Box traceback strings once

It had gotten to the point that a large part of the cost of throwing an
eception was allocating the std::string's to represent the function and
filename.

Most of the other strings have already been converted to being represented
as BoxedString's (ie python strings), so do that conversion here and a few
more places.

src/codegen/ast_interpreter.cpp

@@ -294,7 +294,7 @@ void ASTInterpreter::startJITing(CFGBlock* block, int exit_offset) {
         code_block = code_blocks[code_blocks.size() - 1].get();
 
     if (!code_block || code_block->shouldCreateNewBlock()) {
-        code_blocks.push_back(std::unique_ptr<JitCodeBlock>(new JitCodeBlock(source_info->getName())));
+        code_blocks.push_back(std::unique_ptr<JitCodeBlock>(new JitCodeBlock(source_info->getName()->s())));
         code_block = code_blocks[code_blocks.size() - 1].get();
         exit_offset = 0;
     }
@@ -334,7 +334,7 @@ Box* ASTInterpreter::execJITedBlock(CFGBlock* b) {
 
         auto source = getCL()->source.get();
         stmt->cxx_exception_count++;
-        caughtCxxException(LineInfo(stmt->lineno, stmt->col_offset, source->fn, source->getName()), &e);
+        caughtCxxException(LineInfo(stmt->lineno, stmt->col_offset, source->getFn(), source->getName()), &e);
 
         next_block = ((AST_Invoke*)stmt)->exc_dest;
         last_exception = e;
@@ -791,7 +791,7 @@ Value ASTInterpreter::visit_invoke(AST_Invoke* node) {
 
         auto source = getCL()->source.get();
         node->cxx_exception_count++;
-        caughtCxxException(LineInfo(node->lineno, node->col_offset, source->fn, source->getName()), &e);
+        caughtCxxException(LineInfo(node->lineno, node->col_offset, source->getFn(), source->getName()), &e);
 
         next_block = node->exc_dest;
         last_exception = e;
@@ -925,7 +925,7 @@ Value ASTInterpreter::visit_stmt(AST_stmt* node) {
 #endif
 
     if (0) {
-        printf("%20s % 2d ", source_info->getName().data(), current_block->idx);
+        printf("%20s % 2d ", source_info->getName()->c_str(), current_block->idx);
         print_ast(node);
         printf("\n");
     }

src/codegen/codegen.cpp

@@ -31,6 +31,7 @@
 #include "codegen/compvars.h"
 #include "core/ast.h"
 #include "core/util.h"
+#include "runtime/types.h"
 
 namespace pyston {
 
@@ -60,16 +61,18 @@ CLFunction::CLFunction(int num_args, int num_defaults, bool takes_varargs, bool
 }
 
 SourceInfo::SourceInfo(BoxedModule* m, ScopingAnalysis* scoping, FutureFlags future_flags, AST* ast,
-                       std::vector<AST_stmt*> body, std::string fn)
+                       std::vector<AST_stmt*> body, BoxedString* fn)
     : parent_module(m),
       scoping(scoping),
       scope_info(NULL),
       future_flags(future_flags),
       ast(ast),
       cfg(NULL),
-      fn(std::move(fn)),
       body(std::move(body)) {
-    assert(this->fn.size());
+    assert(fn->size());
+    // TODO: we should track this reference correctly rather than making it a root
+    gc::registerPermanentRoot(fn, true);
+    this->fn = fn;
 
     switch (ast->type) {
         case AST_TYPE::ClassDef:

src/codegen/irgen.cpp

@@ -934,18 +934,18 @@ static llvm::MDNode* setupDebugInfo(SourceInfo* source, llvm::Function* f, std::
 
     llvm::DIBuilder builder(*g.cur_module);
 
-    const std::string& fn = source->fn;
+    BoxedString* fn = source->getFn();
     std::string dir = "";
     std::string producer = "pyston; git rev " STRINGIFY(GITREV);
 
-    llvm::DIFile file = builder.createFile(fn, dir);
+    llvm::DIFile file = builder.createFile(fn->s(), dir);
     llvm::DITypeArray param_types = builder.getOrCreateTypeArray(llvm::None);
     llvm::DICompositeType func_type = builder.createSubroutineType(file, param_types);
     llvm::DISubprogram func_info = builder.createFunction(file, f->getName(), f->getName(), file, lineno, func_type,
                                                           false, true, lineno + 1, 0, true, f);
 
     llvm::DICompileUnit compile_unit
-        = builder.createCompileUnit(llvm::dwarf::DW_LANG_Python, fn, dir, producer, true, "", 0);
+        = builder.createCompileUnit(llvm::dwarf::DW_LANG_Python, fn->s(), dir, producer, true, "", 0);
 
     builder.finalize();
     return func_info;
@@ -967,7 +967,7 @@ static std::string getUniqueFunctionName(std::string nameprefix, EffortLevel eff
 
 CompiledFunction* doCompile(CLFunction* clfunc, SourceInfo* source, ParamNames* param_names,
                             const OSREntryDescriptor* entry_descriptor, EffortLevel effort,
-                            ExceptionStyle exception_style, FunctionSpecialization* spec, std::string nameprefix) {
+                            ExceptionStyle exception_style, FunctionSpecialization* spec, llvm::StringRef nameprefix) {
     Timer _t("in doCompile");
     Timer _t2;
     long irgen_us = 0;

src/codegen/irgen.h

@@ -113,7 +113,7 @@ bool isIsDefinedName(llvm::StringRef name);
 
 CompiledFunction* doCompile(CLFunction* clfunc, SourceInfo* source, ParamNames* param_names,
                             const OSREntryDescriptor* entry_descriptor, EffortLevel effort,
-                            ExceptionStyle exception_style, FunctionSpecialization* spec, std::string nameprefix);
+                            ExceptionStyle exception_style, FunctionSpecialization* spec, llvm::StringRef nameprefix);
 
 // A common pattern is to branch based off whether a variable is defined but only if it is
 // potentially-undefined.  If it is potentially-undefined, we have to generate control-flow

src/codegen/irgen/irgenerator.cpp

@@ -2961,7 +2961,7 @@ CLFunction* wrapFunction(AST* node, AST_arguments* args, const std::vector<AST_s
     CLFunction*& cl = made[node];
     if (cl == NULL) {
         std::unique_ptr<SourceInfo> si(
-            new SourceInfo(source->parent_module, source->scoping, source->future_flags, node, body, source->fn));
+            new SourceInfo(source->parent_module, source->scoping, source->future_flags, node, body, source->getFn()));
         if (args)
             cl = new CLFunction(args->args.size(), args->defaults.size(), args->vararg.s().size(),
                                 args->kwarg.s().size(), std::move(si));

src/codegen/unwinding.cpp

@@ -490,7 +490,7 @@ static const LineInfo lineInfoForFrame(PythonFrameIteratorImpl* frame_it) {
 
     auto source = cl->source.get();
 
-    return LineInfo(current_stmt->lineno, current_stmt->col_offset, source->fn, source->getName());
+    return LineInfo(current_stmt->lineno, current_stmt->col_offset, source->getFn(), source->getName());
 }
 
 // A class that converts a C stack trace to a Python stack trace.
@@ -1146,7 +1146,7 @@ std::string getCurrentPythonLine() {
 
         auto current_stmt = frame_iter->getCurrentStatement();
 
-        stream << source->fn << ":" << current_stmt->lineno;
+        stream << source->getFn()->c_str() << ":" << current_stmt->lineno;
         return stream.str();
     }
     return "unknown:-1";

src/core/cfg.cpp

@@ -2733,7 +2733,7 @@ CFG* computeCFG(SourceInfo* source, std::vector<AST_stmt*> body) {
         if (b->predecessors.size() == 0) {
             if (b != rtn->getStartingBlock()) {
                 rtn->print();
-                printf("%s\n", source->getName().data());
+                printf("%s\n", source->getName()->c_str());
             }
             ASSERT(b == rtn->getStartingBlock(), "%d", b->idx);
         }

src/core/types.h

@@ -304,6 +304,8 @@ class ScopeInfo;
 class InternedStringPool;
 class LivenessAnalysis;
 class SourceInfo {
+private:
+    BoxedString* fn; // equivalent of code.co_filename
 public:
     BoxedModule* parent_module;
     ScopingAnalysis* scoping;
@@ -312,7 +314,6 @@ public:
     AST* ast;
     CFG* cfg;
     bool is_generator;
-    std::string fn; // equivalent of code.co_filename
 
     InternedStringPool& getInternedStrings();
 
@@ -323,13 +324,15 @@ public:
     // body and we have to create one.  Ideally, we'd be able to avoid the space duplication for non-lambdas.
     const std::vector<AST_stmt*> body;
 
-    llvm::StringRef getName();
+    BoxedString* getName();
+    BoxedString* getFn() { return fn; }
+
     InternedString mangleName(InternedString id);
 
     Box* getDocString();
 
     SourceInfo(BoxedModule* m, ScopingAnalysis* scoping, FutureFlags future_flags, AST* ast,
-               std::vector<AST_stmt*> body, std::string fn);
+               std::vector<AST_stmt*> body, BoxedString* fn);
     ~SourceInfo();
 
 private:
@@ -727,9 +730,9 @@ void raiseSyntaxErrorHelper(llvm::StringRef file, llvm::StringRef func, AST* nod
 struct LineInfo {
 public:
     int line, column;
-    std::string file, func;
+    BoxedString* file, *func;
 
-    LineInfo(int line, int column, llvm::StringRef file, llvm::StringRef func)
+    LineInfo(int line, int column, BoxedString* file, BoxedString* func)
         : line(line), column(column), file(file), func(func) {}
 };
 

src/runtime/code.cpp

@@ -44,12 +44,12 @@ public:
 
     static Box* name(Box* b, void*) {
         RELEASE_ASSERT(b->cls == code_cls, "");
-        return boxString(static_cast<BoxedCode*>(b)->f->source->getName());
+        return static_cast<BoxedCode*>(b)->f->source->getName();
     }
 
     static Box* filename(Box* b, void*) {
         RELEASE_ASSERT(b->cls == code_cls, "");
-        return boxString(static_cast<BoxedCode*>(b)->f->source->fn);
+        return static_cast<BoxedCode*>(b)->f->source->getFn();
     }
 
     static Box* firstlineno(Box* b, void*) {

src/runtime/cxx_unwind.cpp

@@ -359,8 +359,8 @@ static void print_frame(unw_cursor_t* cursor, const unw_proc_info_t* pip) {
     if (frame_type == INTERPRETED && cf && cur_stmt) {
         auto source = cf->clfunc->source.get();
         // FIXME: dup'ed from lineInfoForFrame
-        LineInfo line(cur_stmt->lineno, cur_stmt->col_offset, source->fn, source->getName());
-        printf("      File \"%s\", line %d, in %s\n", line.file.c_str(), line.line, line.func.c_str());
+        LineInfo line(cur_stmt->lineno, cur_stmt->col_offset, source->getFn(), source->getName());
+        printf("      File \"%s\", line %d, in %s\n", line.file->c_str(), line.line, line.func->c_str());
     }
 }
 

src/runtime/exceptions.cpp

@@ -37,7 +37,7 @@ void raiseExc(Box* exc_obj) {
 void raiseSyntaxError(const char* msg, int lineno, int col_offset, llvm::StringRef file, llvm::StringRef func) {
     Box* exc = runtimeCall(SyntaxError, ArgPassSpec(1), boxString(msg), NULL, NULL, NULL, NULL);
 
-    auto tb = new BoxedTraceback(LineInfo(lineno, col_offset, file, func), None);
+    auto tb = new BoxedTraceback(LineInfo(lineno, col_offset, boxString(file), boxString(func)), None);
     assert(!PyErr_Occurred());
     throw ExcInfo(exc->cls, exc, tb);
 }
@@ -243,7 +243,8 @@ extern "C" void caughtCapiException(AST_stmt* stmt, void* _source_info) {
     SourceInfo* source = static_cast<SourceInfo*>(_source_info);
     PyThreadState* tstate = PyThreadState_GET();
 
-    exceptionAtLine(LineInfo(stmt->lineno, stmt->col_offset, source->fn, source->getName()), &tstate->curexc_traceback);
+    exceptionAtLine(LineInfo(stmt->lineno, stmt->col_offset, source->getFn(), source->getName()),
+                    &tstate->curexc_traceback);
 }
 
 extern "C" void reraiseCapiExcAsCxx() {

src/runtime/generator.cpp

@@ -468,7 +468,7 @@ Box* generatorName(Box* _self, void* context) {
     assert(isSubclass(_self->cls, generator_cls));
     BoxedGenerator* self = static_cast<BoxedGenerator*>(_self);
 
-    return boxString(self->function->f->source->getName());
+    return self->function->f->source->getName();
 }
 
 void generatorDestructor(Box* b) {

src/runtime/objmodel.cpp

@@ -2958,7 +2958,7 @@ static CompiledFunction* pickVersion(CLFunction* f, ExceptionStyle S, int num_ou
 
 static llvm::StringRef getFunctionName(CLFunction* f) {
     if (f->source)
-        return f->source->getName();
+        return f->source->getName()->s();
     else if (f->versions.size()) {
         return "<builtin function>";
         // std::ostringstream oss;

src/runtime/traceback.cpp

@@ -45,6 +45,9 @@ void BoxedTraceback::gcHandler(GCVisitor* v, Box* b) {
     if (self->tb_next)
         v->visit(self->tb_next);
 
+    v->visit(self->line.file);
+    v->visit(self->line.func);
+
     Box::gcHandler(v, b);
 }
 
@@ -59,12 +62,12 @@ void printTraceback(Box* b) {
 
     for (; tb && tb != None; tb = static_cast<BoxedTraceback*>(tb->tb_next)) {
         auto& line = tb->line;
-        fprintf(stderr, "  File \"%s\", line %d, in %s:\n", line.file.c_str(), line.line, line.func.c_str());
+        fprintf(stderr, "  File \"%s\", line %d, in %s:\n", line.file->c_str(), line.line, line.func->c_str());
 
         if (line.line < 0)
             continue;
 
-        FILE* f = fopen(line.file.c_str(), "r");
+        FILE* f = fopen(line.file->c_str(), "r");
         if (f) {
             assert(line.line < 10000000 && "Refusing to try to seek that many lines forward");
             for (int i = 1; i < line.line; i++) {
@@ -104,7 +107,7 @@ Box* BoxedTraceback::getLines(Box* b) {
         BoxedList* lines = new BoxedList();
         for (BoxedTraceback* wtb = tb; wtb && wtb != None; wtb = static_cast<BoxedTraceback*>(wtb->tb_next)) {
             auto& line = wtb->line;
-            auto l = BoxedTuple::create({ boxString(line.file), boxString(line.func), boxInt(line.line) });
+            auto l = BoxedTuple::create({ line.file, line.func, boxInt(line.line) });
             listAppendInternal(lines, l);
         }
         tb->py_lines = lines;

src/runtime/types.cpp

@@ -432,7 +432,7 @@ BoxedFunction::BoxedFunction(CLFunction* f, std::initializer_list<Box*> defaults
     // some builtin functions that are BoxedFunctions but really ought to be a type that
     // we don't have yet.
     if (f->source) {
-        this->name = static_cast<BoxedString*>(boxString(f->source->getName()));
+        this->name = static_cast<BoxedString*>(f->source->getName());
     }
 }
 

src/runtime/util.cpp

@@ -256,7 +256,7 @@ extern "C" void dumpEx(void* p, int levels) {
 
             CLFunction* cl = f->f;
             if (cl->source) {
-                printf("User-defined function '%s'\n", cl->source->getName().data());
+                printf("User-defined function '%s'\n", cl->source->getName()->c_str());
             } else {
                 printf("A builtin function\n");
             }

test/unittests/analysis.cpp

@@ -12,6 +12,7 @@
 #include "codegen/parser.h"
 #include "core/ast.h"
 #include "core/cfg.h"
+#include "runtime/types.h"
 #include "unittests.h"
 
 using namespace pyston;
@@ -39,7 +40,8 @@ TEST_F(AnalysisTest, augassign) {
 
     FutureFlags future_flags = getFutureFlags(module->body, fn.c_str());
 
-    SourceInfo* si = new SourceInfo(createModule("augassign", fn.c_str()), scoping, future_flags, func, func->body, fn);
+    SourceInfo* si = new SourceInfo(createModule("augassign", fn.c_str()), scoping, future_flags, func,
+            func->body, boxString(fn));
 
     CFG* cfg = computeCFG(si, func->body);
     std::unique_ptr<LivenessAnalysis> liveness = computeLivenessInfo(cfg);
@@ -69,7 +71,7 @@ void doOsrTest(bool is_osr, bool i_maybe_undefined) {
 
     ScopeInfo* scope_info = scoping->getScopeInfoForNode(func);
     std::unique_ptr<SourceInfo> si(new SourceInfo(createModule("osr" + std::to_string((is_osr << 1) + i_maybe_undefined),
-                    fn.c_str()), scoping, future_flags, func, func->body, fn));
+                    fn.c_str()), scoping, future_flags, func, func->body, boxString(fn)));
     CLFunction* clfunc = new CLFunction(0, 0, false, false, std::move(si));
 
     CFG* cfg = computeCFG(clfunc->source.get(), func->body);