Разработка компилятора для «typescript»на LLVM#1 Дмитрий Пацура @ovr https://github.com/ovr 

+2 Дмитрий Пацура Fintier UK Пишу на много чем) https://github.com/ovr GHubber PHPSA 

Разработка компилятора для "TypeScript" используя LLVM в качестве backend, и TypeScript в качестве frontend, на TypeScript Часть 1 

Сегодня в программе Теория о компиляторах Немного об LLVM Пишем программы LLVM IR Пишем компилятор Горячие закуски Выпивка +4 

+5 TypeScript как frontend LLVM Null terminate String Про JS будет, но чуток позжее Runtime Library Кодогенерация libstdc libstdc++ Линковшики 

Немножко про компиляторы 

*.ts Идея: Написать компилятор для Typescript Binary Frontend? Backend? Немножко про компиляторы D 

*.cpp *.h Let’s take an example CLang Executable/Binary preprocessor Compile *.cpp -> *.o Linking file1.o file2.o Немножко про компиляторы A 

LLVM C/C++ language BINARY BACKEND Frontend Clang Немножко про компиляторы R 

+10 LLVM - Low Level Virtual Machine The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. Despite its name, LLVM has little to do with traditional virtual machines. The name "LLVM" itself is not an acronym; it is the full name of the project. Немножко про компиляторы T 

+11 Почему LLVM? LLVM IR (optimize, cfg, cfa) Clang (C/C++) Fortran etc… X86 PowerPC ARM Frontend Backend Немножко про компиляторы Ж 

+12 LLVM Language Reference Manual https://llvm.org/docs/LangRef.html LLVM IR - LLVM intermediate representation lllvm-llc LLVM IR Obj Немножко про компиляторы И 

+13 Каждый уважающий себя программист хочет сделать свой собственный компилятор. Мечты сбываются! LLVM — важный шаг, позволяющий избежать велосипедостроения. (C) @gridem Немножко про компиляторы В 

Генерация кода LLVM IR 

+15 Level 0 program that cannot do anything Генерация кода LLVM IR 

+16 int main() { return 0; } program that cannot do anything clang -S -emit-llvm main.cpp Генерация кода LLVM IR 

; ModuleID = 'main.cpp' source_filename = "main.cpp" target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" target triple = "x86_64-apple-macosx10.14.0" ; Function Attrs: noinline norecurse nounwind optnone ssp uwtable define i32 @main() #0 { %1 = alloca i32, align 4 store i32 0, i32* %1, align 4 ret i32 0 } attributes #0 = { noinline norecurse nounwind optnone ssp uwtable «…..» } !llvm.module.flags = !{!0, !1} !llvm.ident = !{!2} !0 = !{i32 1, !"wchar_size", i32 4} !1 = !{i32 7, !"PIC Level", i32 2} !2 = !{!"Apple LLVM version 10.0.0 (clang-1000.11.45.2)"} 

+18 ; ModuleID = 'main.cpp' source_filename = "main.cpp" target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" target triple = "x86_64-apple-macosx10.14.0" Генерация кода LLVM IR 

+19 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" https://llvm.org/docs/LangRef.html#data-layout Specifies that the target lays out data in little-endian form. e m:o : Mach-O mangling: Private symbols get L prefix. Other symbols get a _ prefix. https://llvm.org/devmtg/2014-10/Slides/Prashanth-DLO.pdf Implementing Data Layout Optimizations in the LLVM Framework Генерация кода LLVM IR 

+20 ; Function Attrs: noinline … define i32 @main() #0 { %1 = alloca i32, align 4 store i32 0, i32* %1, align 4 ret i32 0 } clang -S -emit-llvm main.cpp Генерация кода LLVM IR 

+21 ; Function Attrs: norecurse nounwind readnone ssp uwtable define i32 @main() local_unnamed_addr #0 { ret i32 0 } clang -S -emit-llvm main.cpp -O3 Генерация кода LLVM IR 

.section __TEXT,__text,regular,pure_instructions .macosx_version_min 10, 14 .globl _main ## -- Begin function main .p2align 4, 0x90 _main: ## @main .cfi_startproc ## %bb.0: pushq %rbp .cfi_def_cfa_offset 16 .cfi_offset %rbp, -16 movq %rsp, %rbp .cfi_def_cfa_register %rbp xorl %eax, %eax popq %rbp retq .cfi_endproc llc main.ll Генерация кода LLVM IR 

+23 gcc main.s -o main && objdump -d main ovr@MBP-Dmitry  ~/projects/ovr/hlvm   master ● ?  gcc main.s -o main && objdump -d main  2 ↵  8.11.1 ⬢ main: file format Mach-O 64-bit x86-64 Disassembly of section __TEXT,__text: __text: 100000fb0: 55 pushq %rbp 100000fb1: 48 89 e5 movq %rsp, %rbp 100000fb4: 31 c0 xorl %eax, %eax 100000fb6: 5d popq %rbp 100000fb7: c3 retq _main: 100000fb0: 55 pushq %rbp 100000fb1: 48 89 e5 movq %rsp, %rbp 100000fb4: 31 c0 xorl %eax, %eax 100000fb6: 5d popq %rbp 100000fb7: c3 retq Генерация кода LLVM IR 

Level 1 Simple math 

+25 int main() { int a = 5; int b = 6; return a + b; } clang -S -emit-llvm main.cpp -O0 Генерация кода LLVM IR 

define i32 @main() #0 { %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 store i32 0, i32* %1, align 4 store i32 5, i32* %2, align 4 store i32 6, i32* %3, align 4 %4 = load i32, i32* %2, align 4 %5 = load i32, i32* %3, align 4 %6 = add nsw i32 %4, %5 ret i32 %6 } 

+27 %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 Static Single Assignment (SSA) The ‘alloca’ instruction allocates memory on the stack frame of the currently executing function, to be automatically released when this function returns to its caller. The object is always allocated in the address space for allocas indicated in the datalayout . = alloca [inalloca] [, ] [, align ] [, addrspace()] Генерация кода LLVM IR 

Сыграем в игру? define i32 @main() #0 { %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 store i32 0, i32* %1, align 4 store i32 5, i32* %2, align 4 store i32 6, i32* %3, align 4 %4 = load i32, i32* %2, align 4 %5 = load i32, i32* %3, align 4 %6 = add nsw i32 %4, %5 ret i32 %6 } 

define i32 @main() #0 { %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 store i32 0, i32* %1, align 4 store i32 5, i32* %2, align 4 store i32 6, i32* %3, align 4 %4 = load i32, i32* %2, align 4 %5 = load i32, i32* %3, align 4 %6 = add nsw i32 %4, %5 ret i32 %6 } 

define i32 @main() #0 { %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 store i32 0, i32* %1, align 4 store i32 5, i32* %2, align 4 store i32 6, i32* %3, align 4 %4 = load i32, i32* %2, align 4 %5 = load i32, i32* %3, align 4 %6 = add nsw i32 %4, %5 ret i32 %6 } 

define i32 @main() #0 { %1 = alloca i32, align 4 %2 = alloca i32, align 4 %3 = alloca i32, align 4 store i32 0, i32* %1, align 4 store i32 5, i32* %2, align 4 store i32 6, i32* %3, align 4 %4 = load i32, i32* %2, align 4 %5 = load i32, i32* %3, align 4 %6 = add nsw i32 %4, %5 ret i32 %6 } 

define i32 @main() #0 { %a.0 = alloca i32, align 4 %b.0 = alloca i32, align 4 store i32 5, i32* %a.0, align 4 store i32 6, i32* %b.0, align 4 %a.1 = load i32, i32* %a.0, align 4 %b.1 = load i32, i32* %b.0, align 4 %c.0 = add nsw i32 %a.1, %b.1 ret i32 %c.0 } 

define i32 @main() #0 { %a.0 = alloca i32, align 4 %b.0 = alloca i32, align 4 store i32 5, i32* %a.0, align 4 store i32 6, i32* %b.0, align 4 %a.1 = load i32, i32* %a.0, align 4 %b.1 = load i32, i32* %b.0, align 4 %c.0 = add nsw i32 %a.1, %b.1 ret i32 %c.0 } 

define i32 @main() #0 { %a.0 = alloca i32, align 4 %b.0 = alloca i32, align 4 store i32 5, i32* %a.0, align 4 store i32 6, i32* %b.0, align 4 %a.1 = load i32, i32* %a.0, align 4 %b.1 = load i32, i32* %b.0, align 4 %c.0 = add nsw i32 %a.1, %b.1 ret i32 %c.0 } 

Level 2 Hello World 

+36 Что бы распечатать строку, нужно: функция и строка Генерация кода LLVM IR 

+37 Стандартная библиотека языка Си libc #include int puts(const char *); Генерация кода LLVM IR 

+38 Зачем в C/C++ заголовочные файлы? cstdlib.so stdlib.h int puts(const str *) Генерация кода LLVM IR 

+39 Зачем в C/C++ заголовочные файлы? #include int main() { puts("Hello"); return 0; } main.cpp main.o Main compiler linker 

+40 clang -S -emit-llvm main.cpp -O0 declare i32 @puts(i8*) #1 @.str = private unnamed_addr constant [6 x i8] c"Hello\00" define i32 @main() #0 { %1 = getelementptr inbounds ([6 x i8], [6 x i8]* @.str, i32 0, i32 0) %2 = call i32 @puts(i8* %1) ret i32 0 } Генерация кода LLVM IR 

+41 @.str = private unnamed_addr constant [6 x i8] c"Hello\00" [6 x i8] c"Hello\00" [6 x i8] c"12345\00" Null-terminated string / С string Генерация кода LLVM IR 

+42 H E L L O \0 const char* str = *str *(str+1) *(str+2) *(str+3) *(str+4) *(str+5) How it works int puts(const str *)? Генерация кода LLVM IR 

Стоп, а где компилятор? 

+44 LLVM TypeScript BACKEND Frontend *.ts Компилятор языка «TypeScript» Binary 

+45 { puts("Hello PiterJS!"); } First Program in Static Script Компилятор - Frontend 

+46 import * as ts from 'typescript'; const options = { lib: [], types: [] }; const files = ['sandbox/do-simple-math.ts']; const host = ts.createCompilerHost(options); const program = ts.createProgram(files, options, host); Компилятор - Frontend 

+47 Let’s use TypeScript diagnostics const diagnostics = ts.getPreEmitDiagnostics(program); if (diagnostics.length) { diagnostics.forEach(diagnostic => { const message = ts.flattenDiagnosticMessageText(diagnostic.messageText, '\n'); if (!diagnostic.file) { console.log(message); return; } const { line, character } = diagnostic.file.getLineAndCharacterOfPosition(diagnostic.start!); console.log(`${diagnostic.file.fileName} (${line + 1},${character + 1}): ${message}`); }); process.exit(1); } Компилятор - Frontend 

ovr@MBP-Dmitry  ~hlvm   master  node ./build/cli.js  SIGINT(2) ↵ Cannot find global type 'Array'. Cannot find global type 'Boolean'. Cannot find global type 'Function'. Cannot find global type 'IArguments'. Cannot find global type 'Number'. Cannot find global type 'Object'. Cannot find global type 'RegExp'. Cannot find global type 'String'. sandbox/do-simple-math.ts (3,5): Cannot find name 'puts'. 

const options = { lib: [ path.join(__dirname, '..', 'language.d.ts') ], types: [] }; interface Boolean {} interface Function {} interface IArguments {} interface Number {} interface Object {} interface RegExp {} interface String {} interface Array {} declare function puts(str: string): void; 

Backend - CodeGeneration 

https://github.com/MichaReiser/llvm-node Micha Reiser MichaReiser 

const llvmContext = new llvm.LLVMContext(); const llvmModule = new llvm.Module("test", this.llvmContext); const block = llvm.BasicBlock.create(llvmContext, "Entry", mainFn); const builder = new llvm.IRBuilder(block); const mainFnType = llvm.FunctionType.get( llvm.Type.getVoidTy(llvmContext), false ); const mainFn = llvm.Function.create( mainFnType, LinkageTypes.ExternalLinkage, "main", llvmModule ); import * as llvm from 'llvm-node'; 

Abstract Syntax Tree 

Statement Expression Literal -> llvm.value -> llvm.value 

export function passStatement(stmt: ts.Statement, ctx: Context, builder: llvm.IRBuilder) { switch (stmt.kind) { case ts.SyntaxKind.Block: passBlockStatement(stmt, ctx, builder); break; case ts.SyntaxKind.ReturnStatement: passReturnStatement(stmt, ctx, builder); break; case ts.SyntaxKind.IfStatement: passIfStatement(stmt, ctx, builder); break; case ts.SyntaxKind.ForStatement: passForStatement(stmt, ctx, builder); break; default: throw new UnsupportedError( stmt, `Unsupported statement: "${stmt.kind}"` ); } } 

Use Context + IRBuilder export class Context { public typeChecker: ts.TypeChecker; public llvmContext: llvm.LLVMContext; public llvmModule: llvm.Module; public scope: Scope = new Scope(); } export class Scope { public functions: FunctionsTable = new FunctionsTable(); public variables: VariablesTable = new VariablesTable(); } 

export function buildFromStringLiteral( node: ts.StringLiteral, ctx: Context, builder: llvm.IRBuilder ): llvm.Value { return builder.createGlobalStringPtr( node.text, ); } StringLiteral -> llvm.Value 

NumbericLiteral -> llvm.Value function buildFromNumericLiteral( value: ts.NumericLiteral, ctx: Context, builder: llvm.IRBuilder, ): llvm.Value { return llvm.ConstantFP.get( ctx.llvmContext, parseFloat(value.text) ); } 

Level 3 Runtime Library 

{ function doMath(): number { const a = 5.5; const b = 14.5; return ((a + b) * 50) / 10; } puts("hello"); puts(doMath()); } declare function puts(str: string): void; do-simple-math.ts:11:10 - error TS2345: Argument of type 'number' is not assignable to parameter of type 'string'. 11 puts(doMath()); ~~~~~~~~ 

+61 #include #include #include #include "include/helpers.h" __attribute__ ((visibility ("default"))) const char* number2string(double number) { std::stringstream s; s << number; return s.str().c_str(); } Runtime Library 

+62 cmake_minimum_required(VERSION 3.12) project(hlvm-runtime) set(CMAKE_BUILD_TYPE Release) set(CMAKE_CXX_STANDARD 14) set(CMAKE_CXX_FLAGS "-pedantic -Wextra -O2") set(SOURCE_FILES helpers.cpp) add_library(hlvm-runtime STATIC ${SOURCE_FILES}) add_library( IMPORTED [GLOBAL]) Runtime Library 

define double @doMath() { Entry: %a = alloca double store double 5.500000e+00, double* %a %b = alloca double store double 1.450000e+01, double* %b %0 = load double, double* %a %1 = load double, double* %b %2 = fadd double %0, %1 %3 = fmul double %2, 5.000000e+01 %4 = fdiv double %3, 1.000000e+01 ret double %4 } define void @main() { Entry: %0 = call double @doMath() %1 = call i8* @number2string(double %0) call void @puts(i8* %1) ret void } declare void @puts(i8*) declare i8* @number2string(double) { function doMath(): number { const a = 5.5; const b = 14.5; return ((a + b) * 50) / 10; } puts(number2string(doMath())); } 

Undefined symbols for architecture x86_64: "_number2string", referenced from: _main in hello_world-0f8a45.o ld: symbol(s) not found for architecture x86_64 clang: error: linker command failed with exit code 1 (use -v to see invocation) Runtime Library 

OVERVIEW: llvm symbol table dumper ovr@MBP-Dmitry  ~/projects/ovr/hlvm/runtime   master ✚ ● ?  nm libhlvm-runtime.a.  ✔ libhlvm-runtime.a(helpers.cpp.o): 0000000000000e4c s GCC_except_table0 0000000000000e94 s GCC_except_table13 U __Unwind_Resume 0000000000000000 T __Z13number2stringd 0000000000000ab0 T __ZNKSt3__115basic_stringbufIcNS_11char_traitsIcEENS_9allocatorIcEEE3strEv U __ZNKSt3__121__basic_string_commonILb1EE20__throw_length_errorEv U __ZNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEE6resizeEmc U __ZNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEE9push_backEc U __ZNSt3__113basic_istreamIcNS_11char_traitsIcEEED0Ev U __ZNSt3__113basic_istreamIcNS_11char_traitsIcEEED1Ev U __ZNSt3__113basic_ostreamIcNS_11char_traitsIcEEED0Ev U __ZNSt3__115basic_streambufIcNS_11char_traitsIcEEE6xsgetnEPcl … 

-demangle Demangle C++ symbol names ovr@MBP-Dmitry  ~/projects/ovr/hlvm/runtime   master ✚ ● ?  nm -C libhlvm-runtime.a.  ✔ libhlvm-runtime.a(helpers.cpp.o): U __Unwind_Resume 0000000000000000 T number2string(double) 0000000000000ab0 T std::__1::basic_stringbuf, std::__1::allocator >::str() const U std::__1::__basic_string_common::__throw_length_error() const U std::__1::basic_string, std::__1::allocator >::resize(unsigned long, char) U std::__1::basic_string, std::__1::allocator >::push_back(char) U std::__1::basic_istream >::~basic_istream() U std::__1::basic_istream >::~basic_istream() U std::__1::basic_ostream >::~basic_ostream() U std::__1::basic_ostream >::~basic_ostream() U std::__1::basic_ostream >::operator<<(double) U std::__1::basic_iostream >::~basic_iostream() U std::__1::basic_iostream >::~basic_iostream() U std::__1::basic_iostream >::~basic_iostream() U std::__1::basic_streambuf >::sync() U std::__1::basic_streambuf >::imbue(std::__1::locale const&) U std::__1::basic_streambuf >::uflow() U std::__1::basic_streambuf >::setbuf(char*, long) U std::__1::basic_streambuf >::xsgetn(char*, long) U std::__1::basic_streambuf >::xsputn(char const*, long) U std::__1::basic_streambuf >::showmanyc() U std::__1::basic_streambuf >::basic_streambuf() U std::__1::basic_streambuf >::~basic_streambuf() 

+67 https://en.wikipedia.org/wiki/Name_mangling In compiler construction, name mangling (also called name decoration) is a technique used to solve various problems caused by the need to resolve unique names for programming entities in many modern programming languages. Name mangling Runtime Library 

define double @doMath() { Entry: %a = alloca double store double 5.500000e+00, double* %a %b = alloca double store double 1.450000e+01, double* %b %0 = load double, double* %a %1 = load double, double* %b %2 = fadd double %0, %1 %3 = fmul double %2, 5.000000e+01 %4 = fdiv double %3, 1.000000e+01 ret double %4 } define void @main() { Entry: %0 = call double @doMath() %1 = call i8* @_Z13number2stringd(double %0) call void @puts(i8* %1) ret void } declare void @puts(i8*) declare i8* @_Z13number2stringd(double) { function doMath(): number { const a = 5.5; const b = 14.5; return ((a + b) * 50) / 10; } puts(number2string(doMath())); } 

ovr@MBP-Dmitry  ~/projects/ovr/hlvm   master ✚ ● ?  ./main  ✔ hello ovr@MBP-Dmitry  ~/projects/ovr/hlvm   master ✚ ● ?   32 ↵ 

LLDB debuging may affect and broken JS developers, please skip this section you are going to be be alive. CONTENT DISCLAMER 

+71 __attribute__ ((visibility ("default"))) const char* number2string(double number) { char* result = new char[100]; sprintf(result, "%f", number); return result; } Runtime Library 

+72 ovr ~/hlvm   master  ./main  ✔ hello 100.000000 { function doMath(): number { const a = 5.5; const b = 14.5; return ((a + b) * 50) / 10; } puts("hello"); puts(_Z13number2stringd(doMath())); } Runtime Library 

ovr@MBP-Dmitry  hlvm  master  llvm-nm -C runtime/libhlvm-runtime.a ✔ runtime/libhlvm-runtime.a(helpers.cpp.o): 0000000000000000 T number2string(double) U operator new[](unsigned long) U _sprintf 

ovr  ~/projects/ovr/hlvm   master  otool -L main  ✔ main: /usr/lib/libc++.1.dylib (compatibility 1.0.0, current version 400.9.4) /usr/lib/libSystem.B.dylib (compatibility 1.0.0, current version 1252.200.5) clang -lstdc++ hello_world.o -lhlvm-runtime -L runtime -o main 

+75 «TypeScript» = StaticScript SSC *.ts BINARY TSC *.ts *.JS 

Стоп, но зачем ты писал компилятор? 

No content

+78 А теперь честно 

+79 Дмитрий Пацура https://github.com/ovr [email protected] https://telegram.me/ovrweb Спасибо за внимание! https://github.com/ovr/StaticScript