Change to canonicalize only when reparse point in included

src/nvim/eval/funcs.c

@@ -6711,8 +6711,10 @@ static void f_resolve(typval_T *argvars, typval_T *rettv, FunPtr fptr)
 #ifdef WIN32
   char *v = os_resolve_shortcut(fname);
   if (v == NULL) {
+    if (os_is_reparse_point_include(fname)) {
       v = os_realpath(fname, v);
     }
+  }
   rettv->vval.v_string = (char_u *)(v == NULL ? xstrdup(fname) : v);
 #else
 # ifdef HAVE_READLINK

src/nvim/os/fs.c

@@ -1257,4 +1257,49 @@ shortcut_end:
   return rfname;
 }
 
+#define is_path_sep(c) ((c) == L'\\' || (c) == L'/')
+/// Returns true if the path contains a reparse point (junction or symbolic
+/// link). Otherwise false in returned.
+bool os_is_reparse_point_include(const char *path)
+{
+  wchar_t *p, *q, *utf16_path;
+  wchar_t buf[MAX_PATH];
+  DWORD attr;
+  bool result = false;
+
+  const int r = utf8_to_utf16(path, -1, &utf16_path);
+  if (r != 0) {
+    EMSG2("utf8_to_utf16 failed: %d", r);
+    return false;
+  }
+
+  p = utf16_path;
+  if (isalpha(p[0]) && p[1] == L':' && is_path_sep(p[2])) {
+    p += 3;
+  } else if (is_path_sep(p[0]) && is_path_sep(p[1])) {
+    p += 2;
+  }
+
+  while (*p != L'\0') {
+    q = wcspbrk(p, L"\\/");
+    if (q == NULL) {
+      p = q = utf16_path + wcslen(utf16_path);
+    } else {
+      p = q + 1;
+    }
+    if (q - utf16_path >= MAX_PATH) {
+      break;
+    }
+    wcsncpy(buf, utf16_path, (size_t)(q - utf16_path));
+    buf[q - utf16_path] = L'\0';
+    attr = GetFileAttributesW(buf);
+    if (attr != INVALID_FILE_ATTRIBUTES
+        && (attr & FILE_ATTRIBUTE_REPARSE_POINT) != 0) {
+      result = true;
+      break;
+    }
+  }
+  xfree(utf16_path);
+  return result;
+}
 #endif