Shorten hash to avoid MAX_PATH limits

The Windows SDK contains some very long paths. When those are added to
the 40-character hash directory and the other path components it is easy
to hit the Windows 260-character MAX_PATH limit. In addition, the 40
character hash makes paths unwieldy in VsChromium search results and
elsewhere.

A ten character hash should be more than enough to avoid collisions - if
we upload a million toolchain packages then there will be about a 50%
chance of a name collision - we won't upload more than a thousand.

This was tested by copying the current toolchain file on Google storage
to a truncated name and then changing to that hash in vs_toolchain.py.
It was also necessary to copy the updates from depot_tools to
third_party\depot_tools. This means that we can't actually start using
short hashes until depot_tools has rolled.

We probably won't use a shorter hash until we next roll the toolchain,
just because changing the toolchain hash is mildly disruptive.

Bug: 1120785
Change-Id: I878b058857cbe9cb72a72b535864404eede33f3f
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/2376030
Reviewed-by: Sébastien Marchand <sebmarchand@chromium.org>
Commit-Queue: Bruce Dawson <brucedawson@chromium.org>

win_toolchain/get_toolchain_if_necessary.py

@@ -183,8 +183,12 @@ def CalculateHash(root, expected_hash):
       digest.update(f.read())
 
   # Save the timestamp file if the calculated hash is the expected one.
-  if digest.hexdigest() == expected_hash:
+  # The expected hash may be shorter, to reduce path lengths, in which case just
+  # compare that many characters.
+  if expected_hash and digest.hexdigest()[:len(expected_hash)] == expected_hash:
     SaveTimestampsAndHash(root, digest.hexdigest())
+    # Return the (potentially truncated) expected_hash.
+    return expected_hash
   return digest.hexdigest()
 
 

win_toolchain/package_from_installed.py

@@ -424,6 +424,10 @@ def RenameToSha1(output):
     zf.extractall(rel_dir)
   print('Hashing...')
   sha1 = get_toolchain_if_necessary.CalculateHash(rel_dir, None)
+  # Shorten from forty characters to ten. This is still enough to avoid
+  # collisions, while being less unwieldy and reducing the risk of MAX_PATH
+  # failures.
+  sha1 = sha1[:10]
   os.chdir(old_dir)
   shutil.rmtree(tempdir)
   final_name = sha1 + '.zip'