[Bug c++/111918] #pragma GCC diagnostic pop does not restore permerror status of -Wnarrowing

[fw at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111918

--- Comment #1 from Florian Weimer <fw at gcc dot gnu.org> ---
diagnostic_classify_diagnostic overwrites DK_UNSPECIFIED in
context->classify_diagnostic[OPT_Wnarrowing] with DK_WARNING here:

      /* Record the command-line status, so we can reset it back on DK_POP. */
      if (old_kind == DK_UNSPECIFIED)
        {
          old_kind = !context->option_enabled (option_index,
                                               context->lang_mask,
                                               context->option_state)
            ? DK_IGNORED : (context->warning_as_error_requested
                            ? DK_ERROR : DK_WARNING);
          context->classify_diagnostic[option_index] = old_kind;
        }

In diagnostic_enabled, we use that to switch the diagnostic kind to DK_WARNING
because update_effective_level_from_pragmas return DK_UNSPECIFIED:

  /* This tests for #pragma diagnostic changes.  */
  diagnostic_t diag_class
    = update_effective_level_from_pragmas (context, diagnostic);

  /* This tests if the user provided the appropriate -Werror=foo
     option.  */
  if (diag_class == DK_UNSPECIFIED
      && (context->classify_diagnostic[diagnostic->option_index]
          != DK_UNSPECIFIED))
    diagnostic->kind
      = context->classify_diagnostic[diagnostic->option_index];

We cannot treat permerrors differently in either function because they are an
emerging property, not associated with the warning option itself. This fixes
the test case, though:

  /* This tests if the user provided the appropriate -Werror=foo
     option.  */
  if (diag_class == DK_UNSPECIFIED
      && (context->classify_diagnostic[diagnostic->option_index] == DK_ERROR))
    diagnostic->kind = DK_ERROR;

I see what it does to the test suite.