Hi, I added a pass right after lower_eh_dispatch and before all_optimizations. In this pass, I'm trying to insert a new function in the call graph by calling cgraph_add_new_function, as the comments in cgraphunit.c mention that this function can be used at an arbitrary point of the middle-end. The comments also say that the new function will be enqueued for later compilation, avoiding the need for the passes to be reentrant.
When my pass executes and cgraph_add_new_function is called, cgraph_state's value is CGRAPH_STATE_EXPANSION. Thus, cgraph_add_new_function decides to execute the lowering and early local passes on the new function. When this happens, the state of some global variables, like dump_file and current_pass, are changed but not restored. As the original pass that called cgraph_add_new_function will need such global variables later, I am saving their current state and restoring them by hand after calling cgraph_add_new_function, otherwise the compilation fails. That being, I'd like to know if that is an indicator that I shouldn't be inserting a new function at this point of compilation, since cgraph_add_new_function wasn't supposed to affect the pass being executed. Thanks in advance for your help, -- Pericles Alves
