You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
notepad-plus-plus/lexilla/lexers/LexX12.cxx

401 lines
12 KiB

// Scintilla Lexer for X12
// @file LexX12.cxx
// Written by Iain Clarke, IMCSoft & Inobiz AB.
// X12 official documentation is behind a paywall, but there's a description of the syntax here:
// http://www.rawlinsecconsulting.com/x12tutorial/x12syn.html
// This code is subject to the same license terms as the rest of the scintilla project:
// The License.txt file describes the conditions under which this software may be distributed.
//
// Header order must match order in scripts/HeaderOrder.txt
#include <cstdlib>
#include <cassert>
#include <cstring>
#include <cctype>
#include <string>
#include <string_view>
#include <vector>
#include <algorithm>
#include "ILexer.h"
#include "Scintilla.h"
#include "SciLexer.h"
#include "LexerModule.h"
#include "DefaultLexer.h"
using namespace Scintilla;
using namespace Lexilla;
class LexerX12 : public DefaultLexer
{
public:
LexerX12();
virtual ~LexerX12() {} // virtual destructor, as we inherit from ILexer
static ILexer5 *Factory() {
return new LexerX12;
}
int SCI_METHOD Version() const override
{
return lvRelease5;
}
void SCI_METHOD Release() override
{
delete this;
}
const char * SCI_METHOD PropertyNames() override
{
return "fold";
}
int SCI_METHOD PropertyType(const char *) override
{
return SC_TYPE_BOOLEAN; // Only one property!
}
const char * SCI_METHOD DescribeProperty(const char *name) override
{
if (!strcmp(name, "fold"))
return "Whether to apply folding to document or not";
return "";
}
Sci_Position SCI_METHOD PropertySet(const char *key, const char *val) override
{
if (!strcmp(key, "fold"))
{
m_bFold = strcmp(val, "0") ? true : false;
return 0;
}
return -1;
}
const char * SCI_METHOD PropertyGet(const char *) override {
return "";
}
const char * SCI_METHOD DescribeWordListSets() override
{
return "";
}
Sci_Position SCI_METHOD WordListSet(int, const char *) override
{
return -1;
}
void SCI_METHOD Lex(Sci_PositionU startPos, Sci_Position length, int initStyle, IDocument *pAccess) override;
void SCI_METHOD Fold(Sci_PositionU startPos, Sci_Position length, int initStyle, IDocument *pAccess) override;
void * SCI_METHOD PrivateCall(int, void *) override
{
return NULL;
}
protected:
struct Terminator
{
int Style = SCE_X12_BAD;
Sci_PositionU pos = 0;
Sci_PositionU length = 0;
int FoldChange = 0;
};
Terminator InitialiseFromISA(IDocument *pAccess);
Sci_PositionU FindPreviousSegmentStart(IDocument *pAccess, Sci_Position startPos) const;
Terminator DetectSegmentHeader(IDocument *pAccess, Sci_PositionU pos) const;
Terminator FindNextTerminator(IDocument *pAccess, Sci_PositionU pos, bool bJustSegmentTerminator = false) const;
bool m_bFold = false;
char m_SeparatorSubElement = 0;
char m_SeparatorElement = 0;
std::string m_SeparatorSegment; // might be multiple characters
std::string m_LineFeed;
};
LexerModule lmX12(SCLEX_X12, LexerX12::Factory, "x12");
///////////////////////////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////
LexerX12::LexerX12() : DefaultLexer("x12", SCLEX_X12)
{
}
void LexerX12::Lex(Sci_PositionU startPos, Sci_Position length, int, IDocument *pAccess)
{
Sci_PositionU posFinish = startPos + length;
Terminator T = InitialiseFromISA(pAccess);
if (T.Style == SCE_X12_BAD)
{
if (T.pos < startPos)
T.pos = startPos; // we may be colouring in batches.
pAccess->StartStyling(startPos);
pAccess->SetStyleFor(T.pos - startPos, SCE_X12_ENVELOPE);
pAccess->SetStyleFor(posFinish - T.pos, SCE_X12_BAD);
return;
}
// Look backwards for a segment start or a document beginning
Sci_PositionU posCurrent = FindPreviousSegmentStart (pAccess, startPos);
// Style buffer, so we're not issuing loads of notifications
pAccess->StartStyling(posCurrent);
while (posCurrent < posFinish)
{
// Look for first element marker, so we can denote segment
T = DetectSegmentHeader(pAccess, posCurrent);
if (T.Style == SCE_X12_BAD)
break;
pAccess->SetStyleFor(T.pos - posCurrent, T.Style);
pAccess->SetStyleFor(T.length, SCE_X12_SEP_ELEMENT);
posCurrent = T.pos + T.length;
while (T.Style != SCE_X12_BAD && T.Style != SCE_X12_SEGMENTEND) // Break on bad or segment ending
{
T = FindNextTerminator(pAccess, posCurrent, false);
if (T.Style == SCE_X12_BAD)
break;
int Style = T.Style;
pAccess->SetStyleFor(T.pos - posCurrent, SCE_X12_DEFAULT);
pAccess->SetStyleFor(T.length, Style);
posCurrent = T.pos + T.length;
}
if (T.Style == SCE_X12_BAD)
break;
}
pAccess->SetStyleFor(posFinish - posCurrent, SCE_X12_BAD);
}
void LexerX12::Fold(Sci_PositionU startPos, Sci_Position length, int, IDocument *pAccess)
{
if (!m_bFold)
return;
// Are we even foldable?
// check for cr,lf,cr+lf.
if (m_LineFeed.empty())
return;
Sci_PositionU posFinish = startPos + length;
// Look backwards for a segment start or a document beginning
startPos = FindPreviousSegmentStart(pAccess, startPos);
Terminator T;
Sci_PositionU currLine = pAccess->LineFromPosition(startPos);
int levelCurrentStyle = SC_FOLDLEVELBASE;
int indentCurrent = 0;
if (currLine > 0)
{
levelCurrentStyle = pAccess->GetLevel(currLine - 1); // bottom 12 bits are level
indentCurrent = levelCurrentStyle & (SC_FOLDLEVELBASE - 1); // indent from previous line
Sci_PositionU posLine = pAccess->LineStart(currLine - 1);
T = DetectSegmentHeader(pAccess, posLine);
indentCurrent += T.FoldChange;
}
while (startPos < posFinish)
{
T = DetectSegmentHeader(pAccess, startPos);
int indentNext = indentCurrent + T.FoldChange;
if (indentNext < 0)
indentNext = 0;
levelCurrentStyle = (T.FoldChange > 0) ? (SC_FOLDLEVELBASE | SC_FOLDLEVELHEADERFLAG) : SC_FOLDLEVELBASE;
currLine = pAccess->LineFromPosition(startPos);
pAccess->SetLevel(currLine, levelCurrentStyle | indentCurrent);
T = FindNextTerminator(pAccess, startPos, true);
if (T.Style == SCE_X12_BAD)
break;
startPos = T.pos + T.length;
indentCurrent = indentNext;
}
}
LexerX12::Terminator LexerX12::InitialiseFromISA(IDocument *pAccess)
{
Sci_Position length = pAccess->Length();
if (length <= 108)
return { SCE_X12_BAD, 0 };
pAccess->GetCharRange(&m_SeparatorElement, 3, 1);
pAccess->GetCharRange(&m_SeparatorSubElement, 104, 1);
Update to Scintilla 5.3.3 and Lexilla 5.2.2 update to https://www.scintilla.org/scintilla533.zip with: 1. Released 8 February 2023. 2. Fix SCI_LINESJOIN bug where carriage returns were incorrectly retained. Bug #2372. 3. Fix SCI_VERTICALCENTRECARET to update the vertical scroll position. 4. When an autocompletion list is shown in response to SCN_CHARADDED, do not process character as fill-up or stop. This avoids closing immediately when a character may both trigger and finish autocompletion. 5. On Cocoa fix character input bug where dotless 'i' and some other extended Latin characters could not be entered. The change also stops SCI_ASSIGNCMDKEY from working with these characters on Cocoa. Bug #2374. 6. On GTK, support IME context. Feature #1476. 7. On GTK on Win32, fix scrolling speed to not be too fast. Bug #2375. 8. On Qt, fix indicator drawing past left of text pane over margin. Bug #2373, Bug #1956. 9. On Qt, allow scrolling with mouse wheel when scroll bar hidden. and https://www.scintilla.org/lexilla522.zip with 1. Released 8 February 2023. 2. C++: Fix keywords that start with non-ASCII. Also affects other lexers. Issue #130. 3. Matlab: Include more prefix and suffix characters in numeric literals. Issue #120. 4. Matlab: More accurate treatment of line ends inside strings. Matlab and Octave are different here. Issue #18. 5. Modula-3: Don't treat identifier suffix that matches keyword as keyword. Issue #129. 6. Modula-3: Fix endless loop in folder. Issue #128. 7. Modula-3: Fix access to lines beyond document end in folder. Issue #131. 8. Python: Don't highlight match and case as keywords in contexts where they probably aren't used as keywords. Pull request #122. 9. X12: Support empty envelopes. Bug #2369. update CMakeLists.txt to latest changes within vcxproj file Close #13082
2 years ago
// Look for GS, as that's the next segment. Anything between 105 and GS/IEA is our segment separator.
Sci_Position posNextSegment;
char bufSegment[3] = { 0 };
for (posNextSegment = 105; posNextSegment < length - 3; posNextSegment++)
{
Update to Scintilla 5.3.3 and Lexilla 5.2.2 update to https://www.scintilla.org/scintilla533.zip with: 1. Released 8 February 2023. 2. Fix SCI_LINESJOIN bug where carriage returns were incorrectly retained. Bug #2372. 3. Fix SCI_VERTICALCENTRECARET to update the vertical scroll position. 4. When an autocompletion list is shown in response to SCN_CHARADDED, do not process character as fill-up or stop. This avoids closing immediately when a character may both trigger and finish autocompletion. 5. On Cocoa fix character input bug where dotless 'i' and some other extended Latin characters could not be entered. The change also stops SCI_ASSIGNCMDKEY from working with these characters on Cocoa. Bug #2374. 6. On GTK, support IME context. Feature #1476. 7. On GTK on Win32, fix scrolling speed to not be too fast. Bug #2375. 8. On Qt, fix indicator drawing past left of text pane over margin. Bug #2373, Bug #1956. 9. On Qt, allow scrolling with mouse wheel when scroll bar hidden. and https://www.scintilla.org/lexilla522.zip with 1. Released 8 February 2023. 2. C++: Fix keywords that start with non-ASCII. Also affects other lexers. Issue #130. 3. Matlab: Include more prefix and suffix characters in numeric literals. Issue #120. 4. Matlab: More accurate treatment of line ends inside strings. Matlab and Octave are different here. Issue #18. 5. Modula-3: Don't treat identifier suffix that matches keyword as keyword. Issue #129. 6. Modula-3: Fix endless loop in folder. Issue #128. 7. Modula-3: Fix access to lines beyond document end in folder. Issue #131. 8. Python: Don't highlight match and case as keywords in contexts where they probably aren't used as keywords. Pull request #122. 9. X12: Support empty envelopes. Bug #2369. update CMakeLists.txt to latest changes within vcxproj file Close #13082
2 years ago
pAccess->GetCharRange(bufSegment, posNextSegment, 3);
if (!memcmp (bufSegment, "GS", 2) || !memcmp(bufSegment, "IEA", 3))
{
Update to Scintilla 5.3.3 and Lexilla 5.2.2 update to https://www.scintilla.org/scintilla533.zip with: 1. Released 8 February 2023. 2. Fix SCI_LINESJOIN bug where carriage returns were incorrectly retained. Bug #2372. 3. Fix SCI_VERTICALCENTRECARET to update the vertical scroll position. 4. When an autocompletion list is shown in response to SCN_CHARADDED, do not process character as fill-up or stop. This avoids closing immediately when a character may both trigger and finish autocompletion. 5. On Cocoa fix character input bug where dotless 'i' and some other extended Latin characters could not be entered. The change also stops SCI_ASSIGNCMDKEY from working with these characters on Cocoa. Bug #2374. 6. On GTK, support IME context. Feature #1476. 7. On GTK on Win32, fix scrolling speed to not be too fast. Bug #2375. 8. On Qt, fix indicator drawing past left of text pane over margin. Bug #2373, Bug #1956. 9. On Qt, allow scrolling with mouse wheel when scroll bar hidden. and https://www.scintilla.org/lexilla522.zip with 1. Released 8 February 2023. 2. C++: Fix keywords that start with non-ASCII. Also affects other lexers. Issue #130. 3. Matlab: Include more prefix and suffix characters in numeric literals. Issue #120. 4. Matlab: More accurate treatment of line ends inside strings. Matlab and Octave are different here. Issue #18. 5. Modula-3: Don't treat identifier suffix that matches keyword as keyword. Issue #129. 6. Modula-3: Fix endless loop in folder. Issue #128. 7. Modula-3: Fix access to lines beyond document end in folder. Issue #131. 8. Python: Don't highlight match and case as keywords in contexts where they probably aren't used as keywords. Pull request #122. 9. X12: Support empty envelopes. Bug #2369. update CMakeLists.txt to latest changes within vcxproj file Close #13082
2 years ago
m_SeparatorSegment.resize(posNextSegment - 105);
pAccess->GetCharRange(&m_SeparatorSegment.at(0), 105, posNextSegment - 105);
// Is some of that CR+LF?
size_t nPos = m_SeparatorSegment.find_last_not_of("\r\n");
m_LineFeed = m_SeparatorSegment.substr(nPos + 1);
m_SeparatorSegment = m_SeparatorSegment.substr(0, nPos + 1);
break;
}
}
if (m_SeparatorSegment.empty() && m_LineFeed.empty())
{
return { SCE_X12_BAD, 105 };
}
// Validate we have an element separator, and it's not silly!
if (m_SeparatorElement == '\0' || m_SeparatorElement == '\n' || m_SeparatorElement == '\r')
return { SCE_X12_BAD, 3 };
// Validate we have an element separator, and it's not silly!
if (m_SeparatorSubElement == '\0' || m_SeparatorSubElement == '\n' || m_SeparatorSubElement == '\r')
return { SCE_X12_BAD, 103 };
if (m_SeparatorElement == m_SeparatorSubElement)
return { SCE_X12_BAD, 104 };
for (auto& c : m_SeparatorSegment)
{
if (m_SeparatorElement == c)
return { SCE_X12_BAD, 105 };
if (m_SeparatorSubElement == c)
return { SCE_X12_BAD, 105 };
}
// Check we have element markers at all the right places! ISA element has fixed entries.
std::vector<Sci_PositionU> ElementMarkers = { 3, 6, 17, 20, 31, 34, 50, 53, 69, 76, 81, 83, 89, 99, 101, 103 };
for (auto i : ElementMarkers)
{
char c;
pAccess->GetCharRange(&c, i, 1);
if (c != m_SeparatorElement)
return { SCE_X12_BAD, i };
}
// Check we have no element markers anywhere else!
for (Sci_PositionU i = 0; i < 105; i++)
{
if (std::find(ElementMarkers.begin(), ElementMarkers.end(), i) != ElementMarkers.end())
continue;
char c;
pAccess->GetCharRange(&c, i, 1);
if (c == m_SeparatorElement)
return { SCE_X12_BAD, i };
}
return { SCE_X12_ENVELOPE };
}
Sci_PositionU LexerX12::FindPreviousSegmentStart(IDocument *pAccess, Sci_Position startPos) const
{
Sci_PositionU length = pAccess->Length();
std::string bufTest = m_SeparatorSegment + m_LineFeed; // quick way of making the lengths the same
std::string bufCompare = bufTest;
for (; startPos > 0; startPos--)
{
if (startPos + bufTest.size() > length)
continue;
pAccess->GetCharRange(&bufTest.at(0), startPos, bufTest.size());
if (bufTest == bufCompare)
{
return startPos + bufTest.size();
}
}
// We didn't find a ', so just go with the beginning
return 0;
}
LexerX12::Terminator LexerX12::DetectSegmentHeader(IDocument *pAccess, Sci_PositionU pos) const
{
Sci_PositionU Length = pAccess->Length();
Length -= pos;
char c, Buf[4] = { 0 }; // max 3 + separator
for (Sci_PositionU posOffset = 0; posOffset < std::size(Buf) && posOffset < Length; posOffset++)
{
pAccess->GetCharRange(&c, pos + posOffset, 1);
if (c != m_SeparatorElement)
{
Buf[posOffset] = c;
continue;
}
// check for special segments, involved in folding start/stop.
if (memcmp(Buf, "ISA", 3) == 0)
return { SCE_X12_ENVELOPE, pos + posOffset, 1, +1 };
if (memcmp(Buf, "IEA", 3) == 0)
return { SCE_X12_ENVELOPE, pos + posOffset, 1, -1 };
if (memcmp(Buf, "GS", 2) == 0)
return { SCE_X12_FUNCTIONGROUP, pos + posOffset, 1, +1 };
if (memcmp(Buf, "GE", 2) == 0)
return { SCE_X12_FUNCTIONGROUP, pos + posOffset, 1, -1 };
if (memcmp(Buf, "ST", 2) == 0)
return { SCE_X12_TRANSACTIONSET, pos + posOffset, 1, +1 };
if (memcmp(Buf, "SE", 2) == 0)
return { SCE_X12_TRANSACTIONSET, pos + posOffset, 1, -1 };
return { SCE_X12_SEGMENTHEADER, pos + posOffset, 1, 0 };
}
return { SCE_X12_BAD, pos, 0, 0 };
}
LexerX12::Terminator LexerX12::FindNextTerminator(IDocument *pAccess, Sci_PositionU pos, bool bJustSegmentTerminator) const
{
char c;
Sci_PositionU length = pAccess->Length();
std::string bufTestSegment = m_SeparatorSegment; // quick way of making the lengths the same
std::string bufTestLineFeed = m_LineFeed; // quick way of making the lengths the same
while (pos < (Sci_PositionU)length)
{
pAccess->GetCharRange(&c, pos, 1);
if (pos + m_SeparatorSegment.size() > length)
bufTestSegment.clear(); // going up - so once we can't get this, we're done with the buffer.
else if (!bufTestSegment.empty())
pAccess->GetCharRange(&bufTestSegment.at(0), pos, bufTestSegment.size());
if (pos + m_LineFeed.size() > length)
bufTestLineFeed.clear(); // going up - so once we can't get this, we're done with the buffer.
else if (!bufTestLineFeed.empty())
pAccess->GetCharRange(&bufTestLineFeed.at(0), pos, bufTestLineFeed.size());
if (!bJustSegmentTerminator && c == m_SeparatorElement)
return { SCE_X12_SEP_ELEMENT, pos, 1 };
else if (!bJustSegmentTerminator && c == m_SeparatorSubElement)
return { SCE_X12_SEP_SUBELEMENT, pos, 1 };
else if (!m_SeparatorSegment.empty() && bufTestSegment == m_SeparatorSegment)
{
if (m_LineFeed.empty())
return { SCE_X12_SEGMENTEND, pos, m_SeparatorSegment.size() };
// is this the end?
if (pos + m_SeparatorSegment.size() == length)
return { SCE_X12_SEGMENTEND, pos, m_SeparatorSegment.size() };
// Check if we're followed by a linefeed.
if (pos + m_SeparatorSegment.size() + m_LineFeed.size() > length)
return { SCE_X12_BAD, pos };
bufTestSegment = m_LineFeed;
pAccess->GetCharRange(&bufTestSegment.at(0), pos + m_SeparatorSegment.size(), bufTestSegment.size());
if (bufTestSegment == m_LineFeed)
return { SCE_X12_SEGMENTEND, pos, m_SeparatorSegment.size() + m_LineFeed.size() };
break;
}
else if (m_SeparatorSegment.empty() && bufTestLineFeed == m_LineFeed)
{
return { SCE_X12_SEGMENTEND, pos, m_LineFeed.size() };
}
pos++;
}
return { SCE_X12_BAD, pos };
}