On 8/26/21 13:04, Richard Biener wrote:
> Yes, and that flag_complex_method is initialized via the langhook
> mentioned, for example
> c-family/c-opts.c has
> 
> /* Initialize options structure OPTS.  */
> void
> c_common_init_options_struct (struct gcc_options *opts)
> {
>    opts->x_flag_exceptions = c_dialect_cxx ();
>    opts->x_warn_pointer_arith = c_dialect_cxx ();
>    opts->x_warn_write_strings = c_dialect_cxx ();
>    opts->x_flag_warn_unused_result = true;
> 
>    /* By default, C99-like requirements for complex multiply and divide.  */
>    opts->x_flag_complex_method = 2;
> }
> 
> so an attempt to "cancel" a command-line option that adjusted any of
> the above will not
> work because we're not re-initializing global_options appropriately.
> But maybe we can
> just do that?  That is, call
> 
>    /* Initialize global options structures; this must be repeated for
>       each structure used for parsing options.  */
>    init_options_struct (&global_options, &global_options_set);
>    lang_hooks.init_options_struct (&global_options);
> 
> and
> 
>    /* Perform language-specific options initialization.  */
>    lang_hooks.init_options (save_decoded_options_count, save_decoded_options);
> 
> as done by toplev.c?  Or if we do not want to do that store that state away
> to an 'initialized_options/initialized_options_set' set of vars we can
> copy from?

Hello.

I think the proper fix is handling the flag_complex_method-related options in finish_options.
Doing that, we can see 'optimize' pragma works with the current master.

Patch can bootstrap on x86_64-linux-gnu and survives regression tests.

Ready to be installed?
Thanks,
Martin