luci-py: Tools for working with BSD style ar archives.

client/libs/ar/reader.py

Index: client/libs/ar/reader.py
diff --git a/client/libs/ar/reader.py b/client/libs/ar/reader.py
new file mode 100644
index 0000000000000000000000000000000000000000..16f8a752d6d5eccd1276cc4d3ea183f0dc93733e
--- /dev/null
+++ b/client/libs/ar/reader.py
@@ -0,0 +1,84 @@
+# Copyright 2016 The LUCI Authors. All rights reserved.
+# Use of this source code is governed under the Apache License, Version 2.0
+# that can be found in the LICENSE file.
+
+import os
+import struct
+
+from .writer import ArDefaultWriter
+
+def assert_eq(e, a):
+   """
+   >>> assert_eq(1, 1)
+   >>> assert_eq(1, 2)
+   Traceback (most recent call last):
+       ...
+   AssertionError: 1 (1) != 2 (2)
+   >>> assert_eq(1, "1")
+   Traceback (most recent call last):
+       ...
+   AssertionError: 1 (1) != 1 ('1')
+   """
+   assert e == a, "%s (%r) != %s (%r)" % (e, e, a, a)
+
+
+class ArDefaultReader(object):
+  """Read an ar archive created with default values."""
+
+  # BSD Variant AR file
+
+  def __init__(self, ibuf, check=False):
+    self.check = check
+    self.ibuf = ibuf
+    assert self.ibuf.read(8) == "!<arch>\n"
+
+  def __iter__(self):
+    # We generate the default values because comparing strings is cheaper then
+    # parsing them.
+    check = self.check

[M-A Ruel, 2016/06/09 21:50:44]

why?

[mithro, 2016/06/14 12:15:55]

Speed. Looking up check in self repeatedly is quite slow.

[M-A Ruel, 2016/06/14 13:26:16]

We do not use python -O in practice so I don't know if it is relevant. I guess
we could envision switching if you want to drive this effort.

+    if check:
+      dmodtime = "%-6i" % ArDefaultWriter.DEFAULT_MODTIME
+      duid = "%-6i" % ArDefaultWriter.DEFAULT_USER
+      dgid = "%-6i" % ArDefaultWriter.DEFAULT_GROUP
+      dmode = "%-6i" % ArDefaultWriter.DEFAULT_MODE
+
+    while True:
+      header = self.ibuf.read(60)
+      if header == '':

[M-A Ruel, 2016/06/09 21:50:44]

if not header:

[mithro, 2016/06/14 12:15:55]

Done.

+        return
+
+      name, modtime, uid, gid, mode, size, magic = struct.unpack(
+         "16s 12s 6s 6s 8s 10s 2s", header)
+
+      assert name.startswith("#1/"), name

[M-A Ruel, 2016/06/09 21:50:44]

use a real check + raise ValueError()

[mithro, 2016/06/14 12:15:55]

Python -O causes these asserts to totally disappear, which gives roughly a
10-15% speed up of this code.

I could catch the assertion error and convert it to ValueError?

+      filename_size = int(name[3:])
+      if check:
+         assert_eq(dmodtime, modtime)

[M-A Ruel, 2016/06/09 21:50:44]

same

[mithro, 2016/06/14 12:15:55]

See above.

+         assert_eq(duid, uid)
+         assert_eq(dgid, gid)
+         assert_eq(dmode, mode)
+      data_size = int(size)
+      assert_eq("\x60\n", magic)
+
+      file_size = data_size - filename_size
+      fp = self.ibuf.read(filename_size)
+
+      start = self.ibuf.tell()
+      yield fp, file_size, self.ibuf
+      end = self.ibuf.tell()
+
+      read = end - start
+      # If the reader didn't touch the input buffer, seek past the file.
+      if read == 0:
+        self.ibuf.seek(file_size, os.SEEK_CUR)
+      else:
+        assert read == file_size
+
+      if data_size % 2 != 0:
+        padding = self.ibuf.read(1)
+        assert padding == "\n"
+
+
+if __name__ == "__main__":
+  import doctest
+  doctest.testmod()