Increase time we spend in the interpreter

(increase the reopt and osr thresholds by 2x)

This helps django a ton.  will probably hurt our microbenchmarks
but I don't care much about those right now.

src/codegen/parser.cpp

@@ -955,7 +955,10 @@ static std::string getParserCommandLine(const char* fn) {
 
     llvm::sys::path::append(parse_ast_fn, "src/codegen/parse_ast.py");
 
-    return std::string("python -S ") + parse_ast_fn.str().str() + " " + fn;
+    // We may be running in an environment where "python" resolves to pyston (ex in
+    // a virtualenv), so try to hard code the path to CPython.
+    // This should probably be a configure-time check?
+    return std::string("/usr/bin/python -S ") + parse_ast_fn.str().str() + " " + fn;
 }
 
 AST_Module* parse_string(const char* code) {

src/core/options.cpp

@@ -42,8 +42,8 @@ bool ENABLE_PYPA_PARSER = true;
 bool USE_REGALLOC_BASIC = true;
 bool PAUSE_AT_ABORT = false;
 
-int OSR_THRESHOLD_INTERPRETER = 200;
-int REOPT_THRESHOLD_INTERPRETER = 100;
+int OSR_THRESHOLD_INTERPRETER = 500;
+int REOPT_THRESHOLD_INTERPRETER = 200;
 int OSR_THRESHOLD_BASELINE = 10000;
 int REOPT_THRESHOLD_BASELINE = 250;
 int OSR_THRESHOLD_T2 = 10000;

src/runtime/list.cpp

@@ -665,7 +665,7 @@ extern "C" void listIteratorGCHandler(GCVisitor* v, Box* b) {
 }
 
 Box* listNew(BoxedClass* cls, Box* container) {
-    assert(cls->cls == type_cls);
+    assert(isSubclass(cls->cls, type_cls));
     assert(isSubclass(cls, list_cls));
     return new (cls) BoxedList();
 }

test/tests/deopt_namescope_tests.py

@@ -6,6 +6,8 @@ try:
     import __pyston__
     __pyston__.setOption("OSR_THRESHOLD_BASELINE", 50)
     __pyston__.setOption("REOPT_THRESHOLD_BASELINE", 50)
+    __pyston__.setOption("OSR_THRESHOLD_INTERPRETER", 50)
+    __pyston__.setOption("REOPT_THRESHOLD_INTERPRETER", 50)
     __pyston__.setOption("SPECULATION_THRESHOLD", 10)
 except ImportError:
     pass