Cache and reuse 1-length strings

CPython does this and seems like a good thing to do.  Otherwise
iterating over a string would cause lots of allocations.

One difference between this implementation and cpython's is that cpython
doesn't eagerly create them, and so it checks to see if they've been
created or not.

src/runtime/inline/boxing.cpp

@@ -31,6 +31,11 @@ extern "C" Box* createList() {
 }
 
 BoxedString* boxString(llvm::StringRef s) {
+    if (s.size() <= 1) {
+        if (s.size() == 0)
+            return EmptyString;
+        return characters[s.data()[0] & UCHAR_MAX];
+    }
     return new (s.size()) BoxedString(s);
 }
 

src/runtime/str.cpp

@@ -54,6 +54,9 @@ extern "C" PyObject* string__format__(PyObject* self, PyObject* args) noexcept;
 
 namespace pyston {
 
+BoxedString* EmptyString;
+BoxedString* characters[UCHAR_MAX + 1];
+
 BoxedString::BoxedString(const char* s, size_t n) : interned_state(SSTATE_NOT_INTERNED) {
     RELEASE_ASSERT(n != llvm::StringRef::npos, "");
     if (s) {

src/runtime/types.cpp

@@ -739,7 +739,6 @@ BoxedClass* object_cls, *type_cls, *none_cls, *bool_cls, *int_cls, *float_cls,
       *builtin_function_or_method_cls, *attrwrapperiter_cls, *set_cls, *frozenset_cls;
 
 BoxedTuple* EmptyTuple;
-BoxedString* EmptyString;
 }
 
 extern "C" Box* createUserClass(BoxedString* name, Box* _bases, Box* _attr_dict) {
@@ -2529,8 +2528,16 @@ void setupRuntime() {
         BoxedHeapClass(object_cls, BoxedWrapperDescriptor::gcHandler, 0, 0, sizeof(BoxedWrapperDescriptor), false,
                        static_cast<BoxedString*>(boxString("wrapper_descriptor")));
 
-    EmptyString = boxString("");
-    gc::registerPermanentRoot(EmptyString);
+    EmptyString = new (0) BoxedString("");
+    // Call InternInPlace rather than InternFromString since that will
+    // probably try to return EmptyString
+    PyString_InternInPlace((Box**)&EmptyString);
+    for (int i = 0; i <= UCHAR_MAX; i++) {
+        char c = (char)i;
+        BoxedString* s = new (1) BoxedString(llvm::StringRef(&c, 1));
+        PyString_InternInPlace((Box**)&s);
+        characters[i] = s;
+    }
 
     // Kind of hacky, but it's easier to manually construct the mro for a couple key classes
     // than try to make the MRO construction code be safe against say, tuple_cls not having

src/runtime/types.h

@@ -614,6 +614,7 @@ public:
 
 extern "C" BoxedTuple* EmptyTuple;
 extern "C" BoxedString* EmptyString;
+extern BoxedString* characters[UCHAR_MAX + 1];
 
 struct PyHasher {
     size_t operator()(Box*) const;