# HG changeset patch # User Mario de Sousa # Date 1329751466 0 # Node ID eb1b28acec2e0dec406656ea57f7b00838a19cea # Parent 3c6225521059585dfc95b9120a136d212e87d6e5 Fix a few more bugs related to FB/function calls in IL. diff -r 3c6225521059 -r eb1b28acec2e stage3/fill_candidate_datatypes.cc --- a/stage3/fill_candidate_datatypes.cc Sun Feb 19 16:16:45 2012 +0000 +++ b/stage3/fill_candidate_datatypes.cc Mon Feb 20 15:24:26 2012 +0000 @@ -287,11 +287,20 @@ return; } + /* The value being passed to the 'param_name' parameter is actually the prev_il_instruction. + * However, we do not place that object directly in the fake il_param_list_c that we will be + * creating, since the visit(il_fb_call_c *) method will recursively call every object in that list. + * The il_prev_intruction object has already been visited. We DO NOT want to visit it again. + * The easiest way to work around this is to simply use a new object, and copy the relevant details to that object! + */ + symbol_c param_value; + copy_candidate_datatype_list(prev_il_instruction/*from*/, ¶m_value/*to*/); + identifier_c variable_name(param_name); // SYM_REF1(il_assign_operator_c, variable_name) il_assign_operator_c il_assign_operator(&variable_name); // SYM_REF3(il_param_assignment_c, il_assign_operator, il_operand, simple_instr_list) - il_param_assignment_c il_param_assignment(&il_assign_operator, prev_il_instruction/*il_operand*/, NULL); + il_param_assignment_c il_param_assignment(&il_assign_operator, ¶m_value/*il_operand*/, NULL); il_param_list_c il_param_list; il_param_list.add_element(&il_param_assignment); // SYM_REF4(il_fb_call_c, il_call_operator, fb_name, il_operand_list, il_param_list, symbol_c *called_fb_declaration) @@ -812,8 +821,11 @@ // void *visit(instruction_list_c *symbol); void *fill_candidate_datatypes_c::visit(il_instruction_c *symbol) { if (NULL == symbol->il_instruction) { - /* This object has (inherits) the same candidate datatypes as the prev_il_instruction */ - copy_candidate_datatype_list(symbol->prev_il_instruction /*from*/, symbol /*to*/); + /* This empty/null il_instruction does not change the value of the current/default IL variable. + * So it inherits the candidate_datatypes from it's previous IL instructions! + */ + if (NULL != symbol->prev_il_instruction) + copy_candidate_datatype_list(symbol->prev_il_instruction /*from*/, symbol /*to*/); } else { prev_il_instruction = symbol->prev_il_instruction; symbol->il_instruction->accept(*this); @@ -947,6 +959,9 @@ */ symbol->called_fb_declaration = fb_decl; + /* This object has the same candidate datatypes as the prev_il_instruction, since it does not change the value stored in the current/default IL variable. */ + copy_candidate_datatype_list(prev_il_instruction/*from*/, symbol/*to*/); + if (debug) std::cout << "FB [] ==> " << symbol->candidate_datatypes.size() << " result.\n"; return NULL; } diff -r 3c6225521059 -r eb1b28acec2e stage3/narrow_candidate_datatypes.cc --- a/stage3/narrow_candidate_datatypes.cc Sun Feb 19 16:16:45 2012 +0000 +++ b/stage3/narrow_candidate_datatypes.cc Mon Feb 20 15:24:26 2012 +0000 @@ -123,13 +123,13 @@ * may actually be calling an object of the il_instruction_c class. * If that is the case, that same il_instruction_c object will be called again inside the for() loop * of void *narrow_candidate_datatypes_c::visit(instruction_list_c *symbol); - * This is actually safe, as the narrow algorithm for all IL instructions is idem-potent. - * (It is easier to just let it be called twice than to hack the code to guarantee that it is - * only called once - this would also make this hacked code ugly an un-elegant, so we leave - * it as it is for now...) + * Since this is not safe (the prev_il_instruction variable will be overwritten with a wrong value!), + * we only do the recursive call if this parameter does not point to a il_instruction_c object. + * Actually, it is easier to check whether it is not the same as the prev_il_instruction. */ set_datatype(desired_datatype, call_param_value); - call_param_value->accept(*this); + if (call_param_value != prev_il_instruction) + call_param_value->accept(*this); if (NULL != param_name) if (extensible_parameter_highest_index < fp_iterator.extensible_param_index()) @@ -179,8 +179,15 @@ if ((NULL != param_name) && (NULL == desired_datatype)) ERROR; if ((NULL == param_name) && (NULL != desired_datatype)) ERROR; + /* set the desired data type for this parameter */ set_datatype(desired_datatype, call_param_value); - call_param_value->accept(*this); + /* And recursively call that parameter/expression, so it can propagate that info */ + /* However, when handling an implicit IL FB call, the first parameter is fake, and points to the prev_il_instruction. + * In this case, we do not propagate this info down, as that prev_il_instruction will be called later + * (remember, we iterate backwards through the IL instructions) by the for() loop in the instruction_list_c. + */ + if (call_param_value != prev_il_instruction) + call_param_value->accept(*this); if (NULL != param_name) if (extensible_parameter_highest_index < fp_iterator.extensible_param_index()) @@ -441,7 +448,9 @@ /*| instruction_list il_instruction */ // SYM_LIST(instruction_list_c) void *narrow_candidate_datatypes_c::visit(instruction_list_c *symbol) { - /* We need to go through the instructions backwards, so we can not use the base class' visitor */ + /* In order to execute the narrow algoritm correctly, we need to go through the instructions backwards, + * so we can not use the base class' visitor + */ for(int i = symbol->n-1; i >= 0; i--) { symbol->elements[i]->accept(*this); } @@ -457,41 +466,14 @@ if (NULL != symbol->prev_il_instruction) symbol->prev_il_instruction->datatype = symbol->datatype; } else { - prev_il_instruction = symbol->prev_il_instruction; /* Tell the il_instruction the datatype that it must generate - this was chosen by the next il_instruction (remember: we are iterating backwards!) */ symbol->il_instruction->datatype = symbol->datatype; + prev_il_instruction = symbol->prev_il_instruction; symbol->il_instruction->accept(*this); - } - return NULL; -} - - - - -/*************************************************************************************************/ -/* Important NOTE: */ -/* */ -/* The visit() methods for all the IL instructions must be idem-potent, as they may */ -/* potentially be called twice to narrow the same object. In other words, they may be called */ -/* to narrow an object that has already been previously narrowed. */ -/* This occurs when that IL instruction imediately precedes an IL non-formal function */ -/* invocation: */ -/* LD 45.5 */ -/* SIN */ -/* */ -/* In the above case, 'LD 45.5' will be narrowed once when the code that handles the */ -/* SIN function call */ -/* */ -/* narrow_nonformal_call(...), which is called by narrow_function_invocation(...), which is */ -/* in turn called by visit(il_function_call_c *) */ -/* */ -/* calls the call_param_value->accept(*this), where call_param_value will be a pointer */ -/* to the preceding IL instruction (in the above case, 'LD 45.5'). */ -/* */ -/* That same IL instruction will be again narrowed when called by the for() loop in */ -/* the visit(instruction_list_c *) visitor method. */ -/*************************************************************************************************/ - + prev_il_instruction = NULL; + } + return NULL; +} @@ -529,8 +511,7 @@ * the following code is actually correct! */ narrow_function_invocation(symbol, fcall_param); - /* set the desired datatype of the previous il instruction */ - prev_il_instruction->datatype = symbol->datatype; + /* The desired datatype of the previous il instruction was already set by narrow_function_invocation() */ return NULL; } @@ -574,7 +555,8 @@ * If we were to have the following line of code after calling arrow_[non]formal_call(), * we would then be overwriting the datatype with the wrong value! */ - prev_il_instruction->datatype = symbol->datatype; + if (NULL != prev_il_instruction) + prev_il_instruction->datatype = symbol->datatype; /* Note: We do not use the symbol->called_fb_declaration value (set in fill_candidate_datatypes_c) * because we try to identify any other datatype errors in the expressions used in the @@ -597,6 +579,29 @@ if (NULL != symbol->il_operand_list) narrow_nonformal_call(symbol, fb_decl); if (NULL != symbol-> il_param_list) narrow_formal_call(symbol, fb_decl); + /* NOTE: + * When handling these implicit IL calls, the parameter_value being passed to the FB parameter + * (in the previous example, the 'CU' parameter) is actually the prev_il_instruction. + * In this case, the prev_il_instruction->datatype will be set by the narrow_[non]formal_call(), + * using the param_value pointer to this same object. + * + * We must check that the datatype required by the IL instructions following this FB call + * is the same as that required for the first parameter. If not, then we have a semantic error, + * and we set it to NULL. + * + * However, we only do that if: + * - There really exists an il_prev_instruction + * (if it does not exist, it will be a semantic error. But that will be caught by the print_datatypes_error_c) + * - The IL instruction that comes after this IL FB call actually asked this FB call for a specific + * datatype in the current/default vairable, once this IL FB call returns. + * However, sometimes, (for e.g., this FB call is the last in the IL list) the subsequent FB to not aks this + * FB call for any datatype. In that case, then the datatype required to pass to the first parameter of the + * FB call must be left unchanged! + */ + if ((NULL != prev_il_instruction) && (NULL != symbol->datatype)) + if (!is_type_equal(prev_il_instruction->datatype, symbol->datatype)) { + prev_il_instruction->datatype = NULL; + } return NULL; } @@ -616,8 +621,7 @@ }; narrow_function_invocation(symbol, fcall_param); - /* set the desired datatype of the previous il instruction */ - prev_il_instruction->datatype = symbol->datatype; + /* The desired datatype of the previous il instruction was already set by narrow_function_invocation() */ return NULL; }