Fixed saver relative paths for latest_checkpoint

This would be cleaner if we made all paths listed in the "latest"
file relative to the its directory, allowing the removal of the
added `os.path.isabs` checks.

That would make the `os.join` in `saver.latest_checkpoint` much less
surprising.

But at least this way, there is no effect on currently working code.

Fixes #571
Change-Id: I47d8536b9b2ed3dcc193d6e6b7f4573a4e22c9b3

tensorflow/python/training/saver.py

@@ -489,6 +489,16 @@ def update_checkpoint_state(save_dir,
     all_model_checkpoint_paths.append(model_checkpoint_path)
   # Writes the "checkpoint" file for the coordinator for later restoration.
   coord_checkpoint_filename = _GetCheckpointFilename(save_dir, latest_filename)
+
+  # Relative paths need to be rewritten to be relative to the "save_dir".
+  if not os.path.isabs(model_checkpoint_path):
+    model_checkpoint_path = os.path.relpath(model_checkpoint_path, save_dir)
+
+  all_model_checkpoint_paths = [
+      os.path.relpath(p, save_dir) for p in all_model_checkpoint_paths
+      if not os.path.isabs(p)
+  ]
+
   if coord_checkpoint_filename == model_checkpoint_path:
     raise RuntimeError("Save path '%s' conflicts with path used for "
                        "checkpoint state.  Please use a different save path." %
@@ -854,6 +864,10 @@ def save(self, sess, save_path, global_step=None, latest_filename=None):
     """
     if latest_filename is None:
       latest_filename = "checkpoint"
+
+    if os.path.split(latest_filename)[0]:
+      raise ValueError("'latest_filename' must not contain path components")
+
     if global_step is not None:
       if not isinstance(global_step, compat.integral_types):
         global_step = training_util.global_step(sess, global_step)
@@ -905,8 +919,11 @@ def latest_checkpoint(checkpoint_dir, latest_filename=None):
   # Pick the latest checkpoint based on checkpoint state.
   ckpt = get_checkpoint_state(checkpoint_dir, latest_filename)
   if ckpt and ckpt.model_checkpoint_path:
+
+    # If you pass "os.path.join" two absolute paths it returns the second one.
     checkpoint_pattern = os.path.join(
         checkpoint_dir, ckpt.model_checkpoint_path)
+
     if gfile.Glob(checkpoint_pattern):
       return checkpoint_pattern
 

tensorflow/python/training/saver_test.py

@@ -20,6 +20,9 @@
 
 import os.path
 import time
+import contextlib
+import shutil
+import tempfile
 
 import tensorflow.python.platform
 
@@ -583,5 +586,73 @@ def testNonReshape(self):
       self.assertEqual(20.0, v1.eval())
 
 
+class LatestCheckpointWithRelativePaths(tf.test.TestCase):
+
+  @staticmethod
+  @contextlib.contextmanager
+  def tempWorkingDir(temppath):
+    cwd = os.getcwd()
+    os.chdir(temppath)
+    try:
+      yield
+    finally:
+      os.chdir(cwd)
+
+  @staticmethod
+  @contextlib.contextmanager
+  def tempDir():
+      tempdir = tempfile.mkdtemp()
+      try:
+        yield tempdir
+      finally:
+        shutil.rmtree(tempdir)
+
+  def testRelativePath(self):
+    # Make sure we have a clean directory to work in.
+    with self.tempDir() as tempdir:
+
+      # Jump to that directory until this test is done.
+      with self.tempWorkingDir(tempdir):
+
+        # Save training snapshots to a relative path.
+        traindir = 'train/'
+        os.mkdir(traindir)
+
+        filename = 'snapshot'
+        filepath = os.path.join(traindir, filename)
+
+        with self.test_session() as sess:
+          # Build a simple graph.
+          v0 = tf.Variable(0.0)
+          inc = v0.assign_add(1.0)
+
+          save = tf.train.Saver({'v0': v0})
+
+          # Record a short training history.
+          tf.initialize_all_variables().run()
+          save.save(sess, filepath, global_step=0)
+          inc.eval()
+          save.save(sess, filepath, global_step=1)
+          inc.eval()
+          save.save(sess, filepath, global_step=2)
+
+        with self.test_session() as sess:
+          # Build a new graph with different initialization.
+          v0 = tf.Variable(-1.0)
+
+          # Create a new saver.
+          save = tf.train.Saver({'v0': v0})
+          tf.initialize_all_variables().run()
+
+          # Get the most recent checkpoint name from the training history file.
+          name = tf.train.latest_checkpoint(traindir)
+          self.assertIsNotNone(name)
+
+          # Restore "v0" from that checkpoint.
+          save.restore(sess, name)
+          self.assertEquals(v0.eval(), 2.0)
+
+
+
 if __name__ == "__main__":
   tf.test.main()