Add assertions to prevent truncation when using llvm::endian::Writer (#17266)

Also, stop serializing a few constant values, saving a tiny bit of
space in swiftmodule files.

Author: jrose-apple

lib/ClangImporter/SwiftLookupTable.cpp

@@ -999,6 +999,7 @@ namespace {
 
       // # of entries
       writer.write<uint16_t>(data.size());
+      assert(data.size() == static_cast<uint16_t>(data.size()));
 
       bool isModule = Writer.getLangOpts().isCompilingModule();
       for (auto &fullEntry : data) {

lib/Serialization/DeserializeSIL.cpp

@@ -111,8 +111,7 @@ class SILDeserializer::FuncTableInfo {
 
   static std::pair<unsigned, unsigned> ReadKeyDataLength(const uint8_t *&data) {
     unsigned keyLength = endian::readNext<uint16_t, little, unaligned>(data);
-    unsigned dataLength = endian::readNext<uint16_t, little, unaligned>(data);
-    return { keyLength, dataLength };
+    return { keyLength, sizeof(uint32_t) };
   }
 
   static internal_key_type ReadKey(const uint8_t *data, unsigned length) {
@@ -121,7 +120,7 @@ class SILDeserializer::FuncTableInfo {
 
   static data_type ReadData(internal_key_type key, const uint8_t *data,
                             unsigned length) {
-    assert(length == 4 && "Expect a single DeclID.");
+    assert(length == sizeof(uint32_t) && "Expect a single DeclID.");
     data_type result = endian::readNext<uint32_t, little, unaligned>(data);
     return result;
   }

lib/Serialization/ModuleFile.cpp

@@ -594,9 +594,8 @@ class ModuleFile::DeclMembersTableInfo {
   }
 
   static std::pair<unsigned, unsigned> ReadKeyDataLength(const uint8_t *&data) {
-    unsigned keyLength = endian::readNext<uint16_t, little, unaligned>(data);
     unsigned dataLength = endian::readNext<uint16_t, little, unaligned>(data);
-    return { keyLength, dataLength };
+    return { sizeof(uint32_t), dataLength };
   }
 
   static internal_key_type ReadKey(const uint8_t *data, unsigned length) {

lib/Serialization/Serialization.cpp

@@ -118,8 +118,11 @@ namespace {
       if (key.getKind() == DeclBaseName::Kind::Normal) {
         keyLength += key.getIdentifier().str().size(); // The name's length
       }
+      assert(keyLength == static_cast<uint16_t>(keyLength));
 
       uint32_t dataLength = (sizeof(uint32_t) + 1) * data.size();
+      assert(dataLength == static_cast<uint16_t>(dataLength));
+
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
       writer.write<uint16_t>(dataLength);
@@ -195,12 +198,14 @@ namespace {
                                                     key_type_ref key,
                                                     data_type_ref data) {
       uint32_t keyLength = key.str().size();
+      assert(keyLength == static_cast<uint16_t>(keyLength));
       uint32_t dataLength = (sizeof(uint32_t) * 2) * data.size();
       for (auto dataPair : data) {
         int32_t nameData = getNameDataForBase(dataPair.first);
         if (nameData > 0)
           dataLength += nameData;
       }
+      assert(dataLength == static_cast<uint16_t>(dataLength));
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
       writer.write<uint16_t>(dataLength);
@@ -242,9 +247,11 @@ namespace {
                                                     key_type_ref key,
                                                     data_type_ref data) {
       uint32_t keyLength = key.size();
+      assert(keyLength == static_cast<uint16_t>(keyLength));
       uint32_t dataLength = sizeof(uint32_t);
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
+      // No need to write the data length; it's constant.
       return { keyLength, dataLength };
     }
 
@@ -281,7 +288,9 @@ namespace {
                                                     key_type_ref key,
                                                     data_type_ref data) {
       uint32_t keyLength = key.str().size();
+      assert(keyLength == static_cast<uint16_t>(keyLength));
       uint32_t dataLength = (sizeof(uint32_t) * 2) * data.size();
+      assert(dataLength == static_cast<uint16_t>(dataLength));
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
       writer.write<uint16_t>(dataLength);
@@ -334,6 +343,7 @@ namespace {
       if (key.getKind() == DeclBaseName::Kind::Normal) {
         keyLength += key.getIdentifier().str().size(); // The name's length
       }
+      assert(keyLength == static_cast<uint16_t>(keyLength));
       uint32_t dataLength = sizeof(uint32_t);
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
@@ -387,12 +397,11 @@ namespace {
       // This will trap if a single ValueDecl has more than 16383 members
       // with the same DeclBaseName. Seems highly unlikely.
       assert((data.size() < (1 << 14)) && "Too many members");
-      uint32_t keyLength = sizeof(uint32_t); // key DeclID
       uint32_t dataLength = sizeof(uint32_t) * data.size(); // value DeclIDs
       endian::Writer<little> writer(out);
-      writer.write<uint16_t>(keyLength);
+      // No need to write the key length; it's constant.
       writer.write<uint16_t>(dataLength);
-      return { keyLength, dataLength };
+      return { sizeof(uint32_t), dataLength };
     }
 
     void EmitKey(raw_ostream &out, key_type_ref key, unsigned len) {

lib/Serialization/SerializeSIL.cpp

@@ -121,7 +121,7 @@ namespace {
       uint32_t dataLength = sizeof(uint32_t);
       endian::Writer<little> writer(out);
       writer.write<uint16_t>(keyLength);
-      writer.write<uint16_t>(dataLength);
+      // No need to write the data length; it's constant.
       return { keyLength, dataLength };
     }