fix: Fix crash on non-ASCII version in dune-project

doc/changes/fixed/12794.md

@@ -0,0 +1,3 @@
+- Improve error message when invalid version strings are used in `dune-project`
+  files. Non-ASCII characters and malformed versions now show a helpful hint
+  with an example of the correct format. (#12794, fixes #12751, @benodiwal)

src/dune_sexp/versioned_file.ml

@@ -57,11 +57,24 @@ struct
 
     let parse first_line : Instance.t =
       let { First_line.lang = name_loc, name; version = ver_loc, ver } = first_line in
+      let ver_atom =
+        match Atom.parse ver with
+        | Some atom -> atom
+        | None ->
+          let hints =
+            match Table.find langs name with
+            | None -> []
+            | Some lang ->
+              let latest = Syntax.greatest_supported_version_exn lang.syntax in
+              [ Pp.textf "lang dune %s" (Syntax.Version.to_string latest) ]
+          in
+          User_error.raise
+            ~loc:ver_loc
+            ~hints
+            [ Pp.text "Invalid version. Version must be two numbers separated by a dot." ]
+      in
       let dune_lang_ver =
-        Decoder.parse
-          Syntax.Version.decode
-          Univ_map.empty
-          (Atom (ver_loc, Atom.of_string ver))
+        Decoder.parse Syntax.Version.decode Univ_map.empty (Atom (ver_loc, ver_atom))
       in
       match Table.find langs name with
       | None ->

test/blackbox-tests/test-cases/lang-invalid-version.t

@@ -5,27 +5,62 @@ such situations provide a clear error.
   >   cat > dune-project <<EOF
   > (lang dune $1)
   > EOF
-  >   dune build 2>&1 | grep "Internal error"
+  >   dune build
   > }
 
 Invalid version number:
 
   $ test_invalid_version "Ali"
+  File "dune-project", line 1, characters 11-14:
+  1 | (lang dune Ali)
+                 ^^^
+  Error: Atom of the form NNN.NNN expected
   [1]
 
 Test with various non-ASCII characters:
 
+CR-someday benodiwal: The version_loc is greedy and captures the closing
+parenthesis.
+
   $ test_invalid_version "è"
-  Internal error, please report upstream including the contents of _build/log.
+  File "dune-project", line 1, characters 11-13:
+  1 | (lang dune è)
+                 ^^
+  Error: Invalid version. Version must be two numbers separated by a dot.
+  Hint: lang dune 3.21
+  [1]
 
   $ test_invalid_version "π3.14"
-  Internal error, please report upstream including the contents of _build/log.
+  File "dune-project", line 1, characters 11-17:
+  1 | (lang dune π3.14)
+                 ^^^^^^
+  Error: Invalid version. Version must be two numbers separated by a dot.
+  Hint: lang dune 3.21
+  [1]
 
   $ test_invalid_version "α"
-  Internal error, please report upstream including the contents of _build/log.
+  File "dune-project", line 1, characters 11-13:
+  1 | (lang dune α)
+                 ^^
+  Error: Invalid version. Version must be two numbers separated by a dot.
+  Hint: lang dune 3.21
+  [1]
 
   $ test_invalid_version "😀"
-  Internal error, please report upstream including the contents of _build/log.
+  File "dune-project", line 1, characters 11-15:
+  1 | (lang dune 😀)
+                 ^^^^
+  Error: Invalid version. Version must be two numbers separated by a dot.
+  Hint: lang dune 3.21
+  [1]
+
+CR-someday benodiwal: Unicode string lengths are miscomputed in location
+excerpts for East Asian characters.
 
   $ test_invalid_version "中3.16文"
-  Internal error, please report upstream including the contents of _build/log.
+  File "dune-project", line 1, characters 11-21:
+  1 | (lang dune 中3.16文)
+                 ^^^^^^^^^^
+  Error: Invalid version. Version must be two numbers separated by a dot.
+  Hint: lang dune 3.21
+  [1]