthLevel( parentSQItem->GetDepthLevel() + 1 ); - newSeqEntry->SetKey( parentSQItem->GetBaseTagKey() - + newSeqEntry->GetKey() ); + + // newSeqEntry->SetKey( parentSQItem->GetBaseTagKey() + // + newSeqEntry->GetKey() ); } if ( l != 0 ) { // Don't try to parse zero-length sequences ParseSQ( newSeqEntry, newDocEntry->GetOffset(), - l, delim_mode); + l, delim_mode_intern); } - if( !set->AddEntry( newSeqEntry ) ) + if ( !set->AddEntry( newSeqEntry ) ) { + gdcmWarningMacro( "in ParseDES : cannot add a SeqEntry " + << newSeqEntry->GetKey() + << " (at offset : " + << newSeqEntry->GetOffset() << " )" ); used = false; } + if ( !delim_mode && ((long)(Fp->tellg())-offset) >= l_max) { - if( !used ) - delete newDocEntry; - break; + if ( !used ) + delete newDocEntry; + break; } - } + } // end SeqEntry : VR = "SQ" - if( !used ) + if ( !used ) + { delete newDocEntry; - } + } + first = false; + } // end While } /** @@ -960,9 +1176,10 @@ void Document::ParseSQ( SeqEntry *seqEntry, if ( !newDocEntry ) { // FIXME Should warn user + gdcmWarningMacro("in ParseSQ : should never get here!"); break; } - if( delim_mode ) + if ( delim_mode ) { if ( newDocEntry->IsSequenceDelimitor() ) { @@ -977,12 +1194,14 @@ void Document::ParseSQ( SeqEntry *seqEntry, } // create the current SQItem SQItem *itemSQ = new SQItem( seqEntry->GetDepthLevel() ); +/* std::ostringstream newBase; newBase << seqEntry->GetKey() << "/" << SQItemNumber << "#"; itemSQ->SetBaseTagKey( newBase.str() ); +*/ unsigned int l = newDocEntry->GetReadLength(); if ( l == 0xffffffff ) @@ -999,7 +1218,9 @@ void Document::ParseSQ( SeqEntry *seqEntry, delete newDocEntry; Fp->seekg(offsetStartCurrentSQItem, std::ios::beg); // fill up the current SQItem, starting at the beginning of fff0,e000 + ParseDES(itemSQ, offsetStartCurrentSQItem, l+8, dlm_mod); + offsetStartCurrentSQItem = Fp->tellg(); // end try ----------------- @@ -1012,6 +1233,41 @@ void Document::ParseSQ( SeqEntry *seqEntry, } } +/** + * \brief When a private Sequence + Implicit VR is encountered + * we cannot guess it's a Sequence till we find the first + * Item Starter. We then backtrack to do the job. + * @param docEntry Item Starter that warned us + */ +DocEntry *Document::Backtrack(DocEntry *docEntry) +{ + // delete the Item Starter, built erroneously out of any Sequence + // it's not yet in the HTable/chained list + delete docEntry; + + // Get all info we can from PreviousDocEntry + uint16_t group = PreviousDocEntry->GetGroup(); + uint16_t elem = PreviousDocEntry->GetElement(); + uint32_t lgt = PreviousDocEntry->GetLength(); + long offset = PreviousDocEntry->GetOffset(); + + gdcmWarningMacro( "Backtrack :" << std::hex << group + << "|" << elem + << " at offset " << offset ); + RemoveEntry( PreviousDocEntry ); + + // forge the Seq Entry + DocEntry *newEntry = NewSeqEntry(group, elem); + newEntry->SetLength(lgt); + newEntry->SetOffset(offset); + + // Move back to the beginning of the Sequence + Fp->seekg( 0, std::ios::beg); + Fp->seekg(offset, std::ios::cur); + +return newEntry; +} + /** * \brief Loads the element content if its length doesn't exceed * the value specified with Document::SetMaxSizeLoadEntry() @@ -1029,7 +1285,7 @@ void Document::LoadDocEntry(DocEntry *entry) // (fffe e000) tells us an Element is beginning // (fffe e00d) tells us an Element just ended // (fffe e0dd) tells us the current SeQuence just ended - if( group == 0xfffe ) + if ( group == 0xfffe ) { // NO more value field for SQ ! return; @@ -1043,7 +1299,7 @@ void Document::LoadDocEntry(DocEntry *entry) } // The elements whose length is bigger than the specified upper bound - // are not loaded. Instead we leave a short notice of the offset of + // are not loaded. Instead we leave a short notice on the offset of // the element content and it's length. std::ostringstream s; @@ -1051,17 +1307,16 @@ void Document::LoadDocEntry(DocEntry *entry) { if (BinEntry *binEntryPtr = dynamic_cast< BinEntry* >(entry) ) { - //s << "gdcm::NotLoaded (BinEntry)"; s << GDCM_NOTLOADED; - s << " Address:" << (long)entry->GetOffset(); - s << " Length:" << entry->GetLength(); + s << " Ad.:" << (long)entry->GetOffset(); + s << " x(" << std::hex << entry->GetOffset() << ")"; + s << std::dec; + s << " Lgt:" << entry->GetLength(); s << " x(" << std::hex << entry->GetLength() << ")"; binEntryPtr->SetValue(s.str()); } - // Be carefull : a BinEntry IS_A ValEntry ... else if (ValEntry *valEntryPtr = dynamic_cast< ValEntry* >(entry) ) { - // s << "gdcm::NotLoaded. (ValEntry)"; s << GDCM_NOTLOADED; s << " Address:" << (long)entry->GetOffset(); s << " Length:" << entry->GetLength(); @@ -1095,7 +1350,7 @@ void Document::LoadDocEntry(DocEntry *entry) uint32_t NewInt; int nbInt; // When short integer(s) are expected, read and convert the following - // n *two characters properly i.e. consider them as short integers as + // (n * 2) characters properly i.e. consider them as short integers as // opposed to strings. // Elements with Value Multiplicity > 1 // contain a set of integers (not a single one) @@ -1144,11 +1399,11 @@ void Document::LoadDocEntry(DocEntry *entry) str[length] = '\0'; //this is only useful when length is odd // Special DicomString call to properly handle \0 and even length std::string newValue; - if( length % 2 ) + if ( length % 2 ) { newValue = Util::DicomString(str, length+1); gdcmWarningMacro("Warning: bad length: " << length << - ",For string :" << newValue.c_str()); + " For string :" << newValue.c_str()); // Since we change the length of string update it length //entry->SetReadLength(length+1); } @@ -1162,12 +1417,17 @@ void Document::LoadDocEntry(DocEntry *entry) { if ( Fp->fail() || Fp->eof()) { - gdcmWarningMacro("Unread element value"); + if ( Fp->fail() ) + gdcmWarningMacro("--> fail"); + + gdcmWarningMacro("Unread element value " << valEntry->GetKey() + << " lgt : " << valEntry->GetReadLength() + << " at " << std::hex << valEntry->GetOffset()); valEntry->SetValue(GDCM_UNREAD); return; } - if( vr == "UI" ) + if ( vr == "UI" ) { // Because of correspondance with the VR dic valEntry->SetValue(newValue); @@ -1179,7 +1439,9 @@ void Document::LoadDocEntry(DocEntry *entry) } else { - gdcmErrorMacro( "Should have a ValEntry, here !"); + gdcmWarningMacro("Should have a ValEntry, here ! " << valEntry->GetKey() + << " lgt : " << valEntry->GetReadLength() + << " at " << std::hex << valEntry->GetOffset()); } } @@ -1195,7 +1457,7 @@ void Document::FindDocEntryLength( DocEntry *entry ) if ( Filetype == ExplicitVR && !entry->IsImplicitVR() ) { - if ( vr == "OB" || vr == "OW" || vr == "SQ" || vr == "UN" ) + if ( vr == "OB" || vr == "OW" || vr == "SQ" || vr == "UT" || vr == "UN" ) { // The following reserved two bytes (see PS 3.5-2003, section // "7.1.2 Data element structure with explicit vr", p 27) must be @@ -1237,72 +1499,6 @@ void Document::FindDocEntryLength( DocEntry *entry ) // Length is encoded on 2 bytes. length16 = ReadInt16(); - - // FIXME : This heuristic supposes that the first group following - // group 0002 *has* and element 0000. - // BUT ... Element 0000 is optionnal :-( - - - // Fixed using : HandleOutOfGroup0002() - // (first hereafter strategy ...) - - // We can tell the current file is encoded in big endian (like - // Data/US-RGB-8-epicard) when we find the "Transfer Syntax" tag - // and it's value is the one of the encoding of a big endian file. - // In order to deal with such big endian encoded files, we have - // (at least) two strategies: - // * when we load the "Transfer Syntax" tag with value of big endian - // encoding, we raise the proper flags. Then we wait for the end - // of the META group (0x0002) among which is "Transfer Syntax", - // before switching the swap code to big endian. We have to postpone - // the switching of the swap code since the META group is fully encoded - // in little endian, and big endian coding only starts at the next - // group. The corresponding code can be hard to analyse and adds - // many additional unnecessary tests for regular tags. - // * the second strategy consists in waiting for trouble, that shall - // appear when we find the first group with big endian encoding. This - // is easy to detect since the length of a "Group Length" tag (the - // ones with zero as element number) has to be of 4 (0x0004). When we - // encounter 1024 (0x0400) chances are the encoding changed and we - // found a group with big endian encoding. - //---> Unfortunately, element 0000 is optional. - //---> This will not work when missing! - // We shall use this second strategy. In order to make sure that we - // can interpret the presence of an apparently big endian encoded - // length of a "Group Length" without committing a big mistake, we - // add an additional check: we look in the already parsed elements - // for the presence of a "Transfer Syntax" whose value has to be "big - // endian encoding". When this is the case, chances are we have got our - // hands on a big endian encoded file: we switch the swap code to - // big endian and proceed... - -// if ( element == 0x0000 && length16 == 0x0400 ) -// { -// std::string ts = GetTransferSyntax(); -// if ( Global::GetTS()->GetSpecialTransferSyntax(ts) -// != TS::ExplicitVRBigEndian ) -// { -// throw FormatError( "Document::FindDocEntryLength()", -// " not explicit VR." ); -// return; -// } -// length16 = 4; -// SwitchByteSwapCode(); -// -// // Restore the unproperly loaded values i.e. the group, the element -// // and the dictionary entry depending on them. -// uint16_t correctGroup = SwapShort( entry->GetGroup() ); -// uint16_t correctElem = SwapShort( entry->GetElement() ); -// DictEntry *newTag = GetDictEntry( correctGroup, correctElem ); -// if ( !newTag ) -// { -// // This correct tag is not in the dictionary. Create a new one. -// newTag = NewVirtualDictEntry(correctGroup, correctElem); -// } -// // FIXME this can create a memory leaks on the old entry that be -// // left unreferenced. -// entry->SetDictEntry( newTag ); -// } // 0xffff means that we deal with 'No Length' Sequence // or 'No Length' SQItem @@ -1362,12 +1558,14 @@ uint32_t Document::FindDocEntryLengthOBOrOW() if ( group != 0xfffe || ( ( elem != 0xe0dd ) && ( elem != 0xe000 ) ) ) { long filePosition = Fp->tellg(); - gdcmWarningMacro( "Neither an Item tag nor a Sequence delimiter tag on :" + gdcmWarningMacro( + "Neither an Item tag nor a Sequence delimiter tag on :" << std::hex << group << " , " << elem << ") -before- position x(" << filePosition << ")" ); Fp->seekg(positionOnEntry, std::ios::beg); - throw FormatUnexpected( "Neither an Item tag nor a Sequence delimiter tag."); + throw FormatUnexpected( + "Neither an Item tag nor a Sequence delimiter tag."); } if ( elem == 0xe0dd ) { @@ -1409,7 +1607,7 @@ std::string Document::FindDocEntryVR() Fp->read (vr, (size_t)2); vr[2] = 0; - if( !CheckDocEntryVR(vr) ) + if ( !CheckDocEntryVR(vr) ) { Fp->seekg(positionOnEntry, std::ios::beg); return GDCM_UNKNOWN; @@ -1457,14 +1655,14 @@ std::string Document::GetDocEntryValue(DocEntry *entry) // Elements with Value Multiplicity > 1 // contain a set of short integers (not a single one) - if( vr == "US" || vr == "SS" ) + if ( vr == "US" || vr == "SS" ) { uint16_t newInt16; nbInt = length / 2; for (int i=0; i < nbInt; i++) { - if( i != 0 ) + if ( i != 0 ) { s << '\\'; } @@ -1480,14 +1678,14 @@ std::string Document::GetDocEntryValue(DocEntry *entry) // as usual for standard multivaluated filels // Elements with Value Multiplicity > 1 // contain a set of integers (not a single one) - else if( vr == "UL" || vr == "SL" ) + else if ( vr == "UL" || vr == "SL" ) { uint32_t newInt32; nbInt = length / 4; for (int i=0; i < nbInt; i++) { - if( i != 0) + if ( i != 0) { s << '\\'; } @@ -1580,9 +1778,15 @@ void Document::SkipDocEntry(DocEntry *entry) */ void Document::SkipToNextDocEntry(DocEntry *currentDocEntry) { - Fp->seekg((long)(currentDocEntry->GetOffset()), std::ios::beg); + int l = currentDocEntry->GetReadLength(); + if ( l == -1 ) // length = 0xffff shouldn't appear here ... + // ... but PMS imagers happen ! + return; + Fp->seekg((long)(currentDocEntry->GetOffset()), std::ios::beg); if (currentDocEntry->GetGroup() != 0xfffe) // for fffe pb + { Fp->seekg( (long)(currentDocEntry->GetReadLength()),std::ios::cur); + } } /** @@ -1614,7 +1818,7 @@ void Document::FixDocEntryFoundLength(DocEntry *entry, // Allthough not recent many such GE corrupted images are still present // on Creatis hard disks. Hence this fix shall remain when such images // are no longer in use (we are talking a few years, here)... - // Note: XMedCom probably uses such a trick since it is able to read + // Note: XMedCon probably uses such a trick since it is able to read // those pesky GE images ... if ( foundLength == 13) { @@ -1644,12 +1848,12 @@ void Document::FixDocEntryFoundLength(DocEntry *entry, //////// We encountered a 'delimiter' element i.e. a tag of the form // "fffe|xxxx" which is just a marker. Delimiters length should not be // taken into account. - else if( gr == 0xfffe ) + else if ( gr == 0xfffe ) { // According to the norm, fffe|0000 shouldn't exist. BUT the Philips // image gdcmData/gdcm-MR-PHILIPS-16-Multi-Seq.dcm happens to // causes extra troubles... - if( entry->GetElement() != 0x0000 ) + if ( entry->GetElement() != 0x0000 ) { foundLength = 0; } @@ -1665,10 +1869,10 @@ void Document::FixDocEntryFoundLength(DocEntry *entry, */ bool Document::IsDocEntryAnInteger(DocEntry *entry) { - uint16_t elem = entry->GetElement(); - uint16_t group = entry->GetGroup(); - const std::string &vr = entry->GetVR(); - uint32_t length = entry->GetLength(); + uint16_t elem = entry->GetElement(); + uint16_t group = entry->GetGroup(); + const std::string &vr = entry->GetVR(); + uint32_t length = entry->GetLength(); // When we have some semantics on the element we just read, and if we // a priori know we are dealing with an integer, then we shall be @@ -1712,12 +1916,7 @@ bool Document::IsDocEntryAnInteger(DocEntry *entry) * true when we hope ours assuptions are OK */ bool Document::CheckSwap() -{ - // The only guaranted way of finding the swap code is to find a - // group tag since we know it's length has to be of four bytes i.e. - // 0x00000004. Finding the swap code in then straigthforward. Trouble - // occurs when we can't find such group... - +{ uint32_t s32; uint16_t s16; @@ -1732,7 +1931,7 @@ bool Document::CheckSwap() Fp->read(deb, 256); char *entCur = deb + 128; - if( memcmp(entCur, "DICM", (size_t)4) == 0 ) + if ( memcmp(entCur, "DICM", (size_t)4) == 0 ) { gdcmWarningMacro( "Looks like DICOM Version3 (preamble + DCM)" ); @@ -1756,11 +1955,11 @@ bool Document::CheckSwap() // even if elem 0002,0010 (Transfer Syntax) tells us the file is // *Implicit* VR (see former 'gdcmData/icone.dcm') - if( memcmp(entCur, "UL", (size_t)2) == 0 || - memcmp(entCur, "OB", (size_t)2) == 0 || - memcmp(entCur, "UI", (size_t)2) == 0 || - memcmp(entCur, "CS", (size_t)2) == 0 ) // CS, to remove later - // when Write DCM *adds* + if ( memcmp(entCur, "UL", (size_t)2) == 0 || + memcmp(entCur, "OB", (size_t)2) == 0 || + memcmp(entCur, "UI", (size_t)2) == 0 || + memcmp(entCur, "CS", (size_t)2) == 0 ) // CS, to remove later + // when Write DCM *adds* // FIXME // Use Document::dicom_vr to test all the possibilities // instead of just checking for UL, OB and UI !? group 0000 @@ -1788,19 +1987,56 @@ bool Document::CheckSwap() // Position the file position indicator at first tag // (i.e. after the file preamble and the "DICM" string). - Fp->seekg(0, std::ios::beg); + + Fp->seekg(0, std::ios::beg); // FIXME : Is it usefull? + Fp->seekg ( 132L, std::ios::beg); return true; - } // End of DicomV3 + } // ------------------------------- End of DicomV3 ---------------- // Alas, this is not a DicomV3 file and whatever happens there is no file // preamble. We can reset the file position indicator to where the data // is (i.e. the beginning of the file). - gdcmWarningMacro( "Not a DICOM Version3 file"); + + gdcmWarningMacro( "Not a Kosher DICOM Version3 file (no preamble)"); + Fp->seekg(0, std::ios::beg); + // Let's check 'No Preamble Dicom File' : + // Should start with group 0x0002 + // and be Explicit Value Representation + + s16 = *((uint16_t *)(deb)); + SwapCode = 0; + switch ( s16 ) + { + case 0x0002 : + SwapCode = 1234; + entCur = deb + 4; + break; + case 0x0200 : + SwapCode = 4321; + entCur = deb + 6; + } + + if ( SwapCode != 0 ) + { + if ( memcmp(entCur, "UL", (size_t)2) == 0 || + memcmp(entCur, "OB", (size_t)2) == 0 || + memcmp(entCur, "UI", (size_t)2) == 0 || + memcmp(entCur, "SH", (size_t)2) == 0 || + memcmp(entCur, "AE", (size_t)2) == 0 || + memcmp(entCur, "OB", (size_t)2) == 0 ) + { + Filetype = ExplicitVR; + gdcmWarningMacro( "Group 0002 : Explicit Value Representation"); + return true; + } + } +// ------------------------------- End of 'No Preamble' DicomV3 ------------- + // Our next best chance would be to be considering a 'clean' ACR/NEMA file. - // By clean we mean that the length of the first tag is written down. + // By clean we mean that the length of the first group is written down. // If this is the case and since the length of the first group HAS to be // four (bytes), then determining the proper swap code is straightforward. @@ -1809,7 +2045,6 @@ bool Document::CheckSwap() // representation of a 32 bits integer. Hence the following dirty // trick : s32 = *((uint32_t *)(entCur)); - switch( s32 ) { case 0x00040000 : @@ -1883,7 +2118,8 @@ bool Document::CheckSwap() */ void Document::SwitchByteSwapCode() { - gdcmWarningMacro( "Switching Byte Swap code from "<< SwapCode); + gdcmWarningMacro( "Switching Byte Swap code from "<< SwapCode + << " at :" <tellg() ); if ( SwapCode == 1234 ) { SwapCode = 4321; @@ -1921,31 +2157,11 @@ void Document::SetMaxSizeLoadEntry(long newSize) } /** - * \brief Header Elements too long will not be printed - * See comments of \ref Document::MAX_SIZE_PRINT_ELEMENT_VALUE - * @param newSize new size - */ -void Document::SetMaxSizePrintEntry(long newSize) -{ - if ( newSize < 0 ) - { - return; - } - if ((uint32_t)newSize >= (uint32_t)0xffffffff ) - { - MaxSizePrintEntry = 0xffffffff; - return; - } - MaxSizePrintEntry = newSize; -} - - -/** - * \brief Read the next tag but WITHOUT loading it's value + * \brief Read the next tag WITHOUT loading it's value * (read the 'Group Number', the 'Element Number', * gets the Dict Entry * gets the VR, gets the length, gets the offset value) - * @return On succes the newly created DocEntry, NULL on failure. + * @return On succes : the newly created DocEntry, NULL on failure. */ DocEntry *Document::ReadNextDocEntry() { @@ -1974,32 +2190,40 @@ DocEntry *Document::ReadNextDocEntry() std::string vr = FindDocEntryVR(); std::string realVR = vr; - if( vr == GDCM_UNKNOWN) + if ( vr == GDCM_UNKNOWN ) { - DictEntry *dictEntry = GetDictEntry(group,elem); - if( dictEntry ) - realVR = dictEntry->GetVR(); + if ( elem == 0x0000 ) // Group Length + realVR = "UL"; // must be UL + else + { + DictEntry *dictEntry = GetDictEntry(group,elem); + if ( dictEntry ) + { + realVR = dictEntry->GetVR(); + } + } } DocEntry *newEntry; - if( Global::GetVR()->IsVROfSequence(realVR) ) + if ( Global::GetVR()->IsVROfSequence(realVR) ) newEntry = NewSeqEntry(group, elem); - else if( Global::GetVR()->IsVROfStringRepresentable(realVR) ) + else if ( Global::GetVR()->IsVROfStringRepresentable(realVR) ) newEntry = NewValEntry(group, elem,vr); else newEntry = NewBinEntry(group, elem,vr); - if( vr == GDCM_UNKNOWN ) + if ( vr == GDCM_UNKNOWN ) { - if( Filetype == ExplicitVR ) + if ( Filetype == ExplicitVR ) { // We thought this was explicit VR, but we end up with an // implicit VR tag. Let's backtrack. if ( newEntry->GetGroup() != 0xfffe ) { std::string msg; - msg = Util::Format("Entry (%04x,%04x) should be Explicit VR\n", - newEntry->GetGroup(), newEntry->GetElement()); + int offset = Fp->tellg(); + msg = Util::Format("Entry (%04x,%04x) at 0x(%x) should be Explicit VR\n", + newEntry->GetGroup(), newEntry->GetElement(), offset ); gdcmWarningMacro( msg.c_str() ); } } @@ -2013,20 +2237,19 @@ DocEntry *Document::ReadNextDocEntry() catch ( FormatError e ) { // Call it quits - //std::cout << e; delete newEntry; return 0; } newEntry->SetOffset(Fp->tellg()); - + return newEntry; } /** * \brief Handle broken private tag from Philips NTSCAN - * where the endianess is being switch to BigEndian for no - * apparent reason + * where the endianess is being switched to BigEndian + * for no apparent reason * @return no return */ void Document::HandleBrokenEndian(uint16_t &group, uint16_t &elem)