Fix #439

If memory serves, a patch like this was submitted to clap_derive,
but not to structopt. Oops.

structopt-derive/src/ty.rs

@@ -63,6 +63,8 @@ fn subty_if<F>(ty: &syn::Type, f: F) -> Option<&syn::Type>
 where
     F: FnOnce(&PathSegment) -> bool,
 {
+    let ty = strip_group(ty);
+
     only_last_segment(ty)
         .filter(|segment| f(segment))
         .and_then(|segment| {
@@ -96,6 +98,23 @@ pub fn is_simple_ty(ty: &syn::Type, name: &str) -> bool {
         .unwrap_or(false)
 }
 
+// If the struct is placed inside of a macro_rules! declaration,
+// in some circumstances, the tokens inside will be enclosed
+// in `proc_macro::Group` delimited by invisible `proc_macro::Delimiter::None`.
+//
+// In syn speak, this is encoded via `*::Group` variants. We don't really care about
+// that, so let's just strip it.
+//
+// Details: https://doc.rust-lang.org/proc_macro/enum.Delimiter.html#variant.None
+// See also: https://github.com/TeXitoi/structopt/issues/439
+fn strip_group(mut ty: &syn::Type) -> &syn::Type {
+    while let Type::Group(group) = ty {
+        ty = &*group.elem;
+    }
+
+    ty
+}
+
 fn is_generic_ty(ty: &syn::Type, name: &str) -> bool {
     subty_if_name(ty, name).is_some()
 }

tests/regressions.rs

@@ -0,0 +1,25 @@
+use structopt::StructOpt;
+
+mod utils;
+use utils::*;
+
+#[test]
+fn invisible_group_issue_439() {
+    macro_rules! m {
+        ($bool:ty) => {
+            #[derive(Debug, StructOpt)]
+            struct Opts {
+                #[structopt(long = "x")]
+                x: $bool
+            }
+        };
+    }
+
+    m!(bool);
+
+    let help = get_long_help::<Opts>();
+
+    assert!(help.contains("--x"));
+    assert!(!help.contains("--x <x>"));
+    Opts::from_iter_safe(&["test", "--x"]).unwrap();
+}