libexpr/nix2: Completely remove URI literals

Change-Id: I8125e88fd5192f3830385b0960827c8a801c95ca

src/libexpr/parser/grammar2.hh

@@ -39,17 +39,6 @@ struct id_rest : sor<
     one<'_', '\'', '-'>
 > {};
 
-struct uri_scheme_first : ranges<'a', 'z', 'A', 'Z'> {};
-struct uri_scheme_rest : sor<
-    ranges<'a', 'z', 'A', 'Z', '0', '9'>,
-    one<'+', '-', '.'>
-> {};
-struct uri_sep : one<':'> {};
-struct uri_rest : sor<
-    ranges<'a', 'z', 'A', 'Z', '0', '9'>,
-    one<'%', '/', '?', ':', '@', '&', '=', '+', '$', ',', '-', '_', '.', '!', '~', '*', '\''>
-> {};
-
 }
 
 // "tokens". PEGs don't really care about tokens, we merely use them as a convenient
@@ -63,15 +52,8 @@ struct _extend_as_path : seq<
     c::path_sep,
     sor<c::path, TAO_PEGTL_STRING("${")>
 > {};
-struct _extend_as_uri : seq<
-    star<c::uri_scheme_rest>,
-    c::uri_sep,
-    c::uri_rest
-> {};
 
-// keywords might be extended to identifiers, paths, or uris.
-// NOTE this assumes that keywords are a-zA-Z only, otherwise uri schemes would never
-// match correctly.
+// keywords might be extended to identifiers or paths.
 // NOTE not a simple seq<...> because this would report incorrect positions for
 // keywords used inside must<> if a prefix of the keyword matches.
 template<typename S>
@@ -79,8 +61,7 @@ struct _keyword : sor<
     seq<
         S,
         not_at<c::id_rest>,
-        not_at<_extend_as_path>,
-        not_at<_extend_as_uri>
+        not_at<_extend_as_path>
     >,
     failure
 > {};
@@ -123,8 +104,6 @@ struct _not_at_any_keyword : minus<
 
 // identifiers are kind of horrid:
 //
-//   - uri_scheme_first ⊂ id_first
-//   - uri_scheme_first ⊂ uri_scheme_rest ⊂ path
 //   - id_first ⊂ id_rest ∖ { ' } ⊂ path
 //   - id_first ∩ (path ∖ uri_scheme_first) = { _ }
 //   - uri_sep ∉ ⋃ { id_first, id_rest, uri_scheme_first, uri_scheme_rest, path }
@@ -132,7 +111,7 @@ struct _not_at_any_keyword : minus<
 //
 // and we want, without reading the input more than once, a string that
 // matches (id_first id_rest*) and is not followed by any number of
-// characters such that the extended string matches path or uri rules.
+// characters such that the extended string matches path rules.
 //
 // since the first character must be either _ or a uri scheme character
 // we can ignore path-like bits at the beginning. uri_sep cannot appear anywhere
@@ -148,14 +127,7 @@ struct _not_at_any_keyword : minus<
 struct identifier : _not_at_any_keyword<
     // we don't use (at<id_rest>, ...) matches here because identifiers are
     // a really hot path and rewinding as needed by at<> isn't entirely free.
-    sor<
-        seq<
-            c::uri_scheme_first,
-            star<ranges<'a', 'z', 'A', 'Z', '0', '9', '-'>>,
-            not_at<_extend_as_uri>
-        >,
-        one<'_'>
-    >,
+    c::id_first,
     star<sor<ranges<'a', 'z', 'A', 'Z', '0', '9'>, one<'_', '-'>>>,
     not_at<_extend_as_path>,
     star<c::id_rest>
@@ -176,13 +148,6 @@ struct floating : seq<
     not_at<_extend_as_path>
 > {};
 
-struct uri : seq<
-    c::uri_scheme_first,
-    star<c::uri_scheme_rest>,
-    c::uri_sep,
-    plus<c::uri_rest>
-> {};
-
 struct sep : sor<
     plus<one<' ', '\t', '\r', '\n'>>,
     seq<one<'#'>, star<not_one<'\r', '\n'>>>,
@@ -479,7 +444,6 @@ struct _expr {
     struct string : semantic, seq<grammar::v2::string> {};
     struct ind_string : semantic, seq<grammar::v2::ind_string> {};
     struct path : semantic, seq<grammar::v2::path> {};
-    struct uri : semantic, t::uri {};
     struct rec_set : semantic, _attrset<must, t::kw_rec, seps> {};
     struct set : semantic, _attrset<> {};
 
@@ -499,7 +463,6 @@ struct _expr {
         string,
         ind_string,
         path,
-        uri,
         seq<one<'('>, seps, must<expr>, seps, must<one<')'>>>,
         rec_set,
         set,

src/libexpr/parser/parser-impl2.inc.cc

@@ -640,18 +640,6 @@ template<> struct BuildAST<grammar::v2::expr::string> : p::maybe_nothing {};
 template<> struct BuildAST<grammar::v2::expr::ind_string> : p::maybe_nothing {};
 template<> struct BuildAST<grammar::v2::expr::path> : p::maybe_nothing {};
 
-template<> struct BuildAST<grammar::v2::expr::uri> {
-    static void apply(const auto & in, ExprState & s, State & ps) {
-       bool URLLiterals = ps.featureSettings.isEnabled(Dep::UrlLiterals);
-       if (!URLLiterals)
-           throw ParseError({
-               .msg = HintFmt("URL literals are deprecated, allow using them with --extra-deprecated-features=url-literals"),
-               .pos = ps.positions[ps.at(in)]
-           });
-       s.pushExpr<ExprString>(ps.at(in), in.string());
-    }
-};
-
 template<> struct BuildAST<grammar::v2::expr::rec_set> : change_head<BindingsState> {
     static void success(const auto & in, BindingsState & b, ExprState & s, State & ps) {
         // Before inserting new attrs, check for __override and throw an error

src/libutil/experimental-features.cc

@@ -187,6 +187,7 @@ constexpr std::array<ExperimentalFeatureDetails, numXpFeatures> xpFeatureDetails
             Notable changes done compared to the stable Nix language:
 
             - Operators cannot be overridden anymore through `__sub` and friends
+            - URL literals are completely removed from the language. `x:x` is now a function. See also the `url-literals` deprecated feature
             - Floating point literals must start with a digit, literals like `.3` are forbidden
             - Removed ancient `let {` syntax. See also the `ancient-let` deprecated feature
         )",

tests/unit/libexpr/trivial.cc

@@ -103,6 +103,22 @@ namespace nix {
         ASSERT_THROW(eval("https://nixos.org"), Error);
     }
 
+    TEST_F(TrivialExpressionTest, smallestLambda) {
+        ASSERT_THROW(eval("x:x"), Error);
+
+        FeatureSettings mockFeatureSettings;
+        mockFeatureSettings.set("deprecated-features", "url-literals");
+
+        auto v = eval("x:x", true, mockFeatureSettings);
+        ASSERT_THAT(v, IsStringEq("x:x"));
+
+        // With URI literals now fully removed instead of merely throwing an error,
+        // the value `x:x` now becomes the lambda it should always have been
+        mockFeatureSettings.set("experimental-features", "nix-lang2");
+        v = eval("(x:x) 42", true, mockFeatureSettings);
+        ASSERT_THAT(v, IsIntEq(42));
+    }
+
     TEST_F(TrivialExpressionTest, withFound) {
         auto v = eval("with { a = 23; }; a");
         ASSERT_THAT(v, IsIntEq(23));