fix: filetype detection freeze with binary files (#339)

* fix: filetype detection freeze with binary files

Filetype detection used to read the first line for shebang and the last
line for modeline detection. This was really slow with large binary
files that don't contain newlines.

Add Path:readbyterange(offset, length) to allow reading a specific range
and reimplement detect_from_modeline and detect_from_shebang to work
with the first 256 and last 256 bytes of a file.

Fixes #307 and closes #322.

* test: readbyterange

Author: dancek

lua/plenary/filetype.lua

@@ -130,19 +130,21 @@ filetype.detect_from_name = function(filepath)
 end
 
 filetype.detect_from_modeline = function(filepath)
-  local tail = Path:new(filepath):tail(1)
+  local tail = Path:new(filepath):readbyterange(-256, 256)
   if not tail then
     return ""
   end
-  return filetype._parse_modeline(tail)
+  local lines = vim.split(tail, "\n")
+  return filetype._parse_modeline(lines[#lines])
 end
 
 filetype.detect_from_shebang = function(filepath)
-  local head = Path:new(filepath):head(1)
+  local head = Path:new(filepath):readbyterange(0, 256)
   if not head then
     return ""
   end
-  return filetype._parse_shebang(head)
+  local lines = vim.split("", "\n")
+  return filetype._parse_shebang(lines[1])
 end
 
 --- Detect a filetype from a path.

lua/plenary/path.lua

@@ -844,4 +844,36 @@ function Path:iter()
   end
 end
 
+function Path:readbyterange(offset, length)
+  self = check_self(self)
+
+  local fd = uv.fs_open(self:expand(), "r", 438)
+  if not fd then
+    return
+  end
+  local stat = assert(uv.fs_fstat(fd))
+  if stat.type ~= "file" then
+    uv.fs_close(fd)
+    return nil
+  end
+
+  if offset < 0 then
+    offset = stat.size + offset
+  end
+
+  data = ""
+  while #data < length do
+    local read_chunk = assert(uv.fs_read(fd, length - #data, offset))
+    if #read_chunk == 0 then
+      break
+    end
+    data = data .. read_chunk
+    offset = offset + #read_chunk
+  end
+
+  assert(uv.fs_close(fd))
+
+  return data
+end
+
 return Path

tests/plenary/path_spec.lua

@@ -684,6 +684,22 @@ SOFTWARE.]]
       assert.are.same(should, data)
     end)
   end)
+
+  describe("readbyterange", function()
+    it("should read bytes at given offset", function()
+      local p = Path:new "LICENSE"
+      local data = p:readbyterange(13, 10)
+      local should = "Copyright "
+      assert.are.same(should, data)
+    end)
+
+    it("supports negative offset", function()
+      local p = Path:new "LICENSE"
+      local data = p:readbyterange(-10, 10)
+      local should = "SOFTWARE.\n"
+      assert.are.same(should, data)
+    end)
+  end)
 end)
 
 -- function TestPath:testIsDir()