Md4qt

Md4qt

License: MIT

md4qt is a header-only C++ library for parsing Markdown.

md4qt supports CommonMark 0.31.2 Spec, and some GitHub extensions, such as tables, footnotes, tasks lists, strikethroughs, LaTeX Math injections, GitHub's autolinks.

md4qt can be built with Qt6 or with ICU.

This library parses Markdown into tree structure.

Example

#define MD4QT_QT_SUPPORT
#include <md4qt/parser.hpp>
int main()
{
auto doc = p.parse(QStringLiteral("your_markdown.md"));
for (auto it = doc->items().cbegin(), last = doc->items().cend(); it != last; ++it) {
switch ((*it)->type())
{
{
auto a = static_cast<MD::Anchor<MD::QStringTrait>*> (it->get());
qDebug() << a->label();
}
break;
default :
break;
}
}
return 0;
}
Just an anchor.
Definition doc.h:397
Markdown parser.
Definition parser.h:1422
std::shared_ptr< Document< Trait > > parse(const typename Trait::String &fileName, bool recursive=true, const typename Trait::StringList &ext={Trait::latin1ToString("md"), Trait::latin1ToString("markdown")}, bool fullyOptimizeParagraphs=true)
Definition parser.h:2091
@ Anchor
Anchor.

Benchmark

Approximate benchmark with cmark-gfm says, that Qt6 version of md4qt is slower ~13 times. But you will get complete C++ tree structure of the Markdown document with all major extensions and sugar and cherry on the cake.

Markdown library Result
cmark-gfm 0.23 ms
markdown-it (Rust) 2.24 ms
md4qt with Qt6 2.9 ms
md4qt with Qt6 without GitHub autolinks extension 2.5 ms

This measurement done with test file in markdown-it (Rust)

markdown-it (Rust) measurement done with markdown_it::plugins::extra

Playground

You can play in action with md4qt in Markdown Tools. There you can find Markdown editor/viewer/converter to PDF.

And KleverNotes from KDE uses md4qt too.

Release notes

  • Note that version 4.0.0 is API incompatible with 3.0.0. In version 4.0.0 were changed rules with spaces, this version fully supports CommonMark standard in this question. Methods isSpaceBefore(), isSpaceAfter() were removed, and spaces are presented as in Markdown, so keep it in mind.

Known issues

  • In contrast to CommonMark, md4qt has one difference. If in list item first element is HTML CommonMark doesn't apply rule to lazy continuation lines as this is not a paragraph, whereas md4qt does so. For example.

    * <!--
    -->

    In CommonMark will be.

    <ul>
    <li>
    <!--
    </li>
    </ul>
    <p>--&gt;</p>

    And in md4qt it will be.

    <ul>
    <li>
    <!--
    -->
    </li>
    </ul>

    I know about this difference. And I decided to keep it. But if somebody will provide compelling reasons to do exactly as in CommonMark in this case, I will work with this question.

Q/A

Why another AST Markdown parser?

  • When I wrote this library I knew about md4c parser, but not about cmark-gfm. md4c was not suitable for my purposes, whereas cmark-gfm could do everything I needed. But God did it so, so I wrote md4qt and only later knew about cmark-gfm. Ok, code is written and tested. Let it be.

    What I can say yet, is that this library is C++. And for some people can be easier to use C++ code instead of C with freeing memory by hand. Qt do things easier by handling text encoding... So let it be, guys.

    And one more cherry on the cake - md4qt can parse Markdown recursively. What it is described below.

What should I know about links in the document?

  • In some cases in Markdown link's URL is something document related. So, when you got a MD::Link in the document check if the labeled links of the document contains key with URL in the link, and if so, use URL from labeled links, look:

    QString url = item->url();
    const auto it = doc->labeledLinks().find(url);
    if (it != doc->labeledLinks().cend()) {
    url = it->second->url();
    }

What is the second argument of MD::Parser::parse()?

  • Second argument of MD::Parser::parse() is a flag that tells the parser to process Markdown files recursively or not. If parsing is recursive then if in the targeted Markdown file exist links to other Markdown files, then they will be parsed too and will exist in the resulting document.

What is an MD::Anchor?

  • As md4qt supports recursive Markdown parsing, then in the resulting document can be represented more than one Markdown file. Each file in the document starts with MD::Anchor, it just shows that during traversing through the document you reached new file.

Does the library throw exceptions?

  • No. This library doesn't use exceptions. Any text is a valid Markdown, so I don't need to inform user about errors. Qt itself doesn't use exceptions either. So you can catch only standard C++ exceptions, like std::bad_alloc, for example. Possibly with MD::UnicodeStringTrait you will catch more standard exceptions, possibly I missed something somewhere, but I tried to negotiate all possible exceptions.

Why MD::Parser and MD::Document are templates?

  • Since version 2.0.0 md4qt can be built not only with Qt6, but with STL too. The code of the parser is the same in both cases. I just added two ready traits to support different C++ worlds. With STL I use ICU library for Unicode handling, and uriparser library to parse and check URLs. These dependencies can be installed with the Conan package manager.

So, how can I use md4qt with Qt6 and ICU?

  • To build with ICU support you need to define MD4QT_ICU_STL_SUPPORT before including parser.h. In this case you will get access to MD::UnicodeStringTrait, that can be passed to MD::Parser as template parameter. You will receive in dependencies C++ STL, ICU and uriparser.

    To build with Qt6 support you need to define MD4QT_QT_SUPPORT. In this case you will get access to MD::QStringTrait to work with Qt's classes and functions. In this case in dependencies you will receive Qt6.

    You can define both to have ability to use md4qt with Qt6 and ICU.

ICU is slower than Qt6? Really?

  • Don't believe anybody, just build built-in md_benchamrk and have a look. Dry numbers say, that Qt6 QString is ~2 times faster than icu::UnicodeString in such tasks. Markdown parsing implies to check every symbol, and tied to use access to every character in the string with operator [] (...), or member at(...). I do it very often in the parser's code and the profiler says that most of the run-time is spent on such operations. QString is just more optimized for accessing separate characters than icu::UnicodeString...

Why is parsing wrong on Windows with std::ifstream?

  • Such a problem can occur on Windows with MSVC if you open the file in text mode, so for MD::Parser always open std::ifstream with std::ios::binary flag. And yes, I expect to receive UTF-8 encoded content...

How can I convert MD::Document into HTML?

How can I obtain positions of blocks/elements in Markdown file?

  • Done in version 2.0.5. Remember that all positions in md4qt start with 0, where first symbol on first line will have coordinates (0,0). One more important thing is that all ranges of position in md4qt are given inclusive, that mean that last column of any element will point to the last symbol in this element.

How can I easily traverse through the MD::Document?

  • Since version 2.6.0 in visitor.h header implemented MD::Visitor interface with which you can easily walk through the document, all you need is implement/override virtual methods to handle that or another element in the document, like:

    virtual void onHeading(
    //! Heading.
    Heading.
    Definition doc.h:710

Why don't you have an implementation for pure STL with std::string?

  • Because of performance, I did a pure STL implementation where the string class was an std::string with some small third-party library to handle UTF8, and benchmark said that the performance was like with Qt6 QString, so I decided to not support third trait. Maybe because I am lazy?

Is it possible to write custom text plugin for this parser?

  • Since version 3.0.0 in the MD::Parser was added a method for adding custom text plugins.

    //! Add text plugin.
    void addTextPlugin(
    //! ID of a plugin. Use TextPlugin::UserDefinedPluginID value for start ID.
    int id,
    //! Function of a plugin, that will be invoked to processs raw text.
    //! Should this plugin be used in parsing of internals of links?
    bool processInLinks,
    //! User data that will be passed to plugin function.
    const typename Trait::StringList &userData);
    std::function< void(std::shared_ptr< Paragraph< Trait > >, TextParsingOpts< Trait > &, const typename Trait::StringList &)> TextPluginFunc
    Functor type for text plugin.
    Definition parser.h:921

What is a ID of a plugin?

  • ID of a plugin is a regular int that should be (but not mandatory) started from

    enum TextPlugin : int {
    UnknownPluginID = 0,
    GitHubAutoLinkPluginID = 1,
    UserDefinedPluginID = 255
    }; // enum TextPlugin

    MD::UserDefinedPluginID value. Note that plugins will be invoked corresponding to its ID from smallest to largest, so a developer can handle an order of text plugins.

What is a MD::TextPluginFunc<Trait>?

  • Text plugin is a usual function with a signature

    template<class Trait>
    using TextPluginFunc = std::function<void(std::shared_ptr<Paragraph<Trait>>,
    TextParsingOpts<Trait> &, const typename Trait::StringList &)>;

    You will get already parsed MD::Paragraph with all items in it. And you are able to process remaining raw text data and check it for what you need.

    MD::TextParsingOpts is an auxiliary structure with some data. You are interested in bool collectRefLinks;, when this flag is true the parser is in a state of collecting reference links, and on this stage plugin may do nothing.

    A last argument of plugin function is a user data, that was passed to MD::Parser::addTextPlugin() method.

    A most important thing in MD::TextParsingOpts structure is a std::vector<TextData> m_rawTextData;. This vector contains not processed raw text data from Markdown. The size of m_rawTextData is the same as a count of MD::Text items in MD::Paragraph, and theirs sizes should remain equal. So, if you replace one of text item with something, for example link, corresponding text item should be removed from MD::Paragraph and m_rawTextData. Or if you replace just a part of text item - it should be modified in MD::Paragraph and m_rawTextData. Be careful, it's UB, if you will make a mistake here, possibly you will crash.

    One more thing - don't forget to set positions of elements in MD::Document to new values if you change something, and don't forget about such things like MD::ItemWithOpts::openStyles() and MD::ItemWithOpts::closeStyles(). Document should remain correct after your manipulations, so any syntax highlighter, for example, won't do a mistake.

    Note, that MD::TextData is

    struct TextData {
    typename Trait::String m_str;
    long long int m_pos = -1;
    long long int m_line = -1;
    };

    And m_pos and m_line here is relative to MD::MdBlock<Trait> & fr; member of MD::TextParsingOpts, but document require absolute positions in the Markdown text. So when you will set positions to new items, use, for example, a following code.

    setEndColumn(po.fr.data.at(s.line).first.virginPos(s.pos));

    where s is an object of MD::TextData type.

What is processInLinks flag for?

  • processInLinks flag should be set to false if you desire to not process your plugin in link's captions, as, for example, links can't contain other links, so if you are implementing a plugin for new links this flag should be set to false for your plugin.

What for is a userData argument?

  • This list of strings will be passed to plugin function. This is auxiliary data that can be handy for plugin implementation.

Could you show an example of a plugin?

  • In md4qt already exists one text plugin for handling GitHub's autolink. A plugin function is quite simple, look.

    template<class Trait>
    inline void
    githubAutolinkPlugin(std::shared_ptr<Paragraph<Trait>> p,
    TextParsingOpts< Trait > &po)
    {
    if (!po.collectRefLinks) {
    long long int i = 0;
    while (i >= 0 && i < (long long int) po.rawTextData.size()) {
    i = processGitHubAutolinkExtension(p, po, i);
    ++i;
    }
    }
    }

    But MD::processGitHubAutolinkExtension() is not so trivial :) Have a look at its implementation to have a good example, it's placed in parser.h.

    Good luck with plugining. :)

I didn't understand how raw text data correlates with a paragraph.

  • Let I will show you on example how raw text data correlate with paragraph. Just two diagrams and you won't have anymore questions. Look.

    Consider we want to replace any occurence of @X by some kind of a link. Before modifications we had.

And after work of your plugin we should have.

How can I get a string of MD::StyleDelim?

  • Since version 3.0.0 was added a function to get a substring from text fragment with given virgin positions.

    template<class Trait>
    inline typename Trait::String
    virginSubstr(const MdBlock<Trait> &fr, const WithPosition &virginPos);

    And a function to get local position from virgin one.

    template<class Trait>
    inline std::pair<long long int, long long int>
    localPosFromVirgin(const MdBlock<Trait> &fr,
    long long int virginColumn,
    long long int virginLine)

Is it possible to find Markdown item by its position?

How can I walk through the document and find all items of given type?

  • Since version 3.0.0 was added algorithm MD::forEach().

    //! Calls function for each item in the document with the given type.
    template<class Trait>
    inline void
    forEach(
    //! Vector of item's types to be processed.
    const typename Trait::template Vector<ItemType> &types,
    //! Document.
    std::shared_ptr<Document<Trait>> doc,
    //! Functor object.
    ItemFunctor<Trait> func,
    //! Maximun nesting level.
    //! 0 means infinity, 1 - only top level items...
    unsigned int maxNestingLevel = 0);

How can I add and process a custom (user-defined) item in MD::Document?

  • Since version 3.0.0 in MD::ItemType enum appeared MD::UserDefined enumerator. So you can inherit from any MD::Item class and return from type() method value greater or equal MD::ItemType::UserData. To handle user-defined types of items in MD::Visitor class now exists method void onUserDefined(MD::Item<Trait> *item). So you can handle your custom items and do what you need.
This file is part of the KDE documentation.
Documentation copyright © 1996-2024 The KDE developers.
Generated on Mon Nov 18 2024 12:19:15 by doxygen 1.12.0 written by Dimitri van Heesch, © 1997-2006

KDE's Doxygen guidelines are available online.