2.1. Cross Translation Unit (CTU) Analysis¶

Normally, static analysis works in the boundary of one translation unit (TU). However, with additional steps and configuration we can enable the analysis to inline the definition of a function from another TU.

Overview
PCH-based analysis
On-demand analysis

2.1.1. Overview ¶

CTU analysis can be used in a variety of ways. The importing of external TU definitions can work with pre-dumped PCH files or generating the necessary AST structure on-demand, during the analysis of the main TU. Driving the static analysis can also be implemented in multiple ways. The most direct way is to specify the necessary commandline options of the Clang frontend manually (and generate the prerequisite dependencies of the specific import method by hand). This process can be automated by other tools, like CodeChecker and scan-build-py (preference for the former).

2.1.2. PCH-based analysis ¶

The analysis needs the PCH dumps of all the translations units used in the project. These can be generated by the Clang Frontend itself, and must be arranged in a specific way in the filesystem. The index, which maps symbols’ USR names to PCH dumps containing them must also be generated by the clang-extdef-mapping. Entries in the index must have an .ast suffix if the goal is to use PCH-based analysis, as the lack of that extension signals that the entry is to be used as a source-file, and parsed on-demand. This tool uses a compilation database to determine the compilation flags used. The analysis invocation must be provided with the directory which contains the dumps and the mapping files.

2.1.2.1. Manual CTU Analysis ¶

Let’s consider these source files in our minimal example:

// main.cpp
int foo();

int main() {
  return 3 / foo();
}

// foo.cpp
int foo() {
  return 0;
}

And a compilation database:

[
  {
    "directory": "/path/to/your/project",
    "command": "clang++ -c foo.cpp -o foo.o",
    "file": "foo.cpp"
  },
  {
    "directory": "/path/to/your/project",
    "command": "clang++ -c main.cpp -o main.o",
    "file": "main.cpp"
  }
]

We’d like to analyze main.cpp and discover the division by zero bug. In order to be able to inline the definition of foo from foo.cpp first we have to generate the AST (or PCH) file of foo.cpp:

$ pwd $ /path/to/your/project
$ clang++ -emit-ast -o foo.cpp.ast foo.cpp
$ # Check that the .ast file is generated:
$ ls
compile_commands.json  foo.cpp.ast  foo.cpp  main.cpp
$

The next step is to create a CTU index file which holds the USR name and location of external definitions in the source files:

$ clang-extdef-mapping -p . foo.cpp
c:@F@foo# /path/to/your/project/foo.cpp
$ clang-extdef-mapping -p . foo.cpp > externalDefMap.txt

We have to modify externalDefMap.txt to contain the name of the .ast files instead of the source files:

$ sed -i -e "s/.cpp/.cpp.ast/g" externalDefMap.txt

We still have to further modify the externalDefMap.txt file to contain relative paths:

$ sed -i -e "s|$(pwd)/||g" externalDefMap.txt

Now everything is available for the CTU analysis. We have to feed Clang with CTU specific extra arguments:

$ pwd
/path/to/your/project
$ clang++ --analyze \
    -Xclang -analyzer-config -Xclang experimental-enable-naive-ctu-analysis=true \
    -Xclang -analyzer-config -Xclang ctu-dir=. \
    -Xclang -analyzer-output=plist-multi-file \
    main.cpp
main.cpp:5:12: warning: Division by zero
  return 3 / foo();
         ~~^~~~~~~
1 warning generated.
$ # The plist file with the result is generated.
$ ls -F
compile_commands.json  externalDefMap.txt  foo.ast  foo.cpp  foo.cpp.ast  main.cpp  main.plist
$

This manual procedure is error-prone and not scalable, therefore to analyze real projects it is recommended to use CodeChecker or scan-build-py.

2.1.2.2. Automated CTU Analysis with CodeChecker ¶

The CodeChecker project fully supports automated CTU analysis with Clang. Once we have set up the PATH environment variable and we activated the python venv then it is all it takes:

$ CodeChecker analyze --ctu compile_commands.json -o reports
$ ls -F
compile_commands.json  foo.cpp  foo.cpp.ast  main.cpp  reports/
$ tree reports
reports
├── compile_cmd.json
├── compiler_info.json
├── foo.cpp_53f6fbf7ab7ec9931301524b551959e2.plist
├── main.cpp_23db3d8df52ff0812e6e5a03071c8337.plist
├── metadata.json
└── unique_compile_commands.json

0 directories, 6 files
$

The plist files contain the results of the analysis, which may be viewed with the regular analysis tools. E.g. one may use CodeChecker parse to view the results in command line:

$ CodeChecker parse reports
[HIGH] /home/egbomrt/ctu_mini_raw_project/main.cpp:5:12: Division by zero [core.DivideZero]
  return 3 / foo();
           ^

Found 1 defect(s) in main.cpp


----==== Summary ====----
-----------------------
Filename | Report count
-----------------------
main.cpp |            1
-----------------------
-----------------------
Severity | Report count
-----------------------
HIGH     |            1
-----------------------
----=================----
Total number of reports: 1
----=================----

Or we can use CodeChecker parse -e html to export the results into HTML format:

$ CodeChecker parse -e html -o html_out reports
$ firefox html_out/index.html

2.1.2.3. Automated CTU Analysis with scan-build-py (don’t do it)¶

We actively develop CTU with CodeChecker as the driver for this feature, scan-build-py is not actively developed for CTU. scan-build-py has various errors and issues, expect it to work only with the very basic projects only.

Example usage of scan-build-py:

$ /your/path/to/llvm-project/clang/tools/scan-build-py/bin/analyze-build --ctu
analyze-build: Run 'scan-view /tmp/scan-build-2019-07-17-17-53-33-810365-7fqgWk' to examine bug reports.
$ /your/path/to/llvm-project/clang/tools/scan-view/bin/scan-view /tmp/scan-build-2019-07-17-17-53-33-810365-7fqgWk
Starting scan-view at: http://127.0.0.1:8181
  Use Ctrl-C to exit.
[6336:6431:0717/175357.633914:ERROR:browser_process_sub_thread.cc(209)] Waited 5 ms for network service
Opening in existing browser session.
^C
$

2.1.3. On-demand analysis ¶

The analysis produces the necessary AST structure of external TUs during analysis. This requires the exact compiler invocations for each TU, which can be generated by hand, or by tools driving the analyzer. The compiler invocation is a shell command that could be used to compile the TU-s main source file. The mapping from absolute source file paths of a TU to lists of compilation command segments used to compile said TU are given in YAML format referred to as invocation list, and must be passed as an analyer-config argument. The index, which maps function USR names to source files containing them must also be generated by the clang-extdef-mapping. Entries in the index must not have an .ast suffix if the goal is to use On-demand analysis, as that extension signals that the entry is to be used as an PCH-dump. The mapping of external definitions implicitly uses a compilation database to determine the compilation flags used. The analysis invocation must be provided with the directory which contains the mapping files, and the invocation list which is used to determine compiler flags.

2.1.3.1. Manual CTU Analysis ¶

Let’s consider these source files in our minimal example:

// main.cpp
int foo();

int main() {
  return 3 / foo();
}

// foo.cpp
int foo() {
  return 0;
}

The compilation database:

[
  {
    "directory": "/path/to/your/project",
    "command": "clang++ -c foo.cpp -o foo.o",
    "file": "foo.cpp"
  },
  {
    "directory": "/path/to/your/project",
    "command": "clang++ -c main.cpp -o main.o",
    "file": "main.cpp"
  }
]

The invocation list:

"/path/to/your/project/foo.cpp":
  - "clang++"
  - "-c"
  - "/path/to/your/project/foo.cpp"
  - "-o"
  - "/path/to/your/project/foo.o"

"/path/to/your/project/main.cpp":
  - "clang++"
  - "-c"
  - "/path/to/your/project/main.cpp"
  - "-o"
  - "/path/to/your/project/main.o"

We’d like to analyze main.cpp and discover the division by zero bug. As we are using On-demand mode, we only need to create a CTU index file which holds the USR name and location of external definitions in the source files:

$ clang-extdef-mapping -p . foo.cpp
c:@F@foo# /path/to/your/project/foo.cpp
$ clang-extdef-mapping -p . foo.cpp > externalDefMap.txt

Now everything is available for the CTU analysis. We have to feed Clang with CTU specific extra arguments:

$ pwd
/path/to/your/project
$ clang++ --analyze \
    -Xclang -analyzer-config -Xclang experimental-enable-naive-ctu-analysis=true \
    -Xclang -analyzer-config -Xclang ctu-dir=. \
    -Xclang -analyzer-config -Xclang ctu-invocation-list=invocations.yaml \
    -Xclang -analyzer-output=plist-multi-file \
    main.cpp
main.cpp:5:12: warning: Division by zero
  return 3 / foo();
         ~~^~~~~~~
1 warning generated.
$ # The plist file with the result is generated.
$ ls -F
compile_commands.json  externalDefMap.txt  foo.cpp  main.cpp  main.plist
$

This manual procedure is error-prone and not scalable, therefore to analyze real projects it is recommended to use CodeChecker or scan-build-py.

2.1.3.2. Automated CTU Analysis with CodeChecker ¶

The CodeChecker project fully supports automated CTU analysis with Clang. Once we have set up the PATH environment variable and we activated the python venv then it is all it takes:

$ CodeChecker analyze --ctu --ctu-ast-loading-mode on-demand compile_commands.json -o reports
$ ls -F
compile_commands.json  foo.cpp main.cpp  reports/
$ tree reports
reports
├── compile_cmd.json
├── compiler_info.json
├── foo.cpp_53f6fbf7ab7ec9931301524b551959e2.plist
├── main.cpp_23db3d8df52ff0812e6e5a03071c8337.plist
├── metadata.json
└── unique_compile_commands.json

0 directories, 6 files
$

The plist files contain the results of the analysis, which may be viewed with the regular analysis tools. E.g. one may use CodeChecker parse to view the results in command line:

$ CodeChecker parse reports
[HIGH] /home/egbomrt/ctu_mini_raw_project/main.cpp:5:12: Division by zero [core.DivideZero]
  return 3 / foo();
           ^

Found 1 defect(s) in main.cpp


----==== Summary ====----
-----------------------
Filename | Report count
-----------------------
main.cpp |            1
-----------------------
-----------------------
Severity | Report count
-----------------------
HIGH     |            1
-----------------------
----=================----
Total number of reports: 1
----=================----

Or we can use CodeChecker parse -e html to export the results into HTML format:

$ CodeChecker parse -e html -o html_out reports
$ firefox html_out/index.html

2.1.3.3. Automated CTU Analysis with scan-build-py (don’t do it)¶

We actively develop CTU with CodeChecker as the driver for feature, scan-build-py is not actively developed for CTU. scan-build-py has various errors and issues, expect it to work only with the very basic projects only.

Currently On-demand analysis is not supported with scan-build-py.