13.1. How to write basf2 code (coding conventions)#

The following rules and conventions should ensure a good usability and maintainability of the Belle II software. They are written for C++, but have to be followed for Python code as well, unless other rules for Python are given.

13.1.1. General rules#

Try to keep your code as simple as possible:

  1. The interfaces should be easy to use and consistent throughout the entire software. Use inheritance and templates with care: try aggregation instead of inheritance, avoid multiple inheritance or friends, don’t use private inheritance.

  2. Aim for high coherence (only things belonging to a class should be there, and the class should be complete) and low coupling.

  3. The code should compile without warning messages and must be portable (e.g. it should run on x86_64 64 bit machines, we do not yet support ARM processors).

Please, please avoid:

  1. Hacks and fancy features. If you really know what you are doing, and you really, really want to implement a hack like the infamous fast inverse square root hack please provide detailed documentation on what you do and why.

  2. Interfaces to interfaces to interfaces.

  3. Too much dependence on external software; check whether a plug-in mechanism can be used if external software is required.

  4. Flexibility if it significantly increases the possibility of making mistakes: the system should behave correctly by default; a user should not be required to change any parameters from their default value to get a correct result.

  5. Side effects: if code has side effects this must be clearly documented.

  6. Preprocessor definitions if possible; in particular don’t use them for constants.

13.1.2. Directories structure#

By default, one shared library is created per basf2 package and is installed in a top-level lib directory that is included in the user’s library path. The build system treats the package’s contents in pre-defined subdirectories as follows:

  • include and src must always include the header and source files, respectively, of the code written in C++.

  • modules: the code is compiled into a shared library and installed in a top-level module directory, so that it can be dynamically loaded by basf2.

  • tools: C++ code is compiled into an executable and installed in a top-level bin directory that is included in the user’s path; executable scripts, usually written in Python, are symlinked to this directory.

  • dataobjects: these classes define the organization of the data that can be stored in output files; the code is linked in a shared library with the _dataobjects suffix.

  • scripts: Python scripts are installed in a directory that is included in the Python path; this folder should then only contain Python modules.

  • data: all files are symlinked to a top-level data folder; this folder should contain XML files for defining the detector geometry, generators configuration files, etc., but it must not contain files containing calibration constants, particle identification likelihoods or other contents that can be stored in our conditions database.

  • tests: unit and script tests used by our CI/CD pipelines.

  • validation: scripts and reference histograms for validation plots.

  • examples: example scripts that illustrate features of the package.

13.1.3. Naming conventions#

Names should be precise, colorful and in self-explanatory English:

  1. Avoid abbreviations where possible, e.g. use TrackFitter instead of TrFt.

  2. Use plurals for containers (lists, vectors, etc.), e.g. particles, tracks.

  3. When combining several words, use the order corresponding to natural English language, e.g. CDCHit instead of HitCDC.

  4. Use first-letter capitalization instead of underscore for constructing names, e.g. getTotalEnergy() instead of get_total_energy() in C++.

  5. Use the get prefix for methods that return the value of the data member (getters). Use the set prefix for methods that set the value of a data member by parsing an argument (setters). In addition, usage of is instead of get as prefix for methods that return a boolean is recommended, e.g. isPhoton(const auto& someParticle) instead of getPhotonStatus(const auto& someParticle) or isFittedTrack(const auto& track) instead of getTrackFitStatus(const auto& track).

  6. Python: except for class names, use lower case words with underscores, e.g. find_particle().

  7. Specifically:

    • class and type names start with a capital letter (CapWords, same for Python)

    • local variable and method names start with a lower case letter

    • enum types start with E followed by a capital letter

    • template arguments begin with A followed by a capital letter

    • use the prefix m_ for member variables, s_ for static member variables, h_ or m_h for histogram member variables, and c_ for constants

    • Python: use all upper case with underscores for constants; always use self/cls as first argument of instance/class methods

    • module parameter names are equal to the corresponding variable name, but without m_ prefix

  8. If values are not given in standard units the used units have to be stated explicitly in the variable name. Standard units are cm, nanosecond, radian, GeV, Kelvin, Tesla, elementary charge as defined in Unit.h.

  9. Executable that are part of the official interface on which users and other tools can rely on must have the prefix b2 and must be documented in Sphinx. Other executables must have a prefix equal to the name of the package to which they belong.

  10. Avoid the usage of “slave/master”, “blacklist/whitelist”. Prefer instead: “controller/agent”, “primary/secondary”, “denylist/allowlist”.

  11. In general, avoid anything that might cause distress or feelings of exclusion to other collaborators.

In case of doubts, don’t hesitate to contact the coordinators.

In the following table Xyz stands for an arbitrary name and DET for a detector name, like PXD or ARICH.

Name

Description

DET SimHit

MC truth information about an energy deposition

DET Digit

simulated or real detector signal, the raw data information

DET Hit

reconstructed hit, after clustering

DET Digitizer

class (usually a module, see below) that creates digits from simulated hits

DET Clusterizer

class (usually a module, see below) that creates hits from digits

DET Trigger Xyz

classes for detector related triggers

Xyz Collection

a class for an array of objects of class Xyz, if not a StoreArray

Xyz Module

a basf2 module class

Xyz Base

a class whose main purpose is to serve as a base class for derived classes

Abc Xyz

Classes derived from Xyz Base

Xyz Error

an exception class

../../../_images/object_naming_convention.svg

Fig. 13.1 Object naming conventions in basf2.#

Warning

Some subdetector packages may use slightly different naming conventions. In case you need further clarifications, please check the sphinx documentation of the subdetector package (when available) or contact the subdetector librarians.

Tip

Please check the AWESOME subpackage for a full, working, example on how to start writing subdetector-specific code.

13.1.4. Namespaces#

  1. All the definitions should be done in namespace Belle2.

  2. Use sub-namespaces with names equal to the package names for everything (e.g.: namespace Belle2::Something), except for datastore objects and anything in a dataobjects, dbobjects or modules subdirectory.

  3. using namespace is forbidden in header files; include statements must be outside namespaces.

  4. enum class should be used to implement enumerations to contain the constant within the enum’s scope

13.1.5. Classes#

Each class implementation should go into a single source code file.

The ordering in the class definition is public, protected, private, and in each such section elements shall be ordered like:

  • typedefs and enums,

  • constructors and destructor,

  • operators,

  • other methods,

  • data members, while in case of modules the steerable data members before non-steerable ones,

with static methods and members after non-static ones. Public data members are forbidden.

The only exception to the rule above are simple structs with no methods except for a default constructor. In this case public data members (without m_ prefix) are allowed. Use the struct instead of the class keyword for such structures.

Python: follow basically the same rules, but put data members first.

Only very short functions like simple setters and getters should be in the declaration. Implementation of non-template methods should go to the source file.

Methods should be declared const if possible. When overriding a virtual method of a base class, the override keyword should be used to mark this intention. The final keyword may be used for methods that should not be overridden or classes from which no other classes should be derived.

Try to pass function parameters by value, const reference, or const pointer, especially if they are not meant to be changed in the function.

13.1.6. Order of class members#

It is advisable to order the data members in the following order:

  1. std::string (has size of 32 bytes on 64 bit PCs as we usually use)

  2. double and long / unsigned long and pointers (have size of 8 bytes)

  3. float and int / unsigned int (have size of 4 bytes)

  4. short / unsigned short (have size of 2 bytes)

  5. bool (has size of 1 byte)

The reason for this order is how processors and memory work. Modern PCs usually have a so-called cache that holds data (data cache) and instructions (instruction cache) that either are used often, or expected to be used in the next O(100) clock cycles. The content of the cache is retrieved from memory in cache lines of up to 64 bytes. Thus, the data should be ordered such that the bytes in the cache lines are filled. Although a bool technically can be represented by a single bit, it occupies a byte in memory as memory addresses are based on units of 1 byte, thus the smallest addressable chunk of memory is a byte but not a bit.

In a simple example of a cache line of 16 bytes these should either be filled by two double or four float or int. Having a double followed by a bool followed by another double would create 7 empty bytes that cannot be filled properly, and the second double would end up in the next cache line, potentially causing a cache miss. Cache misses can significantly slow down a program, as retrieving data from cache only costs a few CPU clock cycles, while retrieving data from RAM usually costs O(100) CPU clock cycles.

Additionally, ordering the data types as described above will lead to a smaller memory footprint of the class, which is critical for basf2 on the HLT.

Because usually classes contain several data members, there is no clear rule where to add classes as data members in other classes, usually it would be at position 1, 2, or 3.

13.1.7. Initialisation#

Always initialize variables immediately in the same statement where you create them (see explanation).

All data members of a class must be initialized at their declaration. Specific constructors should use the initialization list (after the colon of the constructor declaration) instead of an initialization in the body of the constructor if possible.

13.1.8. Data objects#

  1. Place all classes that should be used in the data store and written to output files into the dataobjects folder.

  2. These classes must not depend on any other basf2 code. The only exception are other data objects classes and the core framework.

  3. Classes that will be stored as single objects (not in TClonesArrays) in the data store must implement the Clear method.

  4. Use normal C++ types for basic types (int, double, etc) and ROOT classes for vectors and matrices (ROOT::Math::XYZVector, TMatrixD, etc.).

  5. Classes that are used for temporary or intermediate objects in the data store and are not written to output files can be put into any folder. They must always be registered as transient in the data store.

13.1.9. Random numbers#

Always use gRandom for generating random numbers.

13.1.10. Error handling#

In case of serious errors an exception should be thrown, otherwise return codes should be used to indicate whether an operation was successful or not.

When an exception is thrown a B2ERROR message should be issued, too.

Use BELLE2_DEFINE_EXCEPTION(name, description) to define exceptions. Define them inside the classes to which they belong.

Use the logging system:

  • B2FATAL(message) for situations were the program execution cannot be continued. (An advantage of this type of message is that it cannot be ignored.)

  • B2ERROR(message) for things that went wrong and must be fixed.

  • B2WARNING(message) for potential problems that should not be ignored and only accepted if understood.

  • B2INFO(message) for informational messages that are relevant to the user.

  • B2DEBUG(level, message) everything else, in particular debug information that is useful for developers. The following debug levels should be used

    • 0-9 for user code. The use case is that a user wants to debug their analysis code. Debug levels below 10 must not be used is basf2 code.

    • 10-19 for analysis package code. The use case is that a user wants to debug problems in analysis jobs with the help of experts.

      • For even more output control: 10 is enabled when running debug with no arguments, so messages 0-10 will be shown when running –debug <no args>

    • 20-29 for simulation/reconstruction code. The use case is that a developer wants to debug their code or the debugging of production jobs by experts.

    • 30-39 for framework code. The use case is that an expert wants to debug problems in the core software.

13.1.11. Files#

For C++, header files have the extension .h, sources .cc with the base file name corresponding to the class name defined / implemented in the file. Each header file should have a multiple inclusion protection using:

#pragma once

close to the beginning of the file. In older files you can still find protections using ifndef:

#ifndef FILENAME_H
#define FILENAME_H
 ...
#endif

which is perfectly valid and doesn’t need to be changed, but for new files #pragma once is preferred.

All include statements must be at the top of the beginning of the file and should use the <> - notation for global available includes of the form:

#include <package/dir/otherclass.h>

Includes should be kept to a minimum: all necessary symbols in the header should be forward declared if possible (to reduce depedencies between the header files), or, if that is not possible, the appropriate header needs to be included.

You must add the following header to all C++ files:

/**************************************************************************
* basf2 (Belle II Analysis Software Framework)                           *
* Author: The Belle II Collaboration                                     *
*                                                                        *
* See git log for contributors and copyright holders.                    *
* This file is licensed under LGPL-3.0, see LICENSE.md.                  *
**************************************************************************/

and the following to all python files:

##########################################################################
# basf2 (Belle II Analysis Software Framework)                           #
# Author: The Belle II Collaboration                                     #
#                                                                        #
# See git log for contributors and copyright holders.                    #
# This file is licensed under LGPL-3.0, see LICENSE.md.                  #
##########################################################################

You do not need to add your name (your work is attributed via the git log) to the file unless as a contact person for a validation script or another operational reason.

Python: All files have the extension .py.

13.1.12. Code Formatting Style#

In general use

  • 2 spaces for indentation,

  • Stroustrup style for brackets, i.e. brackets are attached except for functions,

  • spaces around keywords, operands and parentheses.

Example:

if (x == y) {
  ...
} else if (x > y) {
  ...
} else {
  ....
}

Python: Follow the Style Guide for Python Code.

Tip

You can use the b2code-style-fix tool to format your code according to the style rules. Only code that follows the required style can be committed to the git repository! The b2code-style-check tool can be used to print the changes that the b2code-style-fix tool would apply.

In the definition of pointers and references the * and & symbols have to be attached to the variable type, not the variable name, e.g.:

int* pIndex;
int& index = *pIndex;

Be aware of the (wrong) C++ convention for definitions of multiple variables in one statement. E.g. int* a, b; will define a pointer a and an int variable (not a pointer) b. It is recommended to split multiple pointer definitions to separate statements.

13.1.13. Documentation#

All code has to be documented using Software.DoxyGen style comments – you can find examples in the DoxyGen article. Rules on what should be documented, and how:

  • Each class needs a detailed comment describing the purpose of the class.

  • Each method needs a comment describing at least the input and output quantities.

  • Comments on data members are encouraged, unless their meaning is obvious.

  • Comments on design decisions and implementation details may be added if appropriate. External documentation may be more useful to describe the overall design.

  • Python: Use docstrings instead of comments for the documentation of classes and methods/functions.

  • If a file contains functions that are not methods of a class, the documentation generation has to be enabled by documenting the file. This is done by adding /** @file filename */ at the beginning of the file, where filename should be replaced by the actual file name. Further documentation describing the file can be added in the comment. Python : The corresponding statement for python files is ##@package followed by a package name. This has to be added before the function definitions, e.g. directly after the utf-coding definition. Further documentation describing the package can be added after this line as comments.

  • Documentation about the input and output of modules is automatically generated if both are registered in the initialize method.

13.1.14. Further Suggestions#

  • Use iXxx to indicate the identifying integer for an object, e.g. a variable for asking for a specific layer is iLayer.

  • Use nXxx to indicate the total number of something, e.g. the number of layers, that a subdetector has is nLayer.

  • If you get a warning about too large or unbound stack usage, use std::vector instead of a one-dimensional C array or boost::multi_array instead of multi-dimensional C arrays. See these slides or mail for more information.