summaryrefslogtreecommitdiffstats
path: root/vendor/proc-macro2/tests/comments.rs
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/proc-macro2/tests/comments.rs')
-rw-r--r--vendor/proc-macro2/tests/comments.rs103
1 files changed, 103 insertions, 0 deletions
diff --git a/vendor/proc-macro2/tests/comments.rs b/vendor/proc-macro2/tests/comments.rs
new file mode 100644
index 000000000..71741080c
--- /dev/null
+++ b/vendor/proc-macro2/tests/comments.rs
@@ -0,0 +1,103 @@
+use proc_macro2::{Delimiter, Literal, Spacing, TokenStream, TokenTree};
+
+// #[doc = "..."] -> "..."
+fn lit_of_outer_doc_comment(tokens: &TokenStream) -> Literal {
+ lit_of_doc_comment(tokens, false)
+}
+
+// #![doc = "..."] -> "..."
+fn lit_of_inner_doc_comment(tokens: &TokenStream) -> Literal {
+ lit_of_doc_comment(tokens, true)
+}
+
+fn lit_of_doc_comment(tokens: &TokenStream, inner: bool) -> Literal {
+ let mut iter = tokens.clone().into_iter();
+ match iter.next().unwrap() {
+ TokenTree::Punct(punct) => {
+ assert_eq!(punct.as_char(), '#');
+ assert_eq!(punct.spacing(), Spacing::Alone);
+ }
+ _ => panic!("wrong token {:?}", tokens),
+ }
+ if inner {
+ match iter.next().unwrap() {
+ TokenTree::Punct(punct) => {
+ assert_eq!(punct.as_char(), '!');
+ assert_eq!(punct.spacing(), Spacing::Alone);
+ }
+ _ => panic!("wrong token {:?}", tokens),
+ }
+ }
+ iter = match iter.next().unwrap() {
+ TokenTree::Group(group) => {
+ assert_eq!(group.delimiter(), Delimiter::Bracket);
+ assert!(iter.next().is_none(), "unexpected token {:?}", tokens);
+ group.stream().into_iter()
+ }
+ _ => panic!("wrong token {:?}", tokens),
+ };
+ match iter.next().unwrap() {
+ TokenTree::Ident(ident) => assert_eq!(ident.to_string(), "doc"),
+ _ => panic!("wrong token {:?}", tokens),
+ }
+ match iter.next().unwrap() {
+ TokenTree::Punct(punct) => {
+ assert_eq!(punct.as_char(), '=');
+ assert_eq!(punct.spacing(), Spacing::Alone);
+ }
+ _ => panic!("wrong token {:?}", tokens),
+ }
+ match iter.next().unwrap() {
+ TokenTree::Literal(literal) => {
+ assert!(iter.next().is_none(), "unexpected token {:?}", tokens);
+ literal
+ }
+ _ => panic!("wrong token {:?}", tokens),
+ }
+}
+
+#[test]
+fn closed_immediately() {
+ let stream = "/**/".parse::<TokenStream>().unwrap();
+ let tokens = stream.into_iter().collect::<Vec<_>>();
+ assert!(tokens.is_empty(), "not empty -- {:?}", tokens);
+}
+
+#[test]
+fn incomplete() {
+ assert!("/*/".parse::<TokenStream>().is_err());
+}
+
+#[test]
+fn lit() {
+ let stream = "/// doc".parse::<TokenStream>().unwrap();
+ let lit = lit_of_outer_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\" doc\"");
+
+ let stream = "//! doc".parse::<TokenStream>().unwrap();
+ let lit = lit_of_inner_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\" doc\"");
+
+ let stream = "/** doc */".parse::<TokenStream>().unwrap();
+ let lit = lit_of_outer_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\" doc \"");
+
+ let stream = "/*! doc */".parse::<TokenStream>().unwrap();
+ let lit = lit_of_inner_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\" doc \"");
+}
+
+#[test]
+fn carriage_return() {
+ let stream = "///\r\n".parse::<TokenStream>().unwrap();
+ let lit = lit_of_outer_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\"\"");
+
+ let stream = "/**\r\n*/".parse::<TokenStream>().unwrap();
+ let lit = lit_of_outer_doc_comment(&stream);
+ assert_eq!(lit.to_string(), "\"\\r\\n\"");
+
+ "///\r".parse::<TokenStream>().unwrap_err();
+ "///\r \n".parse::<TokenStream>().unwrap_err();
+ "/**\r \n*/".parse::<TokenStream>().unwrap_err();
+}