SanitizerReports

Introduction

SanitizerReports aims to help you grab various reports from sanitizers when running their unit tests and regression tests, by patching lit (LLVM Integrated Tester) and FileCheck(Flexible pattern matching file verifier).

We also provide some report texts as examples. See ./reports for more details.

SanitizerReports is under the Apache License v2.0 with LLVM Exceptions (same as llvm/llvm-project). See LICENSE for more details.

Tutorial

We are going to take LLVM 16.0.0 Release working on Ubuntu 20.04.4 LTS (x86_64) as an example. The same method is also applicable to other versions of llvm-project and other OS.

Check Requirements

See https://releases.llvm.org/16.0.0/docs/GettingStarted.html#requirements .

Get the Source Code

Download the whole source tree from https://github.com/llvm/llvm-project/releases/tag/llvmorg-16.0.0 .

Remember we need source code of all related subprojects like Clang, so download https://github.com/llvm/llvm-project/releases/download/llvmorg-16.0.0/llvm-project-16.0.0.src.tar.xz .

You can extract the source code with

tar -xf llvm-project-16.0.0.src.tar.xz

after downloaded. Then you will get a directory llvm-project-16.0.0.src.

Configure LLVM

This may sound confusing, but it's the fact - we actually need to start all building stuffs from llvm-project-16.0.0.src/llvm.

cd llvm-project-16.0.0.src;

mkdir mybuild myinstall;

cd mybuild;

cmake -G "Unix Makefiles" \
    -DLLVM_ENABLE_PROJECTS="compiler-rt;clang" \
    -DLLVM_INCLUDE_TESTS=ON \
    -DLLVM_TARGETS_TO_BUILD="X86" \
    -DCMAKE_BUILD_TYPE="Release" \
    -DCMAKE_INSTALL_PREFIX="`pwd`/../myinstall" \
    -DCMAKE_C_COMPILER="/usr/bin/gcc" \
    -DCMAKE_CXX_COMPILER="/usr/bin/g++" \
    "`pwd`/../llvm";

Here we enable both compiler-rt and clang, which is necessary for running the test cases. See https://releases.llvm.org/16.0.0/docs/GettingStarted.html#local-llvm-configuration for more details.

Build LLVM

Assuming that we are still in llvm-project-16.0.0.src/mybuild, run

make -j$(nproc)

and wait for all things get done.

Apply Patches

If it has built successfully, let's go to llvm-project-16.0.0.src/mybuild/bin for this step.

Patch FileCheck

Make a backup of llvm-project-16.0.0.src/mybuild/bin/FileCheck before patching:

cp FileCheck FileCheck.bak

Then replace it with our pythonic version, i.e. SanitizerReports/llvm-FileCheck/FileCheck.

And check if it works well:

echo "HelloWorld!" | ./FileCheck

Patch llvm-lit

llvm-project-16.0.0.src/mybuild/bin/llvm-lit is actually a python script, just like our pythonic FileCheck.

You can open and view it as plain text:

cat llvm-lit

Here I get:

#!/usr/bin/python3.8
# -*- coding: utf-8 -*-

import os
import sys

config_map = {}

def map_config(source_dir, site_config):
    global config_map
    source_dir = os.path.realpath(source_dir)
    source_dir = os.path.normcase(source_dir)
    site_config = os.path.normpath(site_config)
    config_map[source_dir] = site_config

# Set up some builtin parameters, so that by default the LLVM test suite
# configuration file knows how to find the object tree.
builtin_parameters = { 'build_mode' : '.' }

# Allow generated file to be relocatable.
from pathlib import Path
def path(p):
    if not p: return ''
    return str((Path(__file__).parent / p).resolve())


map_config(path(r'../../compiler-rt/test/tsan/lit.cfg.py'), path(r'../projects/compiler-rt/test/tsan/X86_64Config/lit.site.cfg.py'))
map_config(path(r'../../compiler-rt/test/tsan/lit.cfg.py'), path(r'../projects/compiler-rt/test/tsan/X86_64LinuxDynamicConfig/lit.site.cfg.py'))
map_config(path(r'../../clang/test/lit.cfg.py'), path(r'../tools/clang/test/lit.site.cfg.py'))
map_config(path(r'../../clang/test/Unit/lit.cfg.py'), path(r'../tools/clang/test/Unit/lit.site.cfg.py'))
map_config(path(r'../../llvm/utils/lit/tests/lit.cfg'), path(r'../utils/lit/lit.site.cfg'))
map_config(path(r'../../llvm/test/lit.cfg.py'), path(r'../test/lit.site.cfg.py'))
map_config(path(r'../../llvm/test/Unit/lit.cfg.py'), path(r'../test/Unit/lit.site.cfg.py'))

builtin_parameters['config_map'] = config_map

# Make sure we can find the lit package.
llvm_source_root = path(r'../../llvm')
sys.path.insert(0, os.path.join(llvm_source_root, 'utils', 'lit'))

if __name__=='__main__':
    from lit.main import main
    main(builtin_parameters)

It indicates that the true body of lit is located at llvm-project-16.0.0.src/llvm/utils/lit/lit.

Go there and locate llvm-project-16.0.0.src/llvm/utils/lit/lit/cl_arguments.py. Now open and edit this file, modified it like what

SanitizerReports/llvm-lit/lit/cl_arguments.py

Lines 239 to 249 in 6156709

	""" SanitizerReports NOTE
	Read env vars as an alternative to --output
	Will append ".pid" in file path in case used
	simultaneously.
	"""
	#opts.reports = filter(None, [opts.output, opts.xunit_xml_output, opts.resultdb_output, opts.time_trace_output])
	report_lst = [opts.output, opts.xunit_xml_output, opts.resultdb_output, opts.time_trace_output]
	tricky_path = os.getenv("LIT_JSON_OUTPUT")
	if (tricky_path is not None):
	report_lst.append(lit.reports.JsonReport("%s.%d"%(tricky_path, os.getpid())))
	opts.reports = filter(None, report_lst)

does, i.e., let llvm-lit receive file path from environment variable LIT_JSON_OUTPUT as an alternative to --output option. Because when running the testcases it's diffcult to change the behavior of llvm-lit by passing command line options.

Run Testcases

Let's check what testcases is needed by us. Back to llvm-project-16.0.0.src/mybuild and run

make help | grep 'check-' | less

and it will show you all regression tests and unit tests we can run. As for sanitizers, I have screened out the needed ones:

• check-lsan
• check-ubsan
• check-asan-dynamic
• check-asan
• check-dfsan
• check-msan
• check-hwasan-lam
• check-hwasan
• check-tsan-dynamic
• check-tsan
• check-ubsan-minimal
• check-gwp_asan

Now we run check-ubsan-minimal as an example:

LIT_JSON_OUTPUT="`pwd`/ubsan_min.json" make check-ubsan-minimal

Access Report

After the last step done, you will get llvm-project-16.0.0.src/mybuild/ubsan_min.json.pid. It's a JSON file which contains fields like:

{
  "__version__": [
    ...,
    ...,
    ...
  ],
  "elapsed": ...,
  "tests": [
    {
      "code":   "...",
      "elapsed": ...,
      "name":   "...",
      "output": "..."
    },
    {
      ...
    },
    ...
  ]
}

Then you can find content of the report printed by the sanitizer tested before in output key.

Parse Report

Text in output key may contain those stuffs not printed by santizer, error messages, or even multiple reports. So additional text processing is usually necessary to get precise content of the report.

Links

• LLVM Testing Infrastructure Guide
• test-suite Guide
• lit - LLVM Integrated Tester
• FileCheck - Flexible pattern matching file verifier