Documentation markup extraction (#1724)

* #include an absolute path didn't work - because paths were taken to always be relative.

* WIP extracting source documentation.

* WIP doc extraction.

* More stuff around doc markup extraction.

* More WIP around doc extraction.

* Fix some indexing issues.

* Initial doc extraction working.

* Renaming of types in markup extraction process.

* Extracting markup content.
Removing indenting.
Other fixes and improvements around document tools.

* WIP support for documentation system.

* Remove some commented out sections.

* Remove some comments that no longer apply.

* Improvements around SourceFile - such that more granularity around line ops.
Made some functionality explicitly work without source.
Improved Doc types nameing.

14 files changed, 1316 insertions, 17 deletions

diff --git a/build/visual-studio/slang/slang.vcxproj b/build/visual-studio/slang/slang.vcxproj
index 6a8798d59..7ddc3a2e0 100644
--- a/build/visual-studio/slang/slang.vcxproj
+++ b/build/visual-studio/slang/slang.vcxproj
@@ -206,6 +206,7 @@
     <ClInclude Include="..\..\..\source\slang\slang-compiler.h" />
     <ClInclude Include="..\..\..\source\slang\slang-diagnostic-defs.h" />
     <ClInclude Include="..\..\..\source\slang\slang-diagnostics.h" />
+    <ClInclude Include="..\..\..\source\slang\slang-doc.h" />
     <ClInclude Include="..\..\..\source\slang\slang-emit-c-like.h" />
     <ClInclude Include="..\..\..\source\slang\slang-emit-cpp.h" />
     <ClInclude Include="..\..\..\source\slang\slang-emit-cuda.h" />
@@ -334,6 +335,7 @@
     <ClCompile Include="..\..\..\source\slang\slang-check.cpp" />
     <ClCompile Include="..\..\..\source\slang\slang-compiler.cpp" />
     <ClCompile Include="..\..\..\source\slang\slang-diagnostics.cpp" />
+    <ClCompile Include="..\..\..\source\slang\slang-doc.cpp" />
     <ClCompile Include="..\..\..\source\slang\slang-dxc-support.cpp" />
     <ClCompile Include="..\..\..\source\slang\slang-emit-c-like.cpp" />
     <ClCompile Include="..\..\..\source\slang\slang-emit-cpp.cpp" />
diff --git a/build/visual-studio/slang/slang.vcxproj.filters b/build/visual-studio/slang/slang.vcxproj.filters
index a6fa6ea0c..f8bc02d51 100644
--- a/build/visual-studio/slang/slang.vcxproj.filters
+++ b/build/visual-studio/slang/slang.vcxproj.filters
@@ -69,6 +69,9 @@
     <ClInclude Include="..\..\..\source\slang\slang-diagnostics.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\source\slang\slang-doc.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
     <ClInclude Include="..\..\..\source\slang\slang-emit-c-like.h">
       <Filter>Header Files</Filter>
     </ClInclude>
@@ -449,6 +452,9 @@
     <ClCompile Include="..\..\..\source\slang\slang-diagnostics.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\source\slang\slang-doc.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="..\..\..\source\slang\slang-dxc-support.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
diff --git a/source/core/slang-string-util.cpp b/source/core/slang-string-util.cpp
index cac7c7cfa..6ce75f3f0 100644
--- a/source/core/slang-string-util.cpp
+++ b/source/core/slang-string-util.cpp
@@ -360,6 +360,28 @@ ComPtr<ISlangBlob> StringUtil::createStringBlob(const String& string)
     }
 }
 
+/* static */UnownedStringSlice StringUtil::trimEndOfLine(const UnownedStringSlice& line)
+{
+    // Strip CR/LF from end of line if present
+
+    const char* begin = line.begin();
+    const char* end = line.end();
+
+    if (end > begin)
+    {
+        const char c = end[-1];
+        // If last char is CR/LF move back a char
+        if (c == '\n' || c == '\r')
+        {
+            --end;
+            // If next char is a match for the CR/LF pair move back an extra char.
+            end -= Index((end > begin) && (c ^ end[-1]) == ('\r' ^ '\n'));
+        }
+    }
+
+    return line.head(Index(end - begin));
+}
+
 /* static */bool StringUtil::areLinesEqual(const UnownedStringSlice& inA, const UnownedStringSlice& inB)
 {
     UnownedStringSlice a(inA), b(inB), lineA, lineB;
diff --git a/source/core/slang-string-util.h b/source/core/slang-string-util.h
index dade8a61d..4672fa1d0 100644
--- a/source/core/slang-string-util.h
+++ b/source/core/slang-string-util.h
@@ -88,6 +88,9 @@ struct StringUtil
         /// Given text, splits into lines stored in outLines. NOTE! That lines is only valid as long as textIn remains valid
     static void calcLines(const UnownedStringSlice& textIn, List<UnownedStringSlice>& lines);
 
+        /// Given a line that may contain cr/lf, returns the the a slice that doesn't have trailing cr/lf
+    static UnownedStringSlice trimEndOfLine(const UnownedStringSlice& slice);
+
         /// Equal if the lines are equal (in effect a way to ignore differences in line breaks)
     static bool areLinesEqual(const UnownedStringSlice& a, const UnownedStringSlice& b);
 
diff --git a/source/slang/slang-compiler.h b/source/slang/slang-compiler.h
index 90b1e30f3..5eedb4db6 100755
--- a/source/slang/slang-compiler.h
+++ b/source/slang/slang-compiler.h
@@ -1522,6 +1522,7 @@ namespace Slang
         bool shouldValidateIR = false;
 
         bool shouldDumpAST = false;
+        bool shouldDocument = false;
 
             /// If true will after lexical analysis output the hierarchy of includes to stdout
         bool outputIncludes = false;
diff --git a/source/slang/slang-doc.cpp b/source/slang/slang-doc.cpp
new file mode 100644
index 000000000..fa3e11030
--- /dev/null
+++ b/source/slang/slang-doc.cpp
@@ -0,0 +1,974 @@
+// slang-doc.cpp
+#include "slang-doc.h"
+
+#include "../core/slang-string-util.h"
+
+namespace Slang {
+
+/* TODO(JS):
+
+* If Decls hand SourceRange, then we could use the range to simplify getting the Post markup, as will be trivial to get to the 'end'
+* Need to handle preceeding * in some markup styles
+* If we want to be able to disable markup we need a mechanism to do this. Probably define source ranges.
+
+* Need a way to take the extracted markup and produce suitable markdown
+** This will need to display the decoration appropriately
+*/
+
+/* Extracts 'markup' from comments in Slang source core. The comments are extracted and associated in declarations. The association
+is held in DocMarkup type. The comment style follows the doxygen style */
+class DocMarkupExtractor 
+{
+public:
+
+    typedef uint32_t MarkupFlags;
+    struct MarkupFlag
+    {
+        enum Enum : MarkupFlags
+        {
+            Before          = 0x1,
+            After           = 0x2,
+            IsMultiToken    = 0x4,          ///< Can use more than one token
+            IsBlock         = 0x8,          ///< 
+        };
+    };
+
+    enum class MarkupType
+    {
+        None,
+        BlockBefore,        /// /**  */ or /*!  */.
+        BlockAfter,         /// /*!< */ or /**< */
+
+        LineBangBefore,      /// //! Can be multiple lines
+        LineSlashBefore,     /// /// Can be multiple lines
+
+        LineBangAfter,       /// //!< Can be multiple lines
+        LineSlashAfter,      /// ///< Can be multiple lines
+    };
+
+    struct IndexRange
+    {
+        SLANG_FORCE_INLINE Index getCount() const { return end - start; }
+
+        Index start;
+        Index end;
+    };
+
+    enum class Location
+    {
+        None,                           ///< No defined location
+        Before,
+        AfterParam,                     ///< Can have trailing , or )
+        AfterSemicolon,                 ///< Can have a trailing ;
+    };
+
+    struct FoundMarkup
+    {
+        void reset()
+        {
+            location = Location::None;
+            type = MarkupType::None;
+            range = IndexRange { 0, 0 };
+        }
+
+        Location location = Location::None;
+        MarkupType type = MarkupType::None;
+        IndexRange range;
+    };
+
+    struct FindInfo
+    {
+
+        SourceView* sourceView;         ///< The source view the tokens were generated from
+        TokenList* tokenList;           ///< The token list
+        Index declTokenIndex;           ///< The token index location (where searches start from)
+        Index declLineIndex;            ///< The line number for the decl
+    };
+
+    SlangResult extract(DocMarkup* doc, ModuleDecl* moduleDecl, SourceManager* sourceManager, DiagnosticSink* sink);
+
+    static MarkupFlags getFlags(MarkupType type);
+    static MarkupType findMarkupType(const Token& tok);
+    static UnownedStringSlice removeStart(MarkupType type, const UnownedStringSlice& comment);
+
+protected:
+        /// returns SLANG_E_NOT_FOUND if not found, SLANG_OK on success else an error
+    SlangResult _findMarkup(const FindInfo& info, Location location, FoundMarkup& out);
+
+        /// Locations are processed in order, and the first successful used. If found in another location will issue a warning.
+        /// returns SLANG_E_NOT_FOUND if not found, SLANG_OK on success else an error
+    SlangResult _findFirstMarkup(const FindInfo& info, const Location* locs, Index locCount, FoundMarkup& out, Index& outIndex);
+
+    SlangResult _findMarkup(const FindInfo& info, const Location* locs, Index locCount, FoundMarkup& out);
+
+        /// Given the decl, the token stream, and the decls tokenIndex, try to find some associated markup
+    SlangResult _findMarkup(const FindInfo& info, Decl* decl, FoundMarkup& out);
+
+        /// Given a found markup location extracts the contents of the tokens into out
+    SlangResult _extractMarkup(const FindInfo& info, const FoundMarkup& foundMarkup, StringBuilder& out);
+
+        /// Given a location, try to find the first token index that could potentially be markup
+        /// Will return -1 if not found
+    Index _findStartIndex(const FindInfo& info, Location location);
+
+        /// True if the tok is 'on' lineIndex. Interpretation of 'on' depends on the markup type.
+    static bool _isTokenOnLineIndex(SourceView* sourceView, MarkupType type, const Token& tok, Index lineIndex);
+
+    void _addDecl(Decl* decl);
+    void _addDeclRec(Decl* decl);
+    void _findDecls(ModuleDecl* moduleDecl);
+    
+    List<Decl*> m_decls;
+
+    DocMarkup* m_doc;
+    ModuleDecl* m_moduleDecl;
+    SourceManager* m_sourceManager;
+    DiagnosticSink* m_sink;
+};
+
+/* static */UnownedStringSlice DocMarkupExtractor::removeStart(MarkupType type, const UnownedStringSlice& comment)
+{
+    switch (type)
+    {
+        case MarkupType::BlockBefore:
+        {
+            if (comment.startsWith(UnownedStringSlice::fromLiteral("/**")) ||
+                comment.startsWith(UnownedStringSlice::fromLiteral("/*!")))
+            {
+                /// /**  */ or /*!  */.
+                return comment.tail(3);
+            }
+            return comment;
+        }   
+        case MarkupType::BlockAfter:
+        {
+            
+            if (comment.startsWith(UnownedStringSlice::fromLiteral("/**<")) ||
+                comment.startsWith(UnownedStringSlice::fromLiteral("/*!<")))
+            {
+                /// /*!< */ or /**< */
+                return comment.tail(4);
+            }
+            return comment;
+        }
+
+        case MarkupType::LineBangBefore:
+        {
+            return comment.startsWith(UnownedStringSlice::fromLiteral("//!")) ? comment.tail(3) : comment;
+        }
+        case MarkupType::LineSlashBefore:
+        {
+            return comment.startsWith(UnownedStringSlice::fromLiteral("///")) ? comment.tail(3) : comment;
+        }
+
+        case MarkupType::LineBangAfter:
+        {
+            /// //!< Can be multiple lines
+            return comment.startsWith(UnownedStringSlice::fromLiteral("//!<")) ? comment.tail(4) : comment;
+        }
+        case MarkupType::LineSlashAfter:
+        {
+            return comment.startsWith(UnownedStringSlice::fromLiteral("///<")) ? comment.tail(4) : comment;
+        }
+        default: break;
+    }
+    return comment;
+}
+
+void DocMarkupExtractor::_addDecl(Decl* decl)
+{
+    if (!decl->loc.isValid())
+    {
+        return;
+    }
+    m_decls.add(decl);
+}
+
+void DocMarkupExtractor::_addDeclRec(Decl* decl)
+{
+    // Just add.
+    // There may be things we don't want to add, but just add them all of now
+    _addDecl(decl);
+
+#if 0
+    if (CallableDecl* callableDecl = as<CallableDecl>(decl))
+    {
+        // For callables (like functions),
+
+        m_decls.add(callableDecl);
+    }
+    else
+#endif
+    if (ContainerDecl* containerDecl = as<ContainerDecl>(decl))
+    {
+        // Add the container - which could be a class, struct, enum, namespace, extension, generic etc.
+        // Now add what the container contains
+        for (Decl* childDecl : containerDecl->members)
+        {
+            _addDeclRec(childDecl);
+        }
+    }
+}
+
+void DocMarkupExtractor::_findDecls(ModuleDecl* moduleDecl)
+{
+    for (Decl* decl : moduleDecl->members)
+    {
+        _addDeclRec(decl);
+    }
+}
+
+static Index _findTokenIndex(SourceLoc loc, const Token* toks, Index numToks)
+{
+    // Use a binary search to find the token
+    Index lo = 0;
+    Index hi = numToks;
+
+    while (lo + 1 < hi)
+    {
+        const Index mid = (hi + lo) >> 1;
+        const Token& midToken = toks[mid];
+
+        if (midToken.loc == loc)
+        {
+            return mid;
+        }
+
+        if (midToken.loc.getRaw() <= loc.getRaw())
+        {
+            lo = mid;
+        }
+        else
+        {
+            hi = mid;
+        }
+    }
+
+    // Not found
+    return -1;
+}
+
+/* static */DocMarkupExtractor::MarkupFlags DocMarkupExtractor::getFlags(MarkupType type)
+{
+    switch (type)
+    {
+        default:
+        case MarkupType::None:              return 0;
+        case MarkupType::BlockBefore:      return MarkupFlag::Before | MarkupFlag::IsBlock; 
+        case MarkupType::BlockAfter:       return MarkupFlag::After  | MarkupFlag::IsBlock;
+
+        case MarkupType::LineBangBefore:     return MarkupFlag::Before | MarkupFlag::IsMultiToken; 
+        case MarkupType::LineSlashBefore:    return MarkupFlag::Before | MarkupFlag::IsMultiToken; 
+
+        case MarkupType::LineBangAfter:      return MarkupFlag::After | MarkupFlag::IsMultiToken; 
+        case MarkupType::LineSlashAfter:     return MarkupFlag::After | MarkupFlag::IsMultiToken;
+    }
+}
+
+/* static */DocMarkupExtractor::MarkupType DocMarkupExtractor::findMarkupType(const Token& tok)
+{
+    switch (tok.type)
+    {
+        case TokenType::BlockComment:
+        {
+            UnownedStringSlice slice = tok.getContent();
+            if (slice.getLength() >= 3 && (slice[2] == '!' || slice[2] == '*'))
+            {
+                return (slice.getLength() >= 4 && slice[3] == '<') ? MarkupType::BlockAfter : MarkupType::BlockBefore;
+            }
+            break;
+        }
+        case TokenType::LineComment:
+        {
+            UnownedStringSlice slice = tok.getContent();
+            if (slice.getLength() >= 3)
+            {
+                if (slice[2] == '!')
+                {
+                    return (slice.getLength() >= 4 && slice[3] == '<') ? MarkupType::LineBangAfter : MarkupType::LineBangBefore;
+                }
+                else if (slice[2] == '/')
+                {
+                    return (slice.getLength() >= 4 && slice[3] == '<') ? MarkupType::LineSlashAfter : MarkupType::LineSlashBefore;
+                }
+            }
+            break;
+        }
+        default: break;
+    }
+    return MarkupType::None;
+}
+
+static Index _calcWhitespaceIndent(const UnownedStringSlice& line)
+{
+    // TODO(JS): For now we ignore tabs and just work out indentation based on spaces/assume ASCII
+    Index indent = 0;
+    const Index count = line.getLength();
+    for (; indent < count && line[indent] == ' '; indent++);
+    return indent;
+}
+
+static Index _calcIndent(const UnownedStringSlice& line)
+{
+    // TODO(JS): For now we just assume no tabs, and that every char is ASCII
+    return line.getLength();
+}
+
+static void _appendUnindenttedLine(const UnownedStringSlice& line, Index maxIndent, StringBuilder& out)
+{
+    Index indent = _calcWhitespaceIndent(line);
+
+    // We want to remove indenting remove no more than maxIndent
+    if (maxIndent >= 0)
+    {
+        indent = (indent > maxIndent) ? maxIndent : indent;
+    }
+
+    // Remove the indenting, and append to out
+    out.append(line.tail(indent));
+}
+
+SlangResult DocMarkupExtractor::_extractMarkup(const FindInfo& info, const FoundMarkup& foundMarkup, StringBuilder& out)
+{
+    SourceView* sourceView = info.sourceView;
+    SourceFile* sourceFile = sourceView->getSourceFile();
+
+    // Here we want to produce the text that is implied by the markup tokens.
+    // We want to removing surrounding markup, and to also keep appropriate indentation
+    
+    switch (foundMarkup.type)
+    {
+        case MarkupType::BlockBefore:
+        case MarkupType::BlockAfter:
+        {
+            // We should only have a single line
+            SLANG_ASSERT(foundMarkup.range.getCount() == 1);
+
+            const auto& tok = info.tokenList->m_tokens[foundMarkup.range.start];
+            uint32_t offset = sourceView->getRange().getOffset(tok.loc);
+
+            const UnownedStringSlice startLine = sourceFile->getLineContainingOffset(offset);
+
+            UnownedStringSlice content = tok.getContent();
+
+            // Split into lines
+            List<UnownedStringSlice> lines;
+
+            StringUtil::calcLines(content, lines);
+
+            Index maxIndent = -1;
+
+            StringBuilder unindentedLine;
+
+            const Index linesCount = lines.getCount();
+            for (Index i = 0; i < linesCount; ++i)
+            {
+                UnownedStringSlice line = lines[i];
+                unindentedLine.Clear();
+
+                if (i == 0)
+                {
+                    if (startLine.isMemoryContained(line.begin()))
+                    {
+                        // For now we'll ignore tabs, and that the indent amount is, the amount of *byte*
+                        // NOTE! This is only appropriate for ASCII without tabs.
+                        maxIndent = _calcIndent(UnownedStringSlice(startLine.begin(), line.begin()));
+
+                        // Let's strip the start stuff
+                        line = removeStart(foundMarkup.type, line);
+                    }
+                }
+
+                if (i == linesCount - 1)
+                {
+                    SLANG_ASSERT(line.tail(line.getLength() - 2) == UnownedStringSlice::fromLiteral("*/"));
+                    // Remove the */ at the end of the line
+                    line = line.head(line.getLength() - 2);
+                }
+
+                if (i > 0)
+                {
+                    _appendUnindenttedLine(line, maxIndent, unindentedLine);
+                }
+                else
+                {
+                    unindentedLine.append(line);
+                }
+                
+                // If the first or last line are all white space, just ignore them
+                if ((i == linesCount - 1 || i == 0) && unindentedLine.getUnownedSlice().trim().getLength() == 0)
+                {
+                    continue;
+                }
+
+                out.append(unindentedLine);
+                out.appendChar('\n');
+            }
+
+            break;
+        }
+        case MarkupType::LineBangBefore:      
+        case MarkupType::LineSlashBefore:
+        case MarkupType::LineBangAfter:
+        case MarkupType::LineSlashAfter:
+        {
+            // Holds the lines extracted, they may have some white space indenting (like the space at the start of //) 
+            List<UnownedStringSlice> lines;
+
+            const auto& range = foundMarkup.range;
+            for (Index i = range.start; i < range.end; ++ i)
+            {
+                const auto& tok = info.tokenList->m_tokens[i];
+                UnownedStringSlice line = tok.getContent();
+                line = removeStart(foundMarkup.type, line);
+
+                // If the first or last line are all white space, just ignore them
+                if ((i == range.start || i == range.end - 1) && line.trim().getLength() == 0)
+                {
+                    continue;
+                }
+                lines.add(line);
+            }
+
+            if (lines.getCount() == 0)
+            {
+                // If there are no lines, theres no content
+                return SLANG_OK;
+            }
+
+            Index minIndent = 0x7fffffff;
+            for (const auto& line : lines)
+            {
+                const Index indent = _calcWhitespaceIndent(line);
+                minIndent = (indent < minIndent) ? indent : minIndent;
+            }
+
+            for (const auto& line : lines)
+            {
+                _appendUnindenttedLine(line, minIndent, out);
+                out.appendChar('\n');
+            }
+
+            break;
+        }
+        default:    return SLANG_FAIL;
+    }
+
+    return SLANG_OK;
+}
+
+Index DocMarkupExtractor::_findStartIndex(const FindInfo& info, Location location)
+{
+    Index openParensCount = 0;
+    Index openBracketCount = 0;
+
+    const TokenList& toks = *info.tokenList;
+    const Index tokIndex = info.declTokenIndex;
+
+    Index direction = (location == Location::Before) ? -1 : 1;
+
+    const Index count = toks.m_tokens.getCount();
+    for (Index i = tokIndex; i >= 0 && i < count; i += direction)
+    {
+        const Token& tok = toks.m_tokens[i];
+
+        switch (tok.type)
+        {
+            case TokenType::LParent:
+            {
+                ++openParensCount;
+                break;
+            }
+            case TokenType::RBracket:
+            {
+                openBracketCount += Index(location == Location::Before);
+                break;
+            }
+            case TokenType::LBracket:
+            {
+                openBracketCount -= Index(location == Location::Before);
+                break;
+            }
+            case TokenType::RParent:
+            {
+                if (openParensCount == 0 &&
+                    location == Location::AfterParam)
+                {
+                    return i + 1;
+                }
+
+                --openParensCount;
+                if (openParensCount < 0)
+                {
+                    // Not found - or weird parens at least 
+                    return -1;
+                }
+                break;
+            }
+            case TokenType::Comma:
+            {
+                if (location == Location::AfterParam)
+                {
+                    return i + 1;
+                }
+                break;
+            }
+            case TokenType::RBrace:
+            {
+                // If we haven't hit a candidate yet before hitting } it's not going to work
+                if (location == Location::Before)
+                {
+                    return -1;
+                }
+                break;
+            }
+            case TokenType::Semicolon:
+            {
+                // If we haven't hit a candidate yet it's not going to work
+                if (location == Location::Before)
+                {
+                    return -1;
+                }
+
+                if (openParensCount == 0 && location == Location::AfterSemicolon)
+                {
+                    return i + 1;
+                }
+                break;
+            }
+            case TokenType::LineComment:
+            case TokenType::BlockComment:
+            {
+                // We hit a comment this could be the markup
+                if (location == Location::Before && openParensCount == 0 && openBracketCount == 0)
+                {
+                    return i;
+                }
+                break;
+            }
+            default: break;
+        }
+    }
+
+    return -1;
+}
+
+/* static */bool DocMarkupExtractor::_isTokenOnLineIndex(SourceView* sourceView, MarkupType type, const Token& tok, Index lineIndex)
+{
+    SourceFile* sourceFile = sourceView->getSourceFile();
+    const int offset = sourceView->getRange().getOffset(tok.loc);
+
+    auto const flags = getFlags(type);
+
+    if (flags & MarkupFlag::IsBlock)
+    {
+        // Either the start or the end of the block have to be on the specified line
+        return sourceFile->isOffsetOnLine(offset, lineIndex) || sourceFile->isOffsetOnLine(offset + tok.charsCount, lineIndex);
+    }
+    else
+    {
+        // Has to be exactly on the specified line
+        return sourceFile->isOffsetOnLine(offset, lineIndex);
+    }
+}
+
+
+SlangResult DocMarkupExtractor::_findMarkup(const FindInfo& info, Location location, FoundMarkup& out)
+{
+    out.reset();
+
+    const auto& toks = info.tokenList->m_tokens;
+    const Index tokIndex = info.declTokenIndex;
+
+    // The starting token index
+    Index startIndex = _findStartIndex(info, location);
+    if (startIndex <= 0)
+    {
+        return SLANG_E_NOT_FOUND;
+    }
+
+    SourceView* sourceView = info.sourceView;
+    SourceFile* sourceFile = sourceView->getSourceFile();
+
+    // Let's lookup the line index where this occurred
+    const int startOffset = sourceView->getRange().getOffset(toks[startIndex - 1].loc);
+
+    // The line index that the markoff starts from 
+    Index lineIndex = sourceFile->calcLineIndexFromOffset(startOffset);
+    if (lineIndex < 0)
+    {
+        return SLANG_E_NOT_FOUND;
+    }
+
+    const Index searchDirection = (location == Location::Before) ? -1 : 1;
+    
+    // Get the type and flags
+    const MarkupType type = findMarkupType(toks[startIndex]);
+    const MarkupFlags flags = getFlags(type);
+
+    const MarkupFlag::Enum requiredFlag = (location == Location::Before) ? MarkupFlag::Before : MarkupFlag::After;
+    if ((flags & requiredFlag) == 0)
+    {
+        return SLANG_E_NOT_FOUND;
+    }
+
+#if 0
+    // The token still isn't accepted, unless it's on the expected line
+    if (_isTokenOnLineIndex(info.sourceView, type, toks[startIndex], expectedLineIndex))
+    {
+        return SLANG_E_NOT_FOUND;
+    }
+#endif
+
+    Index endIndex = startIndex;
+
+    // If it's multiline, so look for the end index
+    if (flags & MarkupFlag::IsMultiToken)
+    {
+        Index expectedLineIndex = lineIndex;
+
+        // TODO(JS):
+        // We should probably do the work here to confirm  indentation - but that
+        // requires knowing something about tabs, so for now we leave.
+
+        while (true)
+        {
+            endIndex += searchDirection;
+            expectedLineIndex += searchDirection;
+
+            if (endIndex < 0 || endIndex >= toks.getCount())
+            {
+                break;
+            }
+
+            // Do we find a token of the right type?
+            if (findMarkupType(toks[endIndex]) != type)
+            {
+                break;
+            }
+
+            // Is it on the right line?
+            if (_isTokenOnLineIndex(info.sourceView, type, toks[startIndex], expectedLineIndex))
+            {
+                break;
+            }
+        }
+
+        // Fix the end index (it's the last one that worked)
+        endIndex -= searchDirection;
+    }
+
+    // Put start < end order
+    if (endIndex < startIndex)
+    {
+        Swap(endIndex, startIndex);
+    }
+    // The range excludes end so increase
+    endIndex++;
+
+    // Okay we've found the markup
+    out.type = type;
+    out.location = location;
+    out.range = IndexRange{ startIndex, endIndex };
+
+    SLANG_ASSERT(out.range.getCount() > 0);
+
+    return SLANG_OK;
+}
+
+SlangResult DocMarkupExtractor::_findFirstMarkup(const FindInfo& info, const Location* locs, Index locCount, FoundMarkup& out, Index& outIndex)
+{
+    Index i = 0;
+    for (; i < locCount; ++i)
+    {
+        SlangResult res = _findMarkup(info, locs[i], out);
+        if (SLANG_SUCCEEDED(res) || (SLANG_FAILED(res) && res != SLANG_E_NOT_FOUND))
+        {
+            outIndex = i;
+            return res;
+        }
+    }
+    return SLANG_E_NOT_FOUND;
+}
+
+SlangResult DocMarkupExtractor::_findMarkup(const FindInfo& info, const Location* locs, Index locCount, FoundMarkup& out)
+{
+    Index foundIndex;
+    SLANG_RETURN_ON_FAIL(_findFirstMarkup(info, locs, locCount, out, foundIndex));
+
+    // Lets see if the remaining ones match
+    {
+        FoundMarkup otherMarkup;
+        for (Index i = foundIndex + 1; i < locCount; ++i)
+        {
+            SlangResult res = _findMarkup(info, locs[i], otherMarkup);
+            if (SLANG_SUCCEEDED(res))
+            {
+                // TODO(JS): Warning found markup in another location
+            }
+        }
+    }
+
+    return SLANG_OK;
+}
+
+SlangResult DocMarkupExtractor::_findMarkup(const FindInfo& info, Decl* decl, FoundMarkup& out)
+{
+    if (auto paramDecl = as<ParamDecl>(decl))
+    {
+        Location locs[] = { Location::Before, Location::AfterParam };
+        return _findMarkup(info, locs, SLANG_COUNT_OF(locs), out);
+    }
+    else if (auto callableDecl = as<CallableDecl>(decl))
+    {
+        // We allow it defined before
+        return _findMarkup(info, Location::Before, out);
+    }
+    else if (as<VarDecl>(decl) || as<TypeDefDecl>(decl) || as<AssocTypeDecl>(decl))
+    {
+        Location locs[] = { Location::Before, Location::AfterSemicolon };
+        return _findMarkup(info, locs, SLANG_COUNT_OF(locs), out);
+    }
+    else
+    {
+        // We'll only allow before
+        return _findMarkup(info, Location::Before, out);
+    }
+}
+
+SlangResult DocMarkupExtractor::extract(DocMarkup* doc, ModuleDecl* moduleDecl, SourceManager* sourceManager, DiagnosticSink* sink)
+{
+    m_doc = doc;
+    m_moduleDecl = moduleDecl;
+    m_sourceManager = sourceManager;
+    m_sink = sink;
+
+    _findDecls(moduleDecl);
+
+    struct Entry
+    {
+        typedef Entry ThisType;
+
+        bool operator<(const ThisType& rhs) const { return locOrOffset < rhs.locOrOffset; }
+
+        Index viewIndex;                    ///< The view/file index this loc is found in
+        SourceLoc::RawValue locOrOffset;    ///< Can be a loc or an offset into the file
+
+        Decl* decl;                         ///< The decl
+    };
+
+    List<Entry> entries;
+
+    {
+        const Index count = m_decls.getCount();
+        entries.setCount(count);
+
+        for (Index i = 0; i < count; ++i)
+        {
+            Entry& entry = entries[i];
+            auto decl = m_decls[i];
+            entry.decl = decl;
+            entry.viewIndex = -1;            //< We don't know what file/view it's in
+            entry.locOrOffset = decl->loc.getRaw();
+        }
+    }
+
+    // We hold one view per *SourceFile*
+    List<SourceView*> views;
+
+    // Sort them into loc order
+    entries.sort([](Entry& a, Entry& b) { return a.locOrOffset < b.locOrOffset; });
+
+    {
+        SourceView* sourceView = nullptr;
+        Index viewIndex = -1;
+
+        for (auto& entry : entries)
+        {
+            const SourceLoc loc = SourceLoc::fromRaw(entry.locOrOffset);
+
+            if (sourceView == nullptr || !sourceView->getRange().contains(loc))
+            {
+                // Find the new view
+                sourceView = m_sourceManager->findSourceView(loc);
+                SLANG_ASSERT(sourceView);
+
+                // We want only one view per SourceFile
+                SourceFile* sourceFile = sourceView->getSourceFile();
+
+                // NOTE! The view found might be different than sourceView. 
+                viewIndex = views.findFirstIndex([&](SourceView* currentView) -> bool { return currentView->getSourceFile() == sourceFile; });
+
+                if (viewIndex < 0)
+                {
+                    viewIndex = views.getCount();
+                    views.add(sourceView);
+                }
+            }
+
+            SLANG_ASSERT(viewIndex >= 0);
+            SLANG_ASSERT(sourceView && sourceView->getRange().contains(loc));
+
+            // Set the file index
+            entry.viewIndex = viewIndex;
+            // Set as the offset within the file 
+            entry.locOrOffset = sourceView->getRange().getOffset(loc);
+        }
+
+        // Sort into view/file and then offset order
+        entries.sort([](Entry& a, Entry& b) { return (a.viewIndex < b.viewIndex) || ((a.viewIndex == b.viewIndex) && a.locOrOffset < b.locOrOffset); });
+    }
+
+    {
+        TokenList tokens;
+
+        MemoryArena memoryArena;
+        RootNamePool rootNamePool;
+        NamePool namePool;
+        namePool.setRootNamePool(&rootNamePool);
+
+        Index viewIndex = -1;
+        SourceView* sourceView = nullptr;
+
+        for (auto& entry : entries)
+        {
+            if (viewIndex != entry.viewIndex)
+            {
+                viewIndex = entry.viewIndex;
+                sourceView = views[viewIndex];
+
+                // Make all memory free again
+                memoryArena.reset();
+
+                // Run the lexer
+                Lexer lexer;
+                lexer.initialize(sourceView, sink, &namePool, &memoryArena, Lexer::OptionFlag::TokenizeComments);
+
+                // Lex everything
+                tokens = lexer.lexAllTokens();
+            }
+
+            // Get the offset within the source file
+            const uint32_t offset = entry.locOrOffset;
+
+            // We need to get the loc in the source views space, so we look up appropriately in the list of tokens (which uses the views loc range)
+            const SourceLoc loc = sourceView->getRange().getSourceLocFromOffset(offset);
+
+            // Work out the line number
+            SourceFile* sourceFile = sourceView->getSourceFile();
+            const Index lineIndex = sourceFile->calcLineIndexFromOffset(int(offset));
+
+            // Okay, lets find the token index with a binary chop
+            Index tokenIndex = _findTokenIndex(loc, tokens.m_tokens.getBuffer(), tokens.m_tokens.getCount());
+            if (tokenIndex >= 0 && lineIndex >= 0)
+            {
+                FindInfo findInfo;
+                findInfo.declTokenIndex = tokenIndex;
+                findInfo.declLineIndex = lineIndex;
+                findInfo.tokenList = &tokens;
+                findInfo.sourceView = sourceView;
+
+                // Okay let's see if we extract some documentation then for this.
+                FoundMarkup foundMarkup;
+                SlangResult res = _findMarkup(findInfo, entry.decl, foundMarkup);
+
+                if (SLANG_SUCCEEDED(res))
+                {
+                    // We need to extract
+                    StringBuilder buf;
+                    SLANG_RETURN_ON_FAIL(_extractMarkup(findInfo, foundMarkup, buf));
+
+                    // Add to the documentation
+                    DocMarkup::Entry& docEntry = m_doc->addEntry(entry.decl);
+                    docEntry.m_markup = buf;
+                }
+                else if (res != SLANG_E_NOT_FOUND)
+                {
+                    return res;
+                }
+            }
+        }
+    }
+   
+    return SLANG_OK;
+}
+
+SlangResult DocMarkup::extract(ModuleDecl* moduleDecl, SourceManager* sourceManager, DiagnosticSink* sink)
+{
+    m_moduleDecl = moduleDecl;
+
+    DocMarkupExtractor context;
+    return context.extract(this, moduleDecl, sourceManager, sink);
+}
+
+/* static */SlangResult DocumentationUtil::writeMarkdown(DocMarkup* markup, StringBuilder& out)
+{
+    for (const auto& entry : markup->getEntries())
+    {
+        NodeBase* node = entry.m_node;
+        Decl* decl = as<Decl>(node);
+        if (!decl)
+        {
+            continue;
+        }
+
+        // Skip these they will be output as part of their respective 'containers'
+        if (as<ParamDecl>(decl) || as<EnumCaseDecl>(decl))
+        {
+            continue;
+        }
+
+        if (CallableDecl* callableDecl = as<CallableDecl>(decl))
+        {
+            out << entry.m_markup;
+
+            // There's code to output sigs in the SemanticsVisitor - we probably need to extract that functionality
+            // out so can be used here
+
+            //   String declString = getDeclSignatureString(item);
+
+            auto params = callableDecl->getParameters();
+            //const auto& returnType = callableDecl->returnType;
+
+            // Let's see if we can get markup on the parameters            
+            for (auto param : params)
+            {
+                DocMarkup::Entry* paramEntry = markup->getEntry(param);
+
+                if (paramEntry)
+                {
+                    out << paramEntry->m_markup;
+
+                    auto type = param->getType();
+
+                    if (type)
+                    {
+                        out << type->toString();
+                    }
+
+                    Name* name = param->getName();
+                    if (name)
+                    {
+                        out << " ";
+                        out << name->text;
+                    }
+                    out << "\n\n";
+                }
+            }
+        }
+        else if (EnumDecl* enumDecl = as<EnumDecl>(decl))
+        {
+
+        }
+        else if (StructDecl* structDecl = as<StructDecl>(decl))
+        {
+        }
+        else if (ClassDecl* classDecl = as<ClassDecl>(decl))
+        {
+        }
+    }
+
+    return SLANG_OK;
+}
+
+} // namespace Slang
diff --git a/source/slang/slang-doc.h b/source/slang/slang-doc.h
new file mode 100644
index 000000000..49251808a
--- /dev/null
+++ b/source/slang/slang-doc.h
@@ -0,0 +1,70 @@
+// slang-doc.h
+#ifndef SLANG_DOC_H
+#define SLANG_DOC_H
+
+#include "../core/slang-basic.h"
+#include "slang-ast-all.h"
+
+namespace Slang {
+
+/* Holds the documentation markup that is associated with each node (typically a decl) from a module */
+class DocMarkup : public RefObject
+{
+public:
+    struct Entry
+    {
+        NodeBase* m_node;           ///< The node this documentation is associated with
+        String m_markup;            ///< The raw contents of of markup associated with the decoration
+    };
+
+        /// Adds an entry, returns the reference to pre-existing node if there is one
+    Entry& addEntry(NodeBase* base);
+        /// Gets an entry for a node. Returns nullptr if there is no markup.
+    Entry* getEntry(NodeBase* base);
+
+        /// Get list of all of the entries in source order
+    const List<Entry>& getEntries() const { return m_entries; }
+
+        /// Given a module extracts all the associated markup.
+    SlangResult extract(ModuleDecl* moduleDecl, SourceManager* sourceManager, DiagnosticSink* sink);
+
+protected:
+
+        /// The module this information was extracted from
+    ModuleDecl* m_moduleDecl;
+        /// Map from AST nodes to documentation entries
+    Dictionary<NodeBase*, Index> m_entryMap;
+        /// All of the documentation entries in source order
+    List<Entry> m_entries;
+};
+
+// ---------------------------------------------------------------------------
+SLANG_INLINE DocMarkup::Entry& DocMarkup::addEntry(NodeBase* base)
+{
+    const Index count = m_entries.getCount();
+    const Index index = m_entryMap.GetOrAddValue(base, count);
+
+    if (index == count)
+    {
+        Entry entry;
+        entry.m_node = base;
+        m_entries.add(entry);
+    }
+    return m_entries[index];
+}
+
+// ---------------------------------------------------------------------------
+SLANG_INLINE DocMarkup::Entry* DocMarkup::getEntry(NodeBase* base)
+{
+    Index* indexPtr = m_entryMap.TryGetValue(base);
+    return (indexPtr) ? &m_entries[*indexPtr] : nullptr;
+}
+
+struct DocumentationUtil
+{
+    static SlangResult writeMarkdown(DocMarkup* markup, StringBuilder& out);
+};
+
+} // namespace Slang
+
+#endif
diff --git a/source/slang/slang-lexer.cpp b/source/slang/slang-lexer.cpp
index b0146c5b0..6c8e9474a 100644
--- a/source/slang/slang-lexer.cpp
+++ b/source/slang/slang-lexer.cpp
@@ -75,27 +75,29 @@ namespace Slang
     // Lexer
 
     void Lexer::initialize(
-        SourceView*     inSourceView,
-        DiagnosticSink* inSink,
-        NamePool*       inNamePool,
-        MemoryArena*    inMemoryArena)
+        SourceView*     sourceView,
+        DiagnosticSink* sink,
+        NamePool*       namePool,
+        MemoryArena*    memoryArena,
+        OptionFlags     optionFlags)
     {
-        m_sourceView  = inSourceView;
-        m_sink        = inSink;
-        m_namePool    = inNamePool;
-        m_memoryArena = inMemoryArena;
+        m_sourceView  = sourceView;
+        m_sink        = sink;
+        m_namePool    = namePool;
+        m_memoryArena = memoryArena;
 
-        auto content = inSourceView->getContent();
+        auto content = sourceView->getContent();
         
         m_begin   = content.begin();
         m_cursor  = content.begin();
         m_end     = content.end();
 
         // Set the start location
-        m_startLoc = inSourceView->getRange().begin;
+        m_startLoc = sourceView->getRange().begin;
 
         m_tokenFlags = TokenFlag::AtStartOfLine | TokenFlag::AfterWhitespace;
         m_lexerFlags = 0;
+        m_optionFlags = optionFlags;
     }
 
     Lexer::~Lexer()
@@ -1231,11 +1233,31 @@ namespace Slang
                 continue;
 
             case TokenType::WhiteSpace:
-            case TokenType::LineComment:
-            case TokenType::BlockComment:
+            {
                 flags |= TokenFlag::AfterWhitespace;
                 continue;
+            }
+            case TokenType::BlockComment:
+            case TokenType::LineComment:
+            {
+                flags |= TokenFlag::AfterWhitespace;
+                if (m_optionFlags & OptionFlag::TokenizeComments)
+                {
+                    // We don't break here, and use the normal token adding logic
+                    // because we want the behavior to be identical (in terms of flags etc)
+                    // as if TokenizeComments is not enabled
+                    char const* textEnd = m_cursor;
+
+                    token.type =  tokenType;
+                    token.flags = m_tokenFlags;
+                    token.setContent(UnownedStringSlice(textBegin, textEnd));
 
+                    return token;
+                }
+
+                continue;
+            }
+            
             // We don't want to skip the end-of-file token, but we *do*
             // want to make sure it has appropriate flags to make our life easier
             case TokenType::EndOfFile:
diff --git a/source/slang/slang-lexer.h b/source/slang/slang-lexer.h
index 957d05fec..f1fe89516 100644
--- a/source/slang/slang-lexer.h
+++ b/source/slang/slang-lexer.h
@@ -90,19 +90,29 @@ namespace Slang
     typedef unsigned int LexerFlags;
     enum
     {
-        kLexerFlag_InDirective      = 1 << 0, ///< Turn end-of-line and end-of-file into end-of-directive
-        kLexerFlag_ExpectFileName   = 1 << 1, ///< Support `<>` style strings for file paths
-        kLexerFlag_IgnoreInvalid    = 1 << 2, ///< Suppress errors about invalid/unsupported characters
-        kLexerFlag_ExpectDirectiveMessage = 1 << 3, ///< Don't lexer ordinary tokens, and instead consume rest of line as a string
+        kLexerFlag_InDirective              = 1 << 0, ///< Turn end-of-line and end-of-file into end-of-directive
+        kLexerFlag_ExpectFileName           = 1 << 1, ///< Support `<>` style strings for file paths
+        kLexerFlag_IgnoreInvalid            = 1 << 2, ///< Suppress errors about invalid/unsupported characters
+        kLexerFlag_ExpectDirectiveMessage   = 1 << 3, ///< Don't lexer ordinary tokens, and instead consume rest of line as a string
     };
 
     struct Lexer
     {
+        typedef uint32_t OptionFlags;
+        struct OptionFlag
+        {
+            enum Enum : OptionFlags
+            {
+                TokenizeComments         = 1 << 0, ///< If set comments will be output to the token stream
+            };
+        };
+
         void initialize(
             SourceView*     sourceView,
             DiagnosticSink* sink,
             NamePool*       namePool,
-            MemoryArena*    memoryArena);
+            MemoryArena*    memoryArena,
+            OptionFlags     optionFlags = 0);
 
         ~Lexer();
 
@@ -134,6 +144,7 @@ namespace Slang
 
         TokenFlags      m_tokenFlags;
         LexerFlags      m_lexerFlags;
+        OptionFlags     m_optionFlags;
 
         MemoryArena*    m_memoryArena;
     };
diff --git a/source/slang/slang-options.cpp b/source/slang/slang-options.cpp
index d28b50b88..fb7ba79c6 100644
--- a/source/slang/slang-options.cpp
+++ b/source/slang/slang-options.cpp
@@ -531,6 +531,10 @@ struct OptionsParser
                 {
                     requestImpl->getFrontEndReq()->shouldDumpAST = true;
                 }
+                else if (argStr == "-doc")
+                {
+                    requestImpl->getFrontEndReq()->shouldDocument = true;
+                }
                 else if (argStr == "-dump-repro")
                 {
                     SLANG_RETURN_ON_FAIL(tryReadCommandLineArgument(sink, arg, &argCursor, argEnd, requestImpl->m_dumpRepro));
diff --git a/source/slang/slang-source-loc.cpp b/source/slang/slang-source-loc.cpp
index 4b9e16b8e..3f9f8ad31 100644
--- a/source/slang/slang-source-loc.cpp
+++ b/source/slang/slang-source-loc.cpp
@@ -279,6 +279,55 @@ const List<uint32_t>& SourceFile::getLineBreakOffsets()
     return m_lineBreakOffsets;
 }
 
+SourceFile::OffsetRange SourceFile::getOffsetRangeAtLineIndex(Index lineIndex)
+{
+    const List<uint32_t>& offsets = getLineBreakOffsets();
+    const Index count = offsets.getCount();
+
+    if (lineIndex >= count - 1)
+    {
+        // Work out the line start
+        const uint32_t offsetEnd = uint32_t(getContentSize());
+        const uint32_t offsetStart = (lineIndex >= count) ? offsetEnd : offsets[lineIndex];
+        // The line is the span from start, to the end of the content
+        return OffsetRange{ offsetStart, offsetEnd };
+    }
+    else
+    {
+        const uint32_t offsetStart = offsets[lineIndex];
+        const uint32_t offsetEnd = offsets[lineIndex + 1];
+        return OffsetRange { offsetStart, offsetEnd };
+    }
+}
+
+UnownedStringSlice SourceFile::getLineAtIndex(Index lineIndex)
+{
+    const OffsetRange range = getOffsetRangeAtLineIndex(lineIndex);
+
+    if (range.isValid() && hasContent())
+    {
+        const UnownedStringSlice content = getContent();
+        SLANG_ASSERT(range.end <= uint32_t(content.getLength()));
+
+        const char*const text = content.begin();
+        return UnownedStringSlice(text + range.start, text + range.end);
+    }
+
+    return UnownedStringSlice();
+}
+
+UnownedStringSlice SourceFile::getLineContainingOffset(uint32_t offset)
+{
+    const Index lineIndex = calcLineIndexFromOffset(offset);
+    return getLineAtIndex(lineIndex);
+}
+
+bool SourceFile::isOffsetOnLine(uint32_t offset, Index lineIndex)
+{
+    const OffsetRange range = getOffsetRangeAtLineIndex(lineIndex);
+    return range.isValid() && range.containsInclusive(offset);
+}
+
 int SourceFile::calcLineIndexFromOffset(int offset)
 {
     SLANG_ASSERT(UInt(offset) <= getContentSize());
diff --git a/source/slang/slang-source-loc.h b/source/slang/slang-source-loc.h
index ef8b49c3b..54811918f 100644
--- a/source/slang/slang-source-loc.h
+++ b/source/slang/slang-source-loc.h
@@ -139,6 +139,9 @@ struct SourceRange
         /// Get the offset of a loc in this range
     int getOffset(SourceLoc loc) const { SLANG_ASSERT(contains(loc)); return int(loc.getRaw() - begin.getRaw()); }
 
+        /// Convert an offset to a loc
+    SourceLoc getSourceLocFromOffset(uint32_t offset) const { SLANG_ASSERT(offset <= getSize()); return begin + Int(offset); }
+
     SourceRange()
     {}
 
@@ -156,6 +159,7 @@ struct SourceRange
     SourceLoc end;
 };
 
+
 // Pre-declare
 struct SourceManager;
 
@@ -165,10 +169,45 @@ class SourceFile
 {
 public:
 
+    struct OffsetRange
+    {
+            /// We need a value to indicate an invalid range. We can't use 0 as that is valid for an offset range
+            /// We can't use a negative number, and don't want to make signed so we get the full 32-bits.
+            /// So we just use the max value as invalid
+        static const uint32_t kInvalid = 0xffffffff;
+
+            /// True if the range is valid
+        SLANG_FORCE_INLINE bool isValid() const { return end >= start && start != kInvalid; }
+            /// True if offset is within range (inclusively)
+        SLANG_FORCE_INLINE bool containsInclusive(uint32_t offset) const { return offset >= start && offset <= end; }
+
+            /// Get the count
+        SLANG_FORCE_INLINE uint32_t getCount() const { return end - start; }
+
+            /// Return an invalid range. 
+        static OffsetRange makeInvalid() { return OffsetRange{ kInvalid, kInvalid }; }
+
+        uint32_t start;
+        uint32_t end;
+    };
+
         /// Returns the line break offsets (in bytes from start of content)
         /// Note that this is lazily evaluated - the line breaks are only calculated on the first request 
     const List<uint32_t>& getLineBreakOffsets();
 
+        /// Returns true if the offset is on the specified line
+        /// NOTE! If offsets are not fully setup (because we don't have source), will only be correct for lines that have offsets
+    bool isOffsetOnLine(uint32_t offset, Index lineIndex);
+
+        /// Get the line containing the offset. Requires that content is available, else will return an empty slice. 
+    UnownedStringSlice getLineContainingOffset(uint32_t offset);
+
+        /// Get the line at the specified line index. Requires that content is available, else will return an empty slice. 
+    UnownedStringSlice getLineAtIndex(Index lineIndex);
+
+        /// Get the offset range at the specified line index. Works without content.
+    OffsetRange getOffsetRangeAtLineIndex(Index lineIndex);
+
         /// Set the line break offsets
     void setLineBreakOffsets(const uint32_t* offsets, UInt numOffsets);
 
diff --git a/source/slang/slang.cpp b/source/slang/slang.cpp
index 14f30c632..e02715015 100644
--- a/source/slang/slang.cpp
+++ b/source/slang/slang.cpp
@@ -30,6 +30,8 @@
 #include "slang-serialize-ir.h"
 #include "slang-serialize-container.h"
 
+#include "slang-doc.h"
+
 #include "slang-check-impl.h"
 
 #include "../../slang-tag-version.h"
@@ -1637,6 +1639,26 @@ void FrontEndCompileRequest::parseTranslationUnit(
             }
         }
 
+        if (shouldDocument)
+        {
+            RefPtr<DocMarkup> markup(new DocMarkup);
+            markup->extract(translationUnit->getModuleDecl(), getSourceManager(), getSink());
+
+            // Extract to a file
+
+            const String& path = sourceFile->getPathInfo().foundPath;
+            if (path.getLength())
+            {
+                String fileName = Path::getFileNameWithoutExt(path);
+                fileName.append(".md");
+
+                StringBuilder buf;
+                DocumentationUtil::writeMarkdown(markup, buf);
+
+                File::writeAllText(fileName, buf);
+            }
+        }
+
 #if 0
         // Test serialization
         {
diff --git a/tests/doc/doc.slang b/tests/doc/doc.slang
new file mode 100644
index 000000000..a2298ec4c
--- /dev/null
+++ b/tests/doc/doc.slang
@@ -0,0 +1,74 @@
+//DISABLE_TEST:SIMPLE:-entry computeMain -target spirv -stage compute -doc
+
+    /// Let's test indent
+    /// 
+    /// ```
+    /// {
+    ///     imIndented();
+    /// }   
+    /// ```
+    ///
+RWStructuredBuffer<int> inputBuffer;       
+
+/// A struct with some fields
+struct SomeStruct
+{
+    /// A field
+    int aField;
+    /// Multi-line 
+    /// is a thing
+    int anotherField;
+    int yetAnother; ///< A field with stuff 
+};
+
+/// An enum
+enum AnEnum
+{
+    Value, ///< A value
+    /// Another value
+    /// With a multi-line comment
+    AnotherValue,
+};
+
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<int> outputBuffer;       ///< An output buffer
+
+/// doThing!
+int doThing(int a,  ///< a parameter
+            int b)  ///< b parameter
+{
+    while (b >= 0)
+    {
+        a 
+        += 
+        a;
+    }
+    
+    return a;
+}
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int a = dispatchThreadID.x;
+    int b = dispatchThreadID.y;
+    int c = dispatchThreadID.z;
+    int d = a + b * c;
+    int e = d + c / 2;
+    
+    for (int i = 0; i < b; ++i)
+    {
+        if (e > 10 && i & 2)
+        {
+            a += b; b -= c; c += c; d = d + e + a; e = a;
+        }
+        else
+        {
+            a = e; b = c + c; d += d + __SyntaxError(); e = doThing(e, dispatchThreadID.x);
+            
+        }
+    }
+    
+    outputBuffer[dispatchThreadID.x] = a + b + c + d + e;
+}