Changeset b81fd95 for src/ResolvExpr

Timestamp:

Jul 29, 2020, 11:29:01 PM (4 years ago)

Author:

Michael Brooks <mlbrooks@…>

Branches:

ADT, arm-eh, ast-experimental, enum, forall-pointer-decay, jacob/cs343-translation, master, new-ast, new-ast-unique-expr, pthread-emulation, qualifiedEnum

Children:

0c760db

Parents:

1d17939

Message:

Fix bug where pointer and reference types allow unsound initialization and return. Fixes #189

There are two instances of the same basic change, which is using conversionCost instead of castCost for resolving...
A: an InitExpr?, always; affects variable initializations
B: a CastExpr?, for type-system-generated casts only; affects function returns

Changing the behaviour of the typechecker on initialization (do A) and cast (do B):
src/ResolvExpr/AlternativeFinder.cc
src/SynTree/Expression.h
testsinit1.*

Making type of string literal consistent with how C defines it (accommodate A):
src/Parser/ExpressionNode.cc

Making type system happy with incumbent use of void* (accommodate A):
libcfa/src/concurrency/kernel.cfa
libcfa/src/containers/list.hfa
tests/bugs/66.cfa
tests/avltree/avl1.cfa
tests/concurrent/signal/block.cfa
tests/searchsort.cfa

Making type system happy with incumbent plan-9 downcast (accommodate B):
libcfa/src/containers/list.hfa

Fixing previously incorrect constness of declarations (accommodate A):
tests/exceptions/defaults.cfa
libcfa/src/iostream.hfa

Fixing previously incorrect isGenerated classification of casts that desugaring introduces (accommodate B):
src/Concurrency/Keywords.cc
src/Concurrency/Waitfor.cc

Working around trac #207 (revealed by A):
tests/io2.cfa

Working around trac #208 (speculatively created by B):
libcfa/src/bits/locks.hfa
libcfa/src/concurrency/preemption.cfa

Misc:
tests/exceptions/conditional.cfa (accommodate A)

a _msg function for an exception was declared with wrong return type, so it was not compatible for assignment into the vtable instance

libcfa/src/stdlib.hfa

the compiler now prohibits a prior attempt to call a nonexistent realloc overload; calling alloc_align in its place

File:

• src/ResolvExpr/AlternativeFinder.cc (modified) (2 diffs)

src/ResolvExpr/AlternativeFinder.cc

@@ -1216 +1216 @@
                         unify( castExpr->result, alt.expr->result, alt.env, needAssertions,
                                 haveAssertions, openVars, indexer );
-                        Cost thisCost = castCost( alt.expr->result, castExpr->result, alt.expr->get_lvalue(),
-                                indexer, alt.env );
+                        Cost thisCost =
+                                castExpr->isGenerated
+                                ? conversionCost( alt.expr->result, castExpr->result, alt.expr->get_lvalue(),   indexer, alt.env )
+                                : castCost( alt.expr->result, castExpr->result, alt.expr->get_lvalue(), indexer, alt.env );
                         PRINT(
                                 std::cerr << "working on cast with result: " << castExpr->result << std::endl;
@@ -1698 +1700 @@
 
                                 // unification run for side-effects
-                                unify( toType, alt.expr->result, newEnv, need, have, openVars, indexer );
+                                bool canUnify = unify( toType, alt.expr->result, newEnv, need, have, openVars, indexer );
+                                (void) canUnify;
                                 // xxx - do some inspecting on this line... why isn't result bound to initAlt.type?
 
-                                Cost thisCost = castCost( alt.expr->result, toType, alt.expr->get_lvalue(),
+                                Cost thisCost = computeConversionCost( alt.expr->result, toType, alt.expr->get_lvalue(),
                                         indexer, newEnv );
+
+                                PRINT(
+                                        Cost legacyCost = castCost( alt.expr->result, toType, alt.expr->get_lvalue(),
+                                                indexer, newEnv );
+                                        std::cerr << "Considering initialization:";
+                                        std::cerr << std::endl << "  FROM: "; alt.expr->result->print(std::cerr);
+                                        std::cerr << std::endl << "  TO: ";   toType          ->print(std::cerr);
+                                        std::cerr << std::endl << "  Unification " << (canUnify ? "succeeded" : "failed");
+                                        std::cerr << std::endl << "  Legacy cost " << legacyCost;
+                                        std::cerr << std::endl << "  New cost " << thisCost;
+                                        std::cerr << std::endl;
+                                )
+                                
                                 if ( thisCost != Cost::infinity ) {
                                         // count one safe conversion for each value that is thrown away