xref: /openbsd/gnu/llvm/clang/docs/analyzer/user-docs/CrossTranslationUnit.rst (revision 12c85518)

=====================================
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.

.. contents::
   :local:

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 <https://github.com/Ericsson/codechecker>`_ and scan-build-py
(preference for the former).

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 :doc:`compilation database <../../JSONCompilationDatabase>` to
determine the compilation flags used.
The analysis invocation must be provided with the directory which contains the dumps and the mapping files.


Manual CTU Analysis
###################
Let's consider these source files in our minimal example:

.. code-block:: cpp

  // main.cpp
  int foo();

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

.. code-block:: cpp

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

And a compilation database:

.. code-block:: bash

  [
    {
      "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`:

.. code-block:: bash

  $ 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 in format `<USR-Length>:<USR> <File-Path>`:

.. code-block:: bash

  $ clang-extdef-mapping -p . foo.cpp
  9: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:

.. code-block:: bash

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

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

.. code-block:: bash

  $ 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:

.. code-block:: bash

  $ 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`.

Automated CTU Analysis with CodeChecker
#######################################
The `CodeChecker <https://github.com/Ericsson/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:

.. code-block:: bash

  $ 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:

.. code-block:: bash

  $ 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:

.. code-block:: bash

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

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:

.. code-block:: bash

  $ /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
  $

.. _ctu-on-demand:

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
analyzer-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
:doc:`compilation database <../../JSONCompilationDatabase>` 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.


Manual CTU Analysis
###################

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

.. code-block:: cpp

  // main.cpp
  int foo();

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

.. code-block:: cpp

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

The compilation database:

.. code-block:: bash

  [
    {
      "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`:

.. code-block:: bash

  "/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 in format `<USR-Length>:<USR> <File-Path>`:

.. code-block:: bash

  $ clang-extdef-mapping -p . foo.cpp
  9: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:

.. code-block:: bash

  $ 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`.

Automated CTU Analysis with CodeChecker
#######################################
The `CodeChecker <https://github.com/Ericsson/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:

.. code-block:: bash

  $ 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:

.. code-block:: bash

  $ 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:

.. code-block:: bash

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

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`.