Chromium-recipe: add rmtree before copy for llvm_tools_to_copy.

scripts/slave/recipe_modules/archive/api.py

 # Copyright 2013 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 import re
 import sys
 
 import manual_bisect_files
 from recipe_engine import recipe_api
 
@@ skipping 252 matching lines @@
     cp_branch, cp_number = self.m.commit_position.parse(commit_position)
     build_git_commit = self._get_git_commit(update_properties, primary_project)
 
     if fixed_staging_dir:
       staging_dir = self.m.path['slave_build'].join('chrome_staging')
       self.m.file.rmtree('purge staging dir', staging_dir)
       self.m.file.makedirs('create staging dir', staging_dir)
     else:
       staging_dir = self.m.path.mkdtemp('chrome_staging')
 
+    llvm_tools_to_copy = ['llvm-symbolizer', 'sancov']
+    llvm_bin_dir = self.m.path['checkout'].join('third_party', 'llvm-build',
+                                                'Release+Asserts', 'bin')
+    ext = '.exe' if self.m.platform.is_win else ''
+
+    for tool in llvm_tools_to_copy:
+      tool_name = tool + ext
+      self.m.file.rmtree('Purge ' + tool,

[mmoroz, 2016/10/07 12:20:02]

I'm not sure that it will work fine if the file doesn't exist. I've tried to add
ignore_errors=True argument to the call, but got an error:

https://paste.googleplex.com/5613218135277568

[tandrii(chromium), 2016/10/07 12:23:31]

The error you got says "TypeError: Got extra keys while setting ConfigGroup:
{'ignore_errors': True}"

Are you sure your spelled kwarg correctly that it is accepted by rmtree?

[Oliver Chang, 2016/10/07 16:41:26]

Why rmtree, and not remove?

Also, did you figure out what was going wrong?

I see 2 errors (on lkgr only, and not on our libfuzzer bots):

shutil.Error:
`/b/c/b/ASAN_Release/src/third_party/llvm-build/Release+Asserts/bin/llvm-symbolizer`
and `/b/c/b/ASAN_Release/src/out/Release/llvm-symbolizer` are the same file

and

IOError: [Errno 2] No such file or directory:
'C:\\b\\c\\b\\Win_SyzyASAN_LKGR\\src\\third_party\\llvm-build\\Release+Asserts\\bin\\llvm-symbolizer.exe'

on syzyasan win bots.

The first error is weird, as I think that would imply the one in llvm-symbolizer
is a hard link of llvm-symbolizer. Not sure why it happens only on LKGR builds.

For the second one, I think we'll need to check if the binary exists before we
copy.

Maybe we should just add checks to make sure that the source exists, and the
destination doesn't, before doing the copy. The file recipe_module doesn't
actually pass your kwargs to the copy call:
https://cs.chromium.org/chromium/build/scripts/slave/recipe_modules/file/api.py

[mmoroz, 2016/10/07 16:50:22]

Basically, I was working on this CL while waiting for someone to LGTM the
revert. I thought that I could implement some good fix faster then get a LGTM
for the revert. Since the revert has been landed, I would be happy to delete
this CL and leave this problem up to you :)


rmtree() because I haven't figured out that there is remove() function :)

Adding checks sounds good to me.

+                         self.m.path.join(build_dir, tool_name))
+      self.m.file.copy('Copy ' + tool,
+                       self.m.path.join(llvm_bin_dir, tool_name), build_dir)
+
     # Build the list of files to archive.
     zip_file_list = [f for f in self.m.file.listdir('build_dir', build_dir)
                      if self._cf_should_package_file(f)]
 
     # Use the legacy platform name as Clusterfuzz has some expectations on
     # this (it only affects Windows, where it replace 'win' by 'win32').
     pieces = [self.legacy_platform_name(), target.lower()]
     if archive_subdir_suffix:
       pieces.append(archive_subdir_suffix)
     subdir = '-'.join(pieces)
@@ skipping 153 matching lines @@
   def legacy_download_url(self, gs_bucket_name, extra_url_components=None):
     """Returns a url suitable for downloading a Chromium build from
     Google Storage.
 
     extra_url_components, if specified, should be a string without a
     trailing '/' which is inserted in the middle of the URL.
 
     The builder_name, or parent_buildername, is always automatically
     inserted into the URL."""
     return self._legacy_url(True, gs_bucket_name, extra_url_components)