feat(input)!: validate INPUT and read whole lines into STRING

Scalar INPUT now reads one whitespace-delimited token through module-local
IR helpers instead of scanf, then skips spaces and tabs and consumes one
newline. INTEGER, REAL, BOOLEAN (TRUE or FALSE in any case) and CHAR are
validated, and bad input or end of input is a runtime error. STRING INPUT
reads the rest of the line. The unused scanf and REAL printf format globals
are removed.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

1 files changed, 328 insertions, 0 deletions

diff --git a/lib/CodeGen/InputGen.cc b/lib/CodeGen/InputGen.cc
new file mode 100644
index 0000000..349e7ce
--- /dev/null
+++ b/lib/CodeGen/InputGen.cc
@@ -0,0 +1,328 @@
+#include "cps/CodeGen.h"
+#include "cps/RuntimeCheck.h"
+#include "llvm/IR/Intrinsics.h"
+
+using namespace llvm;
+using namespace cps;
+
+namespace {
+// Defines (once per module) `ptr __cps_input_token()`: reads one whitespace-delimited token from
+// stdin into a malloc'd NUL-terminated buffer, or returns null at end of input. After the token it
+// skips spaces, tabs and carriage returns and consumes one newline if it follows, so the next
+// INPUT of a STRING starts on the following line. The caller frees the buffer.
+FunctionCallee getInputTokenHelper(Module &M) {
+    LLVMContext &Ctx = M.getContext();
+    Type *I8 = Type::getInt8Ty(Ctx), *I32 = Type::getInt32Ty(Ctx), *I64 = Type::getInt64Ty(Ctx);
+    PointerType *Ptr = PointerType::getUnqual(Ctx);
+    FunctionType *HelperTy = FunctionType::get(Ptr, {}, false);
+    if (Function *Existing = M.getFunction("__cps_input_token")) return {HelperTy, Existing};
+
+    FunctionCallee Getchar = M.getOrInsertFunction("getchar", FunctionType::get(I32, {}, false));
+    FunctionCallee Ungetc = M.getOrInsertFunction("ungetc", FunctionType::get(I32, {I32, Ptr}, false));
+    FunctionCallee Isspace = M.getOrInsertFunction("isspace", FunctionType::get(I32, {I32}, false));
+    FunctionCallee Realloc = M.getOrInsertFunction("realloc", FunctionType::get(Ptr, {Ptr, I64}, false));
+
+    Function *F = Function::Create(HelperTy, Function::InternalLinkage, "__cps_input_token", &M);
+    BasicBlock *Entry = BasicBlock::Create(Ctx, "entry", F);
+    BasicBlock *SkipHead = BasicBlock::Create(Ctx, "skip.head", F);
+    BasicBlock *SkipTest = BasicBlock::Create(Ctx, "skip.test", F);
+    BasicBlock *Append = BasicBlock::Create(Ctx, "append", F);
+    BasicBlock *Grow = BasicBlock::Create(Ctx, "grow", F);
+    BasicBlock *Store = BasicBlock::Create(Ctx, "store", F);
+    BasicBlock *ReadTest = BasicBlock::Create(Ctx, "read.test", F);
+    BasicBlock *Finish = BasicBlock::Create(Ctx, "finish", F);
+    BasicBlock *TrailHead = BasicBlock::Create(Ctx, "trail.head", F);
+    BasicBlock *TrailSkip = BasicBlock::Create(Ctx, "trail.skip", F);
+    BasicBlock *TrailEnd = BasicBlock::Create(Ctx, "trail.end", F);
+    BasicBlock *PushBack = BasicBlock::Create(Ctx, "pushback", F);
+    BasicBlock *Done = BasicBlock::Create(Ctx, "done", F);
+    BasicBlock *Eof = BasicBlock::Create(Ctx, "eof", F);
+
+    IRBuilder<> B(Entry);
+    Value *Buf = B.CreateAlloca(Ptr, nullptr, "buf");
+    Value *Len = B.CreateAlloca(I64, nullptr, "len");
+    Value *Cap = B.CreateAlloca(I64, nullptr, "cap");
+    Value *Cur = B.CreateAlloca(I32, nullptr, "cur");
+    B.CreateStore(ConstantPointerNull::get(Ptr), Buf);
+    B.CreateStore(ConstantInt::get(I64, 0), Len);
+    B.CreateStore(ConstantInt::get(I64, 0), Cap);
+    B.CreateBr(SkipHead);
+
+    // Leading whitespace.
+    B.SetInsertPoint(SkipHead);
+    Value *First = B.CreateCall(Getchar);
+    B.CreateStore(First, Cur);
+    B.CreateCondBr(B.CreateICmpEQ(First, ConstantInt::get(I32, -1)), Eof, SkipTest);
+    B.SetInsertPoint(SkipTest);
+    Value *IsSpace = B.CreateICmpNE(B.CreateCall(Isspace, {First}), ConstantInt::get(I32, 0));
+    B.CreateCondBr(IsSpace, SkipHead, Append);
+
+    // Append Cur to the buffer, keeping room for the byte and the terminator.
+    B.SetInsertPoint(Append);
+    Value *L = B.CreateLoad(I64, Len), *C = B.CreateLoad(I64, Cap);
+    B.CreateCondBr(B.CreateICmpUGE(B.CreateAdd(L, ConstantInt::get(I64, 2)), C), Grow, Store);
+    B.SetInsertPoint(Grow);
+    Value *NewCap = B.CreateSelect(B.CreateICmpEQ(C, ConstantInt::get(I64, 0)),
+                                   ConstantInt::get(I64, 32), B.CreateMul(C, ConstantInt::get(I64, 2)));
+    B.CreateStore(B.CreateCall(Realloc, {B.CreateLoad(Ptr, Buf), NewCap}), Buf);
+    B.CreateStore(NewCap, Cap);
+    B.CreateBr(Store);
+    B.SetInsertPoint(Store);
+    Value *StoreLen = B.CreateLoad(I64, Len);
+    Value *Slot = B.CreateGEP(I8, B.CreateLoad(Ptr, Buf), {StoreLen});
+    B.CreateStore(B.CreateTrunc(B.CreateLoad(I32, Cur), I8), Slot);
+    B.CreateStore(B.CreateAdd(StoreLen, ConstantInt::get(I64, 1)), Len);
+    Value *Next = B.CreateCall(Getchar);
+    B.CreateStore(Next, Cur);
+    B.CreateCondBr(B.CreateICmpEQ(Next, ConstantInt::get(I32, -1)), Finish, ReadTest);
+    B.SetInsertPoint(ReadTest);
+    Value *NextSpace = B.CreateICmpNE(B.CreateCall(Isspace, {Next}), ConstantInt::get(I32, 0));
+    B.CreateCondBr(NextSpace, Finish, Append);
+
+    // Terminate, then skip trailing spaces/tabs/CRs and one newline.
+    B.SetInsertPoint(Finish);
+    Value *EndLen = B.CreateLoad(I64, Len);
+    B.CreateStore(ConstantInt::get(I8, 0), B.CreateGEP(I8, B.CreateLoad(Ptr, Buf), {EndLen}));
+    B.CreateBr(TrailHead);
+    B.SetInsertPoint(TrailHead);
+    Value *Ch = B.CreateLoad(I32, Cur);
+    Value *Blank = B.CreateOr(B.CreateOr(B.CreateICmpEQ(Ch, ConstantInt::get(I32, ' ')),
+                                         B.CreateICmpEQ(Ch, ConstantInt::get(I32, '\t'))),
+                              B.CreateICmpEQ(Ch, ConstantInt::get(I32, '\r')));
+    B.CreateCondBr(Blank, TrailSkip, TrailEnd);
+    B.SetInsertPoint(TrailSkip);
+    B.CreateStore(B.CreateCall(Getchar), Cur);
+    B.CreateBr(TrailHead);
+    B.SetInsertPoint(TrailEnd);
+    Value *End = B.CreateLoad(I32, Cur);
+    Value *KeepsNothing = B.CreateOr(B.CreateICmpEQ(End, ConstantInt::get(I32, '\n')),
+                                     B.CreateICmpEQ(End, ConstantInt::get(I32, -1)));
+    B.CreateCondBr(KeepsNothing, Done, PushBack);
+    B.SetInsertPoint(PushBack);
+    // The stdin FILE* is a platform-specific symbol; host and target match.
+#if defined(_WIN32)
+    FunctionCallee IobFunc = M.getOrInsertFunction("__acrt_iob_func", FunctionType::get(Ptr, {I32}, false));
+    Value *Stdin = B.CreateCall(IobFunc, {ConstantInt::get(I32, 0)});
+#else
+#if defined(__APPLE__)
+    const char *StdinName = "__stdinp";
+#else
+    const char *StdinName = "stdin";
+#endif
+    GlobalVariable *StdinVar = M.getGlobalVariable(StdinName);
+    if (!StdinVar)
+        StdinVar = new GlobalVariable(M, Ptr, false, GlobalValue::ExternalLinkage, nullptr, StdinName);
+    Value *Stdin = B.CreateLoad(Ptr, StdinVar);
+#endif
+    B.CreateCall(Ungetc, {End, Stdin});
+    B.CreateBr(Done);
+    B.SetInsertPoint(Done);
+    B.CreateRet(B.CreateLoad(Ptr, Buf));
+    B.SetInsertPoint(Eof);
+    B.CreateRet(ConstantPointerNull::get(Ptr));
+    return {HelperTy, F};
+}
+
+// Defines `i1 __cps_parse_int(ptr token, ptr out)`: an optionally signed decimal integer that fills
+// the whole token and fits in INTEGER (what strtoll accepts in base 10, without errno).
+FunctionCallee getParseIntHelper(Module &M) {
+    LLVMContext &Ctx = M.getContext();
+    Type *I1 = Type::getInt1Ty(Ctx), *I8 = Type::getInt8Ty(Ctx), *I64 = Type::getInt64Ty(Ctx);
+    PointerType *Ptr = PointerType::getUnqual(Ctx);
+    FunctionType *HelperTy = FunctionType::get(I1, {Ptr, Ptr}, false);
+    if (Function *Existing = M.getFunction("__cps_parse_int")) return {HelperTy, Existing};
+
+    Function *F = Function::Create(HelperTy, Function::InternalLinkage, "__cps_parse_int", &M);
+    Value *Token = F->getArg(0), *Out = F->getArg(1);
+    BasicBlock *Entry = BasicBlock::Create(Ctx, "entry", F);
+    BasicBlock *Head = BasicBlock::Create(Ctx, "head", F), *Body = BasicBlock::Create(Ctx, "body", F);
+    BasicBlock *Tail = BasicBlock::Create(Ctx, "tail", F), *Negate = BasicBlock::Create(Ctx, "negate", F);
+    BasicBlock *Finish = BasicBlock::Create(Ctx, "finish", F), *Fail = BasicBlock::Create(Ctx, "fail", F);
+
+    IRBuilder<> B(Entry);
+    Value *Acc = B.CreateAlloca(I64, nullptr, "acc");
+    Value *Pos = B.CreateAlloca(Ptr, nullptr, "pos");
+    Value *Sign = B.CreateLoad(I8, Token);
+    Value *IsNeg = B.CreateICmpEQ(Sign, ConstantInt::get(I8, '-'));
+    Value *HasSign = B.CreateOr(IsNeg, B.CreateICmpEQ(Sign, ConstantInt::get(I8, '+')));
+    Value *Start = B.CreateGEP(I8, Token, {B.CreateZExt(HasSign, I64)});
+    B.CreateStore(ConstantInt::get(I64, 0), Acc);
+    B.CreateStore(Start, Pos);
+    B.CreateBr(Head);
+
+    // Accumulate negatively so INTEGER's minimum is representable.
+    B.SetInsertPoint(Head);
+    Value *P = B.CreateLoad(Ptr, Pos);
+    Value *Digit = B.CreateSub(B.CreateLoad(I8, P), ConstantInt::get(I8, '0'));
+    B.CreateCondBr(B.CreateICmpULT(Digit, ConstantInt::get(I8, 10)), Body, Tail);
+    B.SetInsertPoint(Body);
+    Value *Old = B.CreateLoad(I64, Acc);
+    Value *Mul = B.CreateBinaryIntrinsic(Intrinsic::smul_with_overflow, Old, ConstantInt::get(I64, 10));
+    Value *Sub = B.CreateBinaryIntrinsic(Intrinsic::ssub_with_overflow, B.CreateExtractValue(Mul, 0),
+                                         B.CreateZExt(Digit, I64));
+    Value *Overflow = B.CreateOr(B.CreateExtractValue(Mul, 1), B.CreateExtractValue(Sub, 1));
+    B.CreateStore(B.CreateExtractValue(Sub, 0), Acc);
+    B.CreateStore(B.CreateGEP(I8, P, {ConstantInt::get(I64, 1)}), Pos);
+    BasicBlock *Next = BasicBlock::Create(Ctx, "next", F);
+    B.CreateCondBr(Overflow, Fail, Next);
+    B.SetInsertPoint(Next);
+    B.CreateBr(Head);
+
+    B.SetInsertPoint(Tail);
+    Value *AtEnd = B.CreateICmpEQ(B.CreateLoad(I8, P), ConstantInt::get(I8, 0));
+    Value *HasDigits = B.CreateICmpNE(P, Start);
+    BasicBlock *Sign2 = BasicBlock::Create(Ctx, "sign", F);
+    B.CreateCondBr(B.CreateAnd(AtEnd, HasDigits), Sign2, Fail);
+    B.SetInsertPoint(Sign2);
+    Value *Magnitude = B.CreateLoad(I64, Acc);
+    B.CreateCondBr(IsNeg, Finish, Negate);
+    B.SetInsertPoint(Negate);
+    // +x needs -acc, which overflows only for INTEGER's minimum.
+    BasicBlock *Positive = BasicBlock::Create(Ctx, "positive", F);
+    B.CreateCondBr(B.CreateICmpEQ(Magnitude, ConstantInt::get(I64, APInt::getSignedMinValue(64))), Fail, Positive);
+    B.SetInsertPoint(Positive);
+    B.CreateStore(B.CreateNeg(Magnitude), Out);
+    B.CreateRet(ConstantInt::getTrue(Ctx));
+    B.SetInsertPoint(Finish);
+    B.CreateStore(Magnitude, Out);
+    B.CreateRet(ConstantInt::getTrue(Ctx));
+    B.SetInsertPoint(Fail);
+    B.CreateRet(ConstantInt::getFalse(Ctx));
+    return {HelperTy, F};
+}
+
+// Defines `i1 __cps_parse_real(ptr token, ptr out)`: the rule of parseNumber in pseudocode-lib/
+// string.cc (shared by IS_NUM and STR_TO_NUM): strtod must consume the whole token and must not
+// overflow to infinity. A finite overflow is told from an "inf" literal by the missing letter i.
+FunctionCallee getParseRealHelper(Module &M) {
+    LLVMContext &Ctx = M.getContext();
+    Type *I1 = Type::getInt1Ty(Ctx), *I8 = Type::getInt8Ty(Ctx), *I32 = Type::getInt32Ty(Ctx);
+    Type *DoubleTy = Type::getDoubleTy(Ctx);
+    PointerType *Ptr = PointerType::getUnqual(Ctx);
+    FunctionType *HelperTy = FunctionType::get(I1, {Ptr, Ptr}, false);
+    if (Function *Existing = M.getFunction("__cps_parse_real")) return {HelperTy, Existing};
+
+    FunctionCallee Strtod = M.getOrInsertFunction("strtod", FunctionType::get(DoubleTy, {Ptr, Ptr}, false));
+    FunctionCallee Strchr = M.getOrInsertFunction("strchr", FunctionType::get(Ptr, {Ptr, I32}, false));
+    Function *F = Function::Create(HelperTy, Function::InternalLinkage, "__cps_parse_real", &M);
+    Value *Token = F->getArg(0), *Out = F->getArg(1);
+    IRBuilder<> B(BasicBlock::Create(Ctx, "entry", F));
+    Value *EndSlot = B.CreateAlloca(Ptr, nullptr, "end");
+    Value *Number = B.CreateCall(Strtod, {Token, EndSlot});
+    Value *Consumed = B.CreateICmpEQ(B.CreateLoad(I8, B.CreateLoad(Ptr, EndSlot)), ConstantInt::get(I8, 0));
+    Value *Progress = B.CreateICmpNE(B.CreateLoad(Ptr, EndSlot), Token);
+    Value *IsInf = B.CreateFCmpOEQ(B.CreateUnaryIntrinsic(Intrinsic::fabs, Number),
+                                   ConstantFP::getInfinity(DoubleTy));
+    Value *NoLowerI = B.CreateIsNull(B.CreateCall(Strchr, {Token, ConstantInt::get(I32, 'i')}));
+    Value *NoUpperI = B.CreateIsNull(B.CreateCall(Strchr, {Token, ConstantInt::get(I32, 'I')}));
+    Value *Overflowed = B.CreateAnd(IsInf, B.CreateAnd(NoLowerI, NoUpperI));
+    B.CreateStore(Number, Out);
+    B.CreateRet(B.CreateAnd(B.CreateAnd(Consumed, Progress), B.CreateNot(Overflowed)));
+    return {HelperTy, F};
+}
+
+// Defines `i32 __cps_parse_bool(ptr token)`: 1 for TRUE, 0 for FALSE in any case, -1 otherwise.
+FunctionCallee getParseBoolHelper(Module &M) {
+    LLVMContext &Ctx = M.getContext();
+    Type *I8 = Type::getInt8Ty(Ctx), *I32 = Type::getInt32Ty(Ctx), *I64 = Type::getInt64Ty(Ctx);
+    PointerType *Ptr = PointerType::getUnqual(Ctx);
+    FunctionType *HelperTy = FunctionType::get(I32, {Ptr}, false);
+    if (Function *Existing = M.getFunction("__cps_parse_bool")) return {HelperTy, Existing};
+
+    FunctionCallee Strlen = M.getOrInsertFunction("strlen", FunctionType::get(I64, {Ptr}, false));
+    Function *F = Function::Create(HelperTy, Function::InternalLinkage, "__cps_parse_bool", &M);
+    Value *Token = F->getArg(0);
+    BasicBlock *Entry = BasicBlock::Create(Ctx, "entry", F);
+    BasicBlock *IsTrue = BasicBlock::Create(Ctx, "true", F), *IsFalse = BasicBlock::Create(Ctx, "false", F);
+    BasicBlock *Other = BasicBlock::Create(Ctx, "other", F), *Not4 = BasicBlock::Create(Ctx, "not4", F);
+    IRBuilder<> B(Entry);
+    Value *Length = B.CreateCall(Strlen, {Token});
+    // ORing 0x20 folds ASCII upper case to lower case; no other byte folds onto a letter.
+    auto Matches = [&](const char *Word) {
+        Value *All = ConstantInt::getTrue(Ctx);
+        for (unsigned I = 0; Word[I]; ++I) {
+            Value *Byte = B.CreateOr(B.CreateLoad(I8, B.CreateGEP(I8, Token, {ConstantInt::get(I64, I)})),
+                                     ConstantInt::get(I8, 0x20));
+            All = B.CreateAnd(All, B.CreateICmpEQ(Byte, ConstantInt::get(I8, Word[I])));
+        }
+        return All;
+    };
+    BasicBlock *Check4 = BasicBlock::Create(Ctx, "check4", F), *Check5 = BasicBlock::Create(Ctx, "check5", F);
+    B.CreateCondBr(B.CreateICmpEQ(Length, ConstantInt::get(I64, 4)), Check4, Not4);
+    B.SetInsertPoint(Check4);
+    B.CreateCondBr(Matches("true"), IsTrue, Other);
+    B.SetInsertPoint(Not4);
+    B.CreateCondBr(B.CreateICmpEQ(Length, ConstantInt::get(I64, 5)), Check5, Other);
+    B.SetInsertPoint(Check5);
+    B.CreateCondBr(Matches("false"), IsFalse, Other);
+    B.SetInsertPoint(IsTrue);
+    B.CreateRet(ConstantInt::get(I32, 1));
+    B.SetInsertPoint(IsFalse);
+    B.CreateRet(ConstantInt::get(I32, 0));
+    B.SetInsertPoint(Other);
+    B.CreateRet(ConstantInt::get(I32, -1));
+    return {HelperTy, F};
+}
+} // namespace
+
+// Reads one token from stdin and stores it, validated, at Addr. A malformed token or the end of
+// input is a runtime error. C scalars follow the same token model; integers must fit the C type.
+void CodeGen::emitScalarInput(Value *Addr, const TypeInfo *Info, int Line) {
+    Module &M = *TheModule;
+    Type *I32 = Type::getInt32Ty(*TheContext), *I64 = Type::getInt64Ty(*TheContext);
+    Type *DoubleTy = Type::getDoubleTy(*TheContext);
+    PointerType *Ptr = PointerType::getUnqual(*TheContext);
+    FunctionCallee Free = M.getOrInsertFunction("free", FunctionType::get(Type::getVoidTy(*TheContext), {Ptr}, false));
+    FunctionCallee Strlen = M.getOrInsertFunction("strlen", FunctionType::get(I64, {Ptr}, false));
+
+    Value *Token = Builder->CreateCall(getInputTokenHelper(M), {}, "input_token");
+    RuntimeChecker->emitCheck(Builder->CreateIsNull(Token),
+        Builder->CreateGlobalString("[Fatal] line %d: INPUT reached the end of input\n", "err_input_eof", 0, &M),
+        {RuntimeChecker->lineValue(Line)});
+    Value *TypeName = Builder->CreateGlobalString(Info->Name, "input_type_name", 0, &M);
+    Value *InvalidMsg = Builder->CreateGlobalString("[Fatal] line %d: Invalid %s input '%s'\n",
+                                                    "err_input_invalid", 0, &M);
+    auto Require = [&](Value *Valid) {
+        RuntimeChecker->emitCheck(Builder->CreateNot(Valid), InvalidMsg,
+                                  {RuntimeChecker->lineValue(Line), TypeName, Token});
+    };
+    Function *Fn = Builder->GetInsertBlock()->getParent();
+
+    if (Info->isChar()) {
+        Require(Builder->CreateICmpEQ(Builder->CreateCall(Strlen, {Token}), ConstantInt::get(I64, 1)));
+        Builder->CreateStore(Builder->CreateLoad(Info->LLVMType, Token), Addr);
+    } else if (Info->isBoolean() && !Info->isC()) {
+        Value *Parsed = Builder->CreateCall(getParseBoolHelper(M), {Token});
+        Require(Builder->CreateICmpSGE(Parsed, ConstantInt::get(I32, 0)));
+        Builder->CreateStore(Builder->CreateICmpNE(Parsed, ConstantInt::get(I32, 0)), Addr);
+    } else if (Info->isReal()) {
+        AllocaInst *Slot = CreateEntryBlockAlloca(Fn, DoubleTy, "input_real");
+        Require(Builder->CreateCall(getParseRealHelper(M), {Token, Slot}));
+        Value *Number = Builder->CreateLoad(DoubleTy, Slot);
+        Type *Target = Info->LLVMType;
+        if (Target->isFloatTy()) Number = Builder->CreateFPTrunc(Number, Target);
+        else if (Target != DoubleTy) Number = Builder->CreateFPExt(Number, Target);
+        Builder->CreateStore(Number, Addr);
+    } else {
+        AllocaInst *Slot = CreateEntryBlockAlloca(Fn, I64, "input_int");
+        Require(Builder->CreateCall(getParseIntHelper(M), {Token, Slot}));
+        Value *Number = Builder->CreateLoad(I64, Slot);
+        if (Info->isBoolean()) {
+            Builder->CreateStore(Builder->CreateICmpNE(Number, ConstantInt::get(I64, 0)), Addr);
+        } else {
+            unsigned Bits = Info->LLVMType->getIntegerBitWidth();
+            Value *Stored = Number;
+            if (Bits < 64) {
+                Type *Narrow = Info->LLVMType;
+                Stored = Builder->CreateTrunc(Number, Narrow);
+                Value *Back = Info->isC() && !Info->CSigned ? Builder->CreateZExt(Stored, I64)
+                                                            : Builder->CreateSExt(Stored, I64);
+                Require(Builder->CreateICmpEQ(Back, Number));
+            }
+            if (Info->isC() && !Info->CSigned)
+                Require(Builder->CreateICmpSGE(Number, ConstantInt::get(I64, 0)));
+            Builder->CreateStore(Stored, Addr);
+        }
+    }
+    Builder->CreateCall(Free, {Token});
+}