refactor(files): separate runtime I/O helper generation

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

1 files changed, 188 insertions, 0 deletions

diff --git a/lib/CodeGen/FileRuntimeIO.cc b/lib/CodeGen/FileRuntimeIO.cc
new file mode 100644
index 0000000..ffe5679
--- /dev/null
+++ b/lib/CodeGen/FileRuntimeIO.cc
@@ -0,0 +1,188 @@
+#include "cps/FileHandler.h"
+#include "cps/RuntimeCheck.h"
+#include "cps/ExceptionHandler.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
+
+using namespace llvm;
+using namespace cps;
+
+// ptr __cps_file_read(ptr name, i32 line): read one line into a fresh heap
+// STRING ('\n' consumed and stripped, trailing '\r' dropped). Reading at EOF
+// throws an exception — CIE programs are expected to guard with EOF().
+void FileHandler::buildReadFn() {
+    PointerType *PtrTy = PointerType::getUnqual(Context);
+    Type *I64 = Type::getInt64Ty(Context);
+    Type *I32 = Type::getInt32Ty(Context);
+    Type *I8 = Type::getInt8Ty(Context);
+
+    ReadFn = createHelper("__cps_file_read", FunctionType::get(PtrTy, {PtrTy, I32}, false));
+    Value *Name = ReadFn->getArg(0);
+    Value *Line = ReadFn->getArg(1);
+    Name->setName("name");
+    Line->setName("line");
+
+    Value *Idx = emitLookupOrFail(Name, Line);
+    Value *Mode = Builder.CreateLoad(I8, emitSlotGEP(FileModes, I8, Idx), "mode");
+    Value *NotRead = Builder.CreateICmpNE(Mode, ConstantInt::get(I8, uint8_t(FileMode::Read)), "not_read");
+    Checker.emitCheck(NotRead, ErrNotRead, {Line, Name});
+    Value *Handle = Builder.CreateLoad(PtrTy, emitSlotGEP(FileHandles, PtrTy, Idx), "handle");
+
+    Value *MinusOne = ConstantInt::get(Context, APInt(32, -1, /*isSigned=*/true));
+    Value *Peek = Builder.CreateCall(FgetcFunc, {Handle}, "peek");
+    Value *AtEof = Builder.CreateICmpEQ(Peek, MinusOne, "at_eof");
+    Checker.emitCheck(AtEof, ErrPastEof, {Line, Name});
+    Builder.CreateCall(UngetcFunc, {Peek, Handle});
+
+    // Growing buffer: cap doubles, always leaving room for byte + NUL.
+    AllocaInst *BufVar = Builder.CreateAlloca(PtrTy, nullptr, "buf");
+    AllocaInst *LenVar = Builder.CreateAlloca(I64, nullptr, "len");
+    AllocaInst *CapVar = Builder.CreateAlloca(I64, nullptr, "cap");
+    AllocaInst *ChVar = Builder.CreateAlloca(I32, nullptr, "ch");
+    Value *InitCap = ConstantInt::get(Context, APInt(64, 64));
+    Builder.CreateStore(Builder.CreateCall(MallocFunc, {InitCap}, "line_buf"), BufVar);
+    Builder.CreateStore(ConstantInt::get(Context, APInt(64, 0)), LenVar);
+    Builder.CreateStore(InitCap, CapVar);
+
+    BasicBlock *LoopBB = BasicBlock::Create(Context, "read_loop", ReadFn);
+    BasicBlock *EnsureBB = BasicBlock::Create(Context, "ensure_cap", ReadFn);
+    BasicBlock *GrowBB = BasicBlock::Create(Context, "grow", ReadFn);
+    BasicBlock *StoreBB = BasicBlock::Create(Context, "store_char", ReadFn);
+    BasicBlock *StripBB = BasicBlock::Create(Context, "strip", ReadFn);
+    BasicBlock *CheckCrBB = BasicBlock::Create(Context, "check_cr", ReadFn);
+    BasicBlock *DropCrBB = BasicBlock::Create(Context, "drop_cr", ReadFn);
+    BasicBlock *DoneBB = BasicBlock::Create(Context, "done", ReadFn);
+
+    Builder.CreateBr(LoopBB);
+
+    Builder.SetInsertPoint(LoopBB);
+    Value *Ch = Builder.CreateCall(FgetcFunc, {Handle}, "c");
+    Builder.CreateStore(Ch, ChVar);
+    Value *IsEof = Builder.CreateICmpEQ(Ch, MinusOne);
+    Value *IsNl = Builder.CreateICmpEQ(Ch, ConstantInt::get(Context, APInt(32, '\n')));
+    Value *Stop = Builder.CreateOr(IsEof, IsNl, "stop");
+    Builder.CreateCondBr(Stop, StripBB, EnsureBB);
+
+    Builder.SetInsertPoint(EnsureBB);
+    Value *Len = Builder.CreateLoad(I64, LenVar, "len");
+    Value *Cap = Builder.CreateLoad(I64, CapVar, "cap");
+    Value *Need = Builder.CreateAdd(Len, ConstantInt::get(Context, APInt(64, 2)), "need");
+    Value *TooSmall = Builder.CreateICmpSGT(Need, Cap, "too_small");
+    Builder.CreateCondBr(TooSmall, GrowBB, StoreBB);
+
+    Builder.SetInsertPoint(GrowBB);
+    Value *NewCap = Builder.CreateShl(Cap, ConstantInt::get(Context, APInt(64, 1)), "new_cap");
+    Builder.CreateStore(NewCap, CapVar);
+    Value *OldBuf = Builder.CreateLoad(PtrTy, BufVar);
+    Builder.CreateStore(Builder.CreateCall(ReallocFunc, {OldBuf, NewCap}, "regrown"), BufVar);
+    Builder.CreateBr(StoreBB);
+
+    Builder.SetInsertPoint(StoreBB);
+    Value *Buf = Builder.CreateLoad(PtrTy, BufVar);
+    Value *CurLen = Builder.CreateLoad(I64, LenVar);
+    Value *DestPtr = Builder.CreateInBoundsGEP(I8, Buf, CurLen);
+    Value *Byte = Builder.CreateTrunc(Builder.CreateLoad(I32, ChVar), I8);
+    Builder.CreateStore(Byte, DestPtr);
+    Builder.CreateStore(Builder.CreateAdd(CurLen, ConstantInt::get(Context, APInt(64, 1))), LenVar);
+    Builder.CreateBr(LoopBB);
+
+    // Normalize CRLF: drop one trailing '\r' (covers reading Windows-authored
+    // files on Linux; Windows text mode has already translated).
+    Builder.SetInsertPoint(StripBB);
+    Value *EndLen = Builder.CreateLoad(I64, LenVar, "end_len");
+    Value *HasChars = Builder.CreateICmpSGT(EndLen, ConstantInt::get(Context, APInt(64, 0)));
+    Builder.CreateCondBr(HasChars, CheckCrBB, DoneBB);
+
+    Builder.SetInsertPoint(CheckCrBB);
+    Value *EndBuf = Builder.CreateLoad(PtrTy, BufVar);
+    Value *LastIdx = Builder.CreateSub(EndLen, ConstantInt::get(Context, APInt(64, 1)));
+    Value *LastByte = Builder.CreateLoad(I8, Builder.CreateInBoundsGEP(I8, EndBuf, LastIdx), "last");
+    Value *IsCr = Builder.CreateICmpEQ(LastByte, ConstantInt::get(I8, '\r'), "is_cr");
+    Builder.CreateCondBr(IsCr, DropCrBB, DoneBB);
+
+    Builder.SetInsertPoint(DropCrBB);
+    Builder.CreateStore(LastIdx, LenVar);
+    Builder.CreateBr(DoneBB);
+
+    Builder.SetInsertPoint(DoneBB);
+    Value *FinalBuf = Builder.CreateLoad(PtrTy, BufVar);
+    Value *FinalLen = Builder.CreateLoad(I64, LenVar);
+    Builder.CreateStore(ConstantInt::get(I8, 0), Builder.CreateInBoundsGEP(I8, FinalBuf, FinalLen));
+    Builder.CreateRet(FinalBuf);
+}
+
+// void __cps_file_write(ptr name, ptr data, i32 line): one line, WRITE/APPEND only.
+void FileHandler::buildWriteFn() {
+    PointerType *PtrTy = PointerType::getUnqual(Context);
+    Type *I32 = Type::getInt32Ty(Context);
+    Type *I8 = Type::getInt8Ty(Context);
+
+    WriteFn = createHelper("__cps_file_write",
+                           FunctionType::get(Type::getVoidTy(Context), {PtrTy, PtrTy, I32}, false));
+    Value *Name = WriteFn->getArg(0);
+    Value *Data = WriteFn->getArg(1);
+    Value *Line = WriteFn->getArg(2);
+    Name->setName("name");
+    Data->setName("data");
+    Line->setName("line");
+
+    Value *Idx = emitLookupOrFail(Name, Line);
+    Value *Mode = Builder.CreateLoad(I8, emitSlotGEP(FileModes, I8, Idx), "mode");
+    Value *IsReadMode = Builder.CreateICmpEQ(Mode, ConstantInt::get(I8, uint8_t(FileMode::Read)), "read_only");
+    Checker.emitCheck(IsReadMode, ErrNotWrite, {Line, Name});
+
+    Value *Handle = Builder.CreateLoad(PtrTy, emitSlotGEP(FileHandles, PtrTy, Idx), "handle");
+    Builder.CreateCall(FputsFunc, {Data, Handle});
+    Builder.CreateCall(FputcFunc, {ConstantInt::get(Context, APInt(32, '\n')), Handle});
+    Builder.CreateRetVoid();
+}
+
+// void __cps_file_close(ptr name, i32 line)
+void FileHandler::buildCloseFn() {
+    PointerType *PtrTy = PointerType::getUnqual(Context);
+    Type *I32 = Type::getInt32Ty(Context);
+    Type *I8 = Type::getInt8Ty(Context);
+
+    CloseFn = createHelper("__cps_file_close",
+                           FunctionType::get(Type::getVoidTy(Context), {PtrTy, I32}, false));
+    Value *Name = CloseFn->getArg(0);
+    Value *Line = CloseFn->getArg(1);
+    Name->setName("name");
+    Line->setName("line");
+
+    Value *Idx = emitLookupOrFail(Name, Line);
+    Value *Handle = Builder.CreateLoad(PtrTy, emitSlotGEP(FileHandles, PtrTy, Idx), "handle");
+    Builder.CreateCall(FcloseFunc, {Handle});
+    Builder.CreateStore(ConstantInt::get(I8, 0), emitSlotGEP(FileModes, I8, Idx));
+    Builder.CreateStore(ConstantPointerNull::get(PtrTy), emitSlotGEP(FileNames, PtrTy, Idx));
+    Builder.CreateStore(ConstantPointerNull::get(PtrTy), emitSlotGEP(FileHandles, PtrTy, Idx));
+    Builder.CreateRetVoid();
+}
+
+// i1 __cps_file_eof(ptr name, i32 line): TRUE iff no more lines can be read.
+void FileHandler::buildEofFn() {
+    PointerType *PtrTy = PointerType::getUnqual(Context);
+    Type *I32 = Type::getInt32Ty(Context);
+    Type *I8 = Type::getInt8Ty(Context);
+
+    EofFn = createHelper("__cps_file_eof",
+                         FunctionType::get(Type::getInt1Ty(Context), {PtrTy, I32}, false));
+    Value *Name = EofFn->getArg(0);
+    Value *Line = EofFn->getArg(1);
+    Name->setName("name");
+    Line->setName("line");
+
+    Value *Idx = emitLookupOrFail(Name, Line);
+    Value *Mode = Builder.CreateLoad(I8, emitSlotGEP(FileModes, I8, Idx), "mode");
+    Value *NotRead = Builder.CreateICmpNE(Mode, ConstantInt::get(I8, uint8_t(FileMode::Read)), "not_read");
+    Checker.emitCheck(NotRead, ErrNotRead, {Line, Name});
+
+    Value *Handle = Builder.CreateLoad(PtrTy, emitSlotGEP(FileHandles, PtrTy, Idx), "handle");
+    Value *Ch = Builder.CreateCall(FgetcFunc, {Handle}, "c");
+    // ungetc(EOF, f) is a defined no-op, so no branch is needed.
+    Builder.CreateCall(UngetcFunc, {Ch, Handle});
+    Value *IsEof = Builder.CreateICmpEQ(Ch, ConstantInt::get(Context, APInt(32, -1, /*isSigned=*/true)), "is_eof");
+    Builder.CreateRet(IsEof);
+}