Handle more mixed warning lines from RBE
* Use android_root or __file__ path to find
the source tree root and its sub directories.
* Use the sub directory list to clean up RBE-returned
warning lines, removing all mixed prefix before
the top sub directories.
* Change some re.match rules and order to save
unnecessary comparisons.
Test: warn.py --url=http://cs/android --separator='?l=' build.log > warnings.html
Test: warn.py --gencsv build.log > warnings.csv
Bug: 198657613
Change-Id: I78cc17d04b8ab9e12935ef04797f3272298d5267
diff --git a/tools/warn/warn_common.py b/tools/warn/warn_common.py
index 61c8676..f24cfb7 100755
--- a/tools/warn/warn_common.py
+++ b/tools/warn/warn_common.py
@@ -228,6 +228,14 @@
return ''
+def find_android_root_top_dirs(root_dir):
+ """Return a list of directories under the root_dir, if it exists."""
+ if not os.path.isdir(root_dir):
+ return None
+ return list(filter(lambda d: os.path.isdir(root_dir + '/' + d),
+ os.listdir(root_dir)))
+
+
def find_android_root(buildlog):
"""Guess android source root from common prefix of file paths."""
# Use the longest common prefix of the absolute file paths
@@ -239,8 +247,8 @@
# We want to find android_root of a local build machine.
# Do not use RBE warning lines, which has '/b/f/w/' path prefix.
# Do not use /tmp/ file warnings.
- if warning_pattern.match(line) and (
- '/b/f/w' not in line and not line.startswith('/tmp/')):
+ if ('/b/f/w' not in line and not line.startswith('/tmp/') and
+ warning_pattern.match(line)):
warning_lines.append(line)
count += 1
if count > 9999:
@@ -251,15 +259,23 @@
path = os.path.normpath(re.sub(':.*$', '', line))
android_root = find_warn_py_and_android_root(path)
if android_root:
- return android_root
+ return android_root, find_android_root_top_dirs(android_root)
# Do not use common prefix of a small number of paths.
+ android_root = ''
if count > 10:
# pytype: disable=wrong-arg-types
root_path = os.path.commonprefix(warning_lines)
# pytype: enable=wrong-arg-types
if len(root_path) > 2 and root_path[len(root_path) - 1] == '/':
- return root_path[:-1]
- return ''
+ android_root = root_path[:-1]
+ if android_root and os.path.isdir(android_root):
+ return android_root, find_android_root_top_dirs(android_root)
+ # When the build.log file is moved to a different machine where
+ # android_root is not found, use the location of this script
+ # to find the android source tree root and its sub directories.
+ # This __file__ is /..../build/make/tools/warn/warn_common.py
+ script_root = __file__.replace('/build/make/tools/warn/warn_common.py', '')
+ return android_root, find_android_root_top_dirs(script_root)
def remove_android_root_prefix(path, android_root):
@@ -310,8 +326,6 @@
warning_pattern = re.compile(chrome_warning_pattern)
# Collect all unique warning lines
- # Remove the duplicated warnings save ~8% of time when parsing
- # one typical build log than before
unique_warnings = dict()
for line in infile:
if warning_pattern.match(line):
@@ -354,7 +368,7 @@
target_variant = 'unknown'
build_id = 'unknown'
use_rbe = False
- android_root = find_android_root(infile)
+ android_root, root_top_dirs = find_android_root(infile)
infile.seek(0)
# rustc warning messages have two lines that should be combined:
@@ -367,24 +381,39 @@
# C/C++ compiler warning messages have line and column numbers:
# some/path/file.c:line_number:column_number: warning: description
warning_pattern = re.compile('(^[^ ]*/[^ ]*: warning: .*)|(^warning: .*)')
- warning_without_file = re.compile('^warning: .*')
rustc_file_position = re.compile('^[ ]+--> [^ ]*/[^ ]*:[0-9]+:[0-9]+')
- # If RBE was used, try to reclaim some warning lines mixed with some
- # leading chars from other concurrent job's stderr output .
+ # If RBE was used, try to reclaim some warning lines (from stdout)
+ # that contain leading characters from stderr.
# The leading characters can be any character, including digits and spaces.
- # It's impossible to correctly identify the starting point of the source
- # file path without the file directory name knowledge.
- # Here we can only be sure to recover lines containing "/b/f/w/".
- rbe_warning_pattern = re.compile('.*/b/f/w/[^ ]*: warning: .*')
- # Collect all unique warning lines
- # Remove the duplicated warnings save ~8% of time when parsing
- # one typical build log than before
+ # If a warning line's source file path contains the special RBE prefix
+ # /b/f/w/, we can remove all leading chars up to and including the "/b/f/w/".
+ bfw_warning_pattern = re.compile('.*/b/f/w/([^ ]*: warning: .*)')
+
+ # When android_root is known and available, we find its top directories
+ # and remove all leading chars before a top directory name.
+ # We assume that the leading chars from stderr do not contain "/".
+ # For example,
+ # 10external/...
+ # 12 warningsexternal/...
+ # 413 warningexternal/...
+ # 5 warnings generatedexternal/...
+ # Suppressed 1000 warnings (packages/modules/...
+ if root_top_dirs:
+ extra_warning_pattern = re.compile(
+ '^.[^/]*((' + '|'.join(root_top_dirs) +
+ ')/[^ ]*: warning: .*)')
+ else:
+ extra_warning_pattern = re.compile('^[^/]* ([^ /]*/[^ ]*: warning: .*)')
+
+ # Collect all unique warning lines
unique_warnings = dict()
+ checked_warning_lines = dict()
line_counter = 0
prev_warning = ''
for line in infile:
+ line_counter += 1
if prev_warning:
if rustc_file_position.match(line):
# must be a rustc warning, combine 2 lines into one warning
@@ -399,14 +428,32 @@
prev_warning, flags, android_root, unique_warnings)
prev_warning = ''
- if use_rbe and rbe_warning_pattern.match(line):
- cleaned_up_line = re.sub('.*/b/f/w/', '', line)
- unique_warnings = add_normalized_line_to_warnings(
- cleaned_up_line, flags, android_root, unique_warnings)
+ # re.match is slow, with several warning line patterns and
+ # long input lines like "TIMEOUT: ...".
+ # We save significant time by skipping non-warning lines.
+ # But do not skip the first 100 lines, because we want to
+ # catch build variables.
+ if line_counter > 100 and line.find('warning: ') < 0:
continue
+ # A large clean build output can contain up to 90% of duplicated
+ # "warning:" lines. If we can skip them quickly, we can
+ # speed up this for-loop 3X to 5X.
+ if line in checked_warning_lines:
+ continue
+ checked_warning_lines[line] = True
+
+ # Clean up extra prefix if RBE is used.
+ if use_rbe:
+ if '/b/f/w/' in line:
+ result = bfw_warning_pattern.search(line)
+ else:
+ result = extra_warning_pattern.search(line)
+ if result is not None:
+ line = result.group(1)
+
if warning_pattern.match(line):
- if warning_without_file.match(line):
+ if line.startswith('warning: '):
# save this line and combine it with the next line
prev_warning = line
else:
@@ -416,7 +463,6 @@
if line_counter < 100:
# save a little bit of time by only doing this for the first few lines
- line_counter += 1
result = re.search('(?<=^PLATFORM_VERSION=).*', line)
if result is not None:
platform_version = result.group(0)