Yantra is a powerful compiler compiler and LALR(1) parser generator written in C++, with the following core features:

• An integrated lexer
• Built-in support for UNICODE/UTF8 input
• Built-in AST builder
• Built-in AST walker(s)
• Bottom-up parsing (being LALR), and top-down walking (traversal)
• Multi-mode lexer, useful for implementing nested multi-line comments, etc.
• Lexer-driven (push-based) parser: reads input one character at a time and feeds tokens to the parser as they complete, useful for processing input as it arrives (e.g. from a socket).
• An optional amalgamated mode, where the entire parser is generated as a single cpp file, along with a full-featured main() function.
• Or, in non-amalgamated mode, the parser is generated as separate .hpp and .cpp files, ready to drop into an existing project.

The name Yantra is Sanskrit for machine, as in state machine in this context.

Yantra has no dependencies beyond the C++ standard library, so building it is a plain CMake build:

git clone [email protected]:TantrixAuto/yantra.git
cd yantra
mkdir build && cd build
cmake ..
cmake --build .

This produces the ycc executable in bin/. Save a grammar file, hello.y:

start := stmts;
stmts := stmts stmt;
stmts := stmt;
stmt := ID;

ID := "[A-Za-z]+";
WS := "\s"!;

Then generate a parser from it:

bin/ycc -c ascii -f hello.y -a

This writes hello.cpp (an amalgamated, self-contained parser with its own main()) and hello.log. Compile it with any C++23 compiler:

# clang
clang++ --std=c++23 -o hello hello.cpp

# gcc
g++ --std=c++23 -o hello hello.cpp

# MSVC (cl.exe, from a Developer Command Prompt)
cl /std:c++23 /EHsc /nologo hello.cpp

The grammar above recognizes one or more whitespace-separated alphabetic words. -s <string> feeds that string directly to the parser as input (as opposed to -f <filename>, which reads from a file, or -i, which reads interactively from the console):

# succeeds silently
$ ./hello -s "hello world"
$ echo $?
0

# -t1 prints the parsed AST
$ ./hello -s "hello world" -t1
0:start_1(1:stmts_1(2:stmts_2(3:stmt_1(4:ID(hello))) 2:stmt_1(3:ID(world))) 1:_tEND())

# fails: ID only matches letters, "123" isn't valid input for this grammar
$ ./hello -s "hello 123"
s1-err:?a1.in(001,007):TOKEN_ERROR{{token: }}
hello 123
$ echo $?
1

Yantra parses the entire input into an AST first, then walks it top-down calling your semantic actions, unlike most parser generators, where actions run bottom-up as each rule is reduced. That ordering is what lets a parent rule's action run before its children are visited. Save this as calc.y:

%class Calculator;

start := expr;

expr := expr(a) PLUS expr(b)
%{
    std::cout << "Adding" << std::endl;
%}

expr := NUMBER(N)
%{
    std::cout << "Number: " << N.text << std::endl;
%}

NUMBER := "\d+";
PLUS := "\+";
WS := "\s+"!;

Generate and compile it the same way as above (bin/ycc -c ascii -f calc.y -a, then any of the three compiler commands), then run it:

$ ./calc -s "1 + 2 + 3"
Adding
Number: 1
Adding
Number: 2
Number: 3

1 + 2 + 3 parses left-associatively as (1 + 2) + 3. So the outer Adding, the root of the tree, prints first, followed by its left child (Number: 1) and then its right child, which is itself another Adding node with its own two children. A hand-written recursive-descent or bottom-up parser would have to build extra AST classes and a separate walking pass to get this ordering. Here it falls out of the grammar directly.

See the Build Instructions and Tutorial below for a real walk-through of the grammar syntax.

• vs. Bison / Yacc / Lemon (the classic LALR(1) family):
  • Lemon, from SQLite, is Yantra's direct stated inspiration.
  • These run semantic actions during parsing, as each rule reduces, bottom-up.
  • Yantra always builds the full AST first, then walks it top-down in a separate pass, so a parent rule's action can run before its children are visited.
  • A single grammar can also define more than one walker (e.g. one that emits C++, another that emits Java, from the same parse).
  • Getting either of those out of the Bison family means hand-building your own AST and walker on top.
• vs. ANTLR:
  • ANTLR walks a fully-built parse tree too, but that comes for free from its LL(*) algorithm, which already builds the tree top-down as it parses.
  • Yantra gets the same top-down walk out of LALR(1), a bottom-up algorithm with no natural "whole tree exists yet" moment during parsing, while keeping LALR(1)'s time and space efficiency over adaptive LL(*).
  • Beyond that, Yantra targets C++ only (ANTLR generates for many languages) and ships its own integrated lexer with mode-stack support instead of a separate lexer generator.
  • ANTLR's own generator tool is Java, so using it from a C++ project means adding a JVM to the build toolchain just to run the generator. Yantra is a native C++ executable with no such dependency.
  • ANTLR is far more mature and widely used. Yantra is a much smaller, newer, single-maintainer project.
• vs. tree-sitter:
  • A different problem entirely. It's built for incremental, error-tolerant parsing embedded in editors and IDEs (what GitHub, Neovim, etc. use it for), not for generating a compiler/codegen backend.
  • Yantra doesn't do incremental reparsing and isn't trying to.

See Known Limitations for an honest list of what Yantra doesn't do yet.

The following are a set of key links to get familiar with Yantra.

It is recommended that they be read in the given order.

See https://github.com/TantrixAuto/lingo for standalone sample project that uses yantra.

This is a language server extension created by Raj Chaudhuri that provides syntax highlighting for Yantra files in vscode, qtcreator, and any other IDE that supports the Language Server Protocol.

https://github.com/rajware/yantra-language-server

Yantra is licensed under the MIT License.

Renji Panicker (@renjipanicker)