dvc: check if cache file inside dir was changed

Fixes #1125

Signed-off-by: Ruslan Kuprieiev <[email protected]>

dvc/remote/local.py

@@ -81,14 +81,35 @@ class RemoteLOCAL(RemoteBase):
         relpath = os.path.relpath(path, self.cache_dir)
         return os.path.dirname(relpath) + os.path.basename(relpath)
 
-    def changed_cache(self, md5):
+    def changed_cache_file(self, md5):
         cache = self.get(md5)
         if self.state.changed(cache, md5=md5):
             if os.path.exists(cache):
-                msg = 'Corrupted cache file {}'
+                msg = 'Corrupted cache file {}.'
                 Logger.warn(msg.format(os.path.relpath(cache)))
                 remove(cache)
             return True
+        return False
+
+    def changed_cache(self, md5):
+        cache = self.get(md5)
+        clist = [(cache, md5)]
+
+        while True:
+            if len(clist) == 0:
+                break
+
+            cache, md5 = clist.pop()
+            if self.changed_cache_file(md5):
+                return True
+
+            if not self.is_dir_cache(cache):
+                continue
+
+            for entry in self.load_dir_cache(md5):
+                md5 = entry[self.PARAM_MD5]
+                cache = self.get(md5)
+                clist.append((cache, md5))
 
         return False
 
@@ -512,7 +533,7 @@ class RemoteLOCAL(RemoteBase):
 
         progress.update_target(title, 90, 100)
 
-        local_exists = [not self.changed_cache(md5) for md5 in md5s]
+        local_exists = [not self.changed_cache_file(md5) for md5 in md5s]
 
         progress.finish_target(title)
 
@@ -539,7 +560,7 @@ class RemoteLOCAL(RemoteBase):
         names = []
         # NOTE: filter files that are not corrupted
         for md5, name in grouped:
-            if self.changed_cache(md5):
+            if self.changed_cache_file(md5):
                 md5s.append(md5)
                 names.append(name)
 
@@ -617,7 +638,7 @@ class RemoteLOCAL(RemoteBase):
 
         # NOTE: verifying that our cache is not corrupted
         def func(info):
-            return not self.changed_cache(info[self.PARAM_MD5])
+            return not self.changed_cache_file(info[self.PARAM_MD5])
         checksum_infos = list(filter(func, checksum_infos))
 
         progress.update_target(title, 20, 100)

tests/test_checkout.py

@@ -6,6 +6,7 @@ import shutil
 import filecmp
 
 from dvc.main import main
+from dvc.project import Project
 from tests.basic_env import TestDvc
 from tests.test_repro import TestRepro
 from dvc.stage import Stage
@@ -65,6 +66,32 @@ class TestCheckoutCorruptedCacheFile(TestRepro):
         self.assertFalse(os.path.isfile(cache))
 
 
+class TestCheckoutCorruptedCacheDir(TestDvc):
+    def test(self):
+        # NOTE: using 'copy' so that cache and link don't have same inode
+        ret = main(['config', 'cache.type', 'copy'])
+        self.assertEqual(ret, 0)
+
+        self.dvc = Project('.')
+        stages = self.dvc.add(self.DATA_DIR)
+        self.assertEqual(len(stages), 1)
+        self.assertEqual(len(stages[0].outs), 1)
+        out = stages[0].outs[0]
+
+        # NOTE: modifying cache file for one of the files inside the directory
+        # to check if dvc will detect that the cache is corrupted.
+        entry = self.dvc.cache.local.load_dir_cache(out.md5)[0]
+        md5 = entry[self.dvc.cache.local.PARAM_MD5]
+        cache = self.dvc.cache.local.get(md5)
+
+        with open(cache, 'w+') as fobj:
+            fobj.write('1')
+
+        self.dvc.checkout()
+
+        self.assertFalse(os.path.exists(cache))
+
+
 class TestCmdCheckout(TestCheckout):
     def test(self):
         ret = main(['checkout'])