PETSc Style and Usage Guide#

The PETSc team uses certain conventions to make the source code consistent and easier to maintain. We will interchangeably use the terminology subclass, implementation, or type [1] to refer to a concrete realization of an abstract base class. For example, KSPGMRES is a type for the base class KSP.

Names#

Consistency of names for variables, functions, and so on is extremely important. We use several conventions

All function names and enum types consist of acronyms or words, each of which is capitalized, for example, KSPSolve() and MatGetOrdering().
All enum elements and macro variables are named with all capital letters. When they consist of several complete words, there is an underscore between each word. For example, MAT_FINAL_ASSEMBLY.
Functions that are private to PETSc (not callable by the application code) either
- have an appended _Private (for example, StashValues_Private) or
- have an appended _Subtype (for example, MatMultSeq_AIJ).
In addition, functions that are not intended for use outside of a particular file are declared static. Also, see the item on symbol visibility in Usage of PETSc Functions and Macros.
Function names in structures (for example, _matops) are the same as the base application function name without the object prefix and in lowercase. For example, MatMultTranspose() has a structure name of multtranspose.
Names of implementations of class functions should begin with the function name, an underscore, and the name of the implementation, for example, KSPSolve_GMRES().
Each application-usable function begins with the name of the class object, followed by any subclass name, for example, ISInvertPermutation(), MatMult(), or KSPGMRESSetRestart().
Functions that PETSc provides as defaults for user-providable functions end with Default (for example, PetscSignalHandlerDefault()).
Options database keys are lower case, have an underscore between words, and match the function name associated with the option without the word “set” or “get”, for example, -ksp_gmres_restart.
Specific XXXType values (for example, MATSEQAIJ) do not have an underscore in them unless they refer to another package that uses an underscore, for example, MATSOLVERSUPERLU_DIST.
Typedefs for functions should end in Fn as in, for example, SNESFunctionFn.

PETSc and standard datatypes#

PetscInt is generally used for array indices, and array lengths. It is a signed 32-bit or 64-bit int depending on the ./configure option --with-64-bit-indices. There is the possibility of integer overflow with the 32-bit version.
PetscCount is ptrdiff_t, (on 64-bit systems it is 64-bits) and should be used for array sizes (number of entries) and indices that may become large. PetscIntCast() should always be used when converting to PetscInt from PetscCount. Since, in most configurations an array of PetscCount requires twice the memory of an array of PetscInt most index arrays (such as ISGetIndices() use PetscInt, when these arrays get too large then --with-64-bit-indices must be used to ./configure PETSc. In most cases it is appropriate to use PetscCount in lieu of PetscInt64.
size_t is used for variables that contain the amount of memory, generally in bytes. It should not be used for the number of entries in an array, or to index into an array, that should be PetscCount, or PetscInt. Though size_t is unsigned and hence can have values larger then those that can be stored in a PetscCount those sizes will never be reached in practice so it is ok to cast with (PetscCount) from a size_t variable to a PetscCount variable, but not a PetscInt. One should not blindly cast from a PetscCount or a PetscInt to size_t since, when the value is negative, it will produce garbage.
Never blindly put in a cast from a potentially longer (in number of bits) to a shorter integer such as
```
PetscInt64 a
PetscInt b
b = (PetscInt)a
```
simply to prevent a compiler warning. Use the appropriate PETSc cast function unless you absolutely know the value will fit in the lesser number of bits.
MPI 4.0 supports the use of MPI_Count (large count) for many MPI functions that previously used int in a new API where the MPI function names end in _c. Since not all installed MPI implementations have such support, use MPIU_XXX() routines that use PetscCount for count arguments and use the large count MPI versions when possible. When not possible they first check the size of the input count arguments and error if they will not fit in the MPI required int, if they fit then the standard MPI functions are automatically called.
```
sizeof(MPI_Count) >= sizeof(PetscCount) >= sizeof(PetscInt)
sizeof(PetscInt64) >= sizeof(PetscCount) >= sizeof(PetscInt)
sizeof(MPI_Count) may be strictly greater than sizeof(PetscInt64)
```

Coding Conventions and Style#

Within the PETSc source code, we adhere to the following guidelines so that the code is uniform and easily maintained.

C Formatting#

The .clang-format file in the PETSc root directory controls the white space and basic layout. You can run the formatter in the entire repository with make clangformat. All merge requests must be properly formatted; this is automatically checked for merge requests with make checkclangformat.

Even with the use of clang-format there are still many decisions about code formatting that must be constantly made. A subset of these is automatically checked for merge requests with make checkbadSource.

The prototypes for functions should not include the names of the variables

PETSC_EXTERN PetscErrorCode MyFunction(PetscInt); // Correct
PETSC_EXTERN PetscErrorCode MyFunction(PetscInt myvalue); // Incorrect

All local variables of a particular type (for example, PetscInt) should be listed on the same line if possible; otherwise, they should be listed on adjacent lines. Note that pointers of different arity (levels of indirection) are considered to be different types. clang-format automatically handles the indenting shown below.
```
// Correct
PetscInt   a, b, c;
PetscInt  *d, *e;
PetscInt **f;

// Incorrect
PetscInt a, b, c, *d, *e, **f;
```

Local variables should be initialized in their declaration when possible

// Correct
PetscInt a = 11;

PetscFunctionBegin;
// use a

// Incorrect
PetscInt a;

PetscFunctionBegin;
a = 11;
// use a

All PETSc subroutine code blocks must start with a single blank line between the local variable declarations followed by PetscFunctionBegin.

// Correct
PetscInt x;

PetscFunctionBegin;

// Incorrect
PetscInt x;
PetscFunctionBegin;

// Incorrect
PetscInt x;
y = 11;

Functions in PETSc examples, including main() should have PetscFunctionBeginUser as the first line after the local variable declarations.
PETSc functions that begin PetscFunctionBegin must always return via PetscFunctionReturn(), or PetscFunctionReturnVoid(), not return. If the function returns a PetscErrorCode, then it must always return with PetscFunctionReturn(PETSC_SUCCESS).
Functions that do use return should use return xx; rather than return(xx);
All PETSc function calls must have their return value checked for errors using the PetscCall() macro. This should be wrapped around the function in question.
```
PetscCall(MyFunction(...)); // Correct
PetscErrorCode ierr = MyFunction(...);PetscCall(ierr); // Incorrect
```
The only exceptions to this rule are begin-end style macros which embed local variables or loops as part of their expansion (e.g. PetscOptionsBegin()/PetscOptionsEnd()). These handle errors internally and do not need error checking.
```
// Correct
PetscOptionsBegin(...);
PetscOptionsEnd();
```
As a rule, always try to wrap the function first; if this fails to compile, you do not need to add the error checking.

Calls to external package functions are generally made with PetscCallExternal() or its variants that are specialized for particular packages, for example PetscCallBLAS()
Single operation if and else commands should not be wrapped in braces. They should be done as follows,
```
if ( ) XXXX;
else YYY;
```
not
```
if ( ) {XXXX;}
else {YYY;}
```
Do not leave sections of commented-out code or dead source code protected with ifdef foo in the source files.
Use classic block comments (/* There must be a space before the first word in the comment and a space at the end */, (/*Do not do this*/) for multi-line comments, and // Comment for single-line comments in source files.
Do not put a * at the beginning or end of each line of a multi-line comment.
Do not use /* ---- ... ----- */ or similar constructs to separate parts of source code files.
Use appropriate grammar and spelling in the comments.

All variables must be declared at the beginning of the code block (C89 style), never mixed in with code. However, when variables are only used in a limited scope, it is encouraged to declare them in that scope. For example:

if (cond) {
  PetscScalar *tmp;

  PetscCall(PetscMalloc1(10, &tmp));
  // use tmp
  PetscCall(PetscFree(tmp));
}

The only exception to this is variables used exclusively within a for loop, which must be declared inside the loop initializer:

// Correct
for (PetscInt i = 0; i < n; ++i) {
  // loop body
}

// Correct, variable used outside of loop
PetscInt i;

for (i = 0; i < n; ++i) {
  // loop body
}
j = i;

// Incorrect
PetscInt i;
...
for (i = 0; i < n; ++i) {
  // loop body
}

Developers can use // to split very long lines when it improves code readability. For example

f[j][i].omega = xdot[j][i].omega + uxx + uyy //
              + (vxp * (u - x[j][i - 1].omega) + vxm * (x[j][i + 1].omega - u)) * hy //
              + (vyp * (u - x[j - 1][i].omega) + vym * (x[j + 1][i].omega - u)) * hx //
              - .5 * grashof * (x[j][i + 1].temp - x[j][i - 1].temp) * hy;

The use of // clang-format off is allowed in the source code but should only be used when necessary. It should not be used when trailing // to split lines works.
```
// clang-format off
f ...
// clang-format on
```
size and rank should be used exclusively for the results of MPI_Comm_size() and MPI_Comm_rank() and other variable names for these values should be avoided unless necessary.

C Usage#

Do not use language features that are not in the intersection of C99, C++11, and MSVC v1900+ (Visual Studio 2015). Examples of such banned features include variable-length arrays. Note that variable-length arrays (including VLA-pointers) are not supported in C++ and were made optional in C11. You may use designated initializers via the PetscDesignatedInitializer() macro.
Array and pointer arguments where the array values are not changed should be labeled as const arguments.
Scalar values passed to functions should never be labeled as const.
Subroutines that would normally have a void ** argument to return a pointer to some data should be prototyped as void *. This prevents the caller from having to put a (void **) cast in each function call. See, for example, DMDAVecGetArray().
Do not use the register directive.
Use if (v == NULL) or if (flg == PETSC_TRUE), instead of using if (!v) or if (flg) or if (!flg).
Avoid #ifdef or #ifndef when possible. Rather, use #if defined or #if !defined. Better, use PetscDefined() (see below). The only exception to this rule is for header guards, where the #ifndef form is preferred (see below).

Header guard macros should be done using #pragma once. This must be the very first non-comment line of the file. There must be no leading or trailing empty (non-comment) lines in the header. For example, do

/*
  It's OK to have

  comments
*/
// before the guard
#pragma once

// OK, other headers included after the guard
#include <petscdm.h>
#include <petscdevice.h>

// OK, other preprocessor symbols defined after the guard
#define FOO_BAR_BAZ

// OK, regular symbols defined after the guard
typedef struct _p_PetscFoo *PetscFoo;
...

Do not do

// ERROR, empty lines at the beginning of the header



// ERROR, included other headers before the guard
#include <petscdm.h>
#include <petscdevice.h>

// ERROR, defined other preprocessor symbols before the guard
#define FOO_BAR_BAZ

// ERROR, defined regular symbols before the guard
typedef struct _p_PetscFoo *PetscFoo;

#pragma once

Never use system random number generators such as rand() in PETSc code or examples because these can produce different results on different systems, thus making portability testing difficult. Instead, use PetscRandom which produces the same results regardless of the system used.
Variadic macros may be used in PETSc, but must work with MSVC v1900+ (Visual Studio 2015). Most compilers have conforming implementations of the C99/C++11 rules for __VA_ARGS__, but MSVC’s implementation is not conforming and may need workarounds. See PetscDefined() for an example of how to work around MSVC’s limitations to write a macro that is usable in both.

Usage of PETSc Functions and Macros#

Lengthy conditional preprocessor blocks should mark any #else or #endif directives with a comment containing (or explaining) either the boolean condition or the macro’s name if the first directive tests whether one is defined. One should be able to read any part of the macroblock and find or deduce the initial #if. That is:
```
#if defined(MY_MACRO)
// many lines of code
#else // MY_MACRO (use name of macro)
// many more lines of code
#endif // MY_MACRO

#if MY_MACRO > 10
// code
#else // MY_MACRO < 10
// more code
#endif // MY_MACRO > 10
```
Public PETSc include files, petsc*.h, should not reference private PETSc petsc/private/*impl.h include files.
Public and private PETSc include files cannot reference include files located in the PETSc source tree.

All public functions must sanity-check their arguments using the appropriate PetscValidXXX() macros. These must appear between PetscFunctionBegin and PetscFunctionReturn() For example

PetscErrorCode PetscPublicFunction(Vec v, PetscScalar *array, PetscInt collectiveInt)
{
  PetscFunctionBegin;
  PetscValidHeaderSpecific(v, VEC_CLASSID, 1);
  PetscAssertPointer(array, 2);
  PetscValidLogicalCollectiveInt(v, collectiveInt, 3);
  ...
  PetscFunctionReturn(PETSC_SUCCESS);
}

See include/petsc/private/petscimpl.h and search for “PetscValid” to see all available checker macros.

When possible, use PetscDefined() instead of preprocessor conditionals. For example, use:
```
if (PetscDefined(USE_DEBUG)) { ... }
```
instead of:
```
#if defined(PETSC_USE_DEBUG)
  ...
#endif
```
The former usage allows syntax and type-checking in all configurations of PETSc, whereas the latter needs to be compiled with and without debugging to confirm that it compiles.

Never put a function call in a return statement; do not write

PetscFunctionReturn( somefunction(...) ); /* Incorrect */

Do not put a blank line immediately after PetscFunctionBegin; or a blank line immediately before PetscFunctionReturn(PETSC_SUCCESS);.
Do not include assert.h in PETSc source code. Do not use assert(), it doesn’t play well in the parallel MPI world. You may use PetscAssert() where appropriate. See PetscCall() documentation for guidance of when to use PetscCheck() vs. PetscAssert().
Make error messages short but informative. The user should be able to reasonably diagnose the greater problem from your error message.
Except in code that may be called before PETSc is fully initialized, always use PetscMallocN() (for example, PetscMalloc1()), PetscCallocN(), PetscNew(), and PetscFree(), not malloc() and free().
MPI routines and macros that are not part of the 2.1 standard should not be used in PETSc without appropriate configure checks and #if PetscDefined() checks. Code should also be provided that works if the MPI feature is not available; for example,
```
#if PetscDefined(HAVE_MPI_REDUCE_LOCAL)
  PetscCallMPI(MPI_Reduce_local(inbuf, inoutbuf, count, MPIU_INT, MPI_SUM));
#else
  PetscCallMPI(MPI_Reduce(inbuf, inoutbuf, count, MPIU_INT, MPI_SUM, 0, PETSC_COMM_SELF);
#endif
```
Do not introduce PETSc routines that provide essentially the same functionality as an available MPI routine. For example, do not write a routine PetscGlobalSum() that takes a scalar value and performs an MPI_Allreduce() on it. Instead, use the MPI routine MPI_Allreduce() directly in the code.
Never use a local variable counter such as PetscInt flops = 0; to accumulate flops and then call PetscLogFlops(); always just call PetscLogFlops() directly when needed.
Library symbols meant to be directly usable by the user should be declared PETSC_EXTERN in their respective public header file. Symbols intended for internal use should instead be declared PETSC_INTERN. Note that doing so is unnecessary in the case of symbols local to a single translation unit; these should be declared static. PETSc can be configured to build a separate shared library for each top-level class (Mat, Vec, KSP, and so on), and that plugin implementations of these classes can be included as separate shared libraries; thus, otherwise private symbols may need to be marked PETSC_SINGLE_LIBRARY_INTERN. For example
- MatStashCreate_Private() is marked PETSC_INTERN as it is used across compilation units, but only within the Mat package;
- all functions, such as KSPCreate(), included in the public headers (include/petsc*.h) should be marked PETSC_EXTERN;
- VecLoad_Default() is marked PETSC_SINGLE_LIBRARY_INTERN as it may be used across library boundaries, but is not intended to be visible to users;
Before removing or renaming an API function, type, or enumerator, PETSC_DEPRECATED_XXX() should be used in the relevant header file to indicate the new usage and the PETSc version number where the deprecation will first appear. The old function or type, with the deprecation warning, should remain for at least one major release. We do not remove support for the deprecated functionality unless there is a specific reason to remove it; it is not removed simply because it has been deprecated for “a long time.”

The function or type’s manual page should be updated (see Manual Page Format). For example,
```
typedef NewType OldType PETSC_DEPRECATED_TYPEDEF("Use NewType (since version 3.9)");

PETSC_DEPRECATED_FUNCTION("Use NewFunction() (since version 3.9)") PetscErrorCode OldFunction();

#define OLD_ENUMERATOR_DEPRECATED  OLD_ENUMERATOR PETSC_DEPRECATED_ENUM("Use NEW_ENUMERATOR (since version 3.9)")
typedef enum {
  OLD_ENUMERATOR_DEPRECATED = 3,
  NEW_ENUMERATOR = 3
} MyEnum;
```
Note that after compiler preprocessing, the enum above would be transformed into something like
```
typedef enum {
  OLD_ENUMERATOR __attribute__((deprecated)) = 3,
  NEW_ENUMERATOR = 3
} MyEnum;
```
Before removing or renaming an options database key, PetscOptionsDeprecated() should be used for at least one major release. We do not remove support for the deprecated functionality unless there is a specific reason to remove it; it is not removed simply because it has been deprecated for “a long time.”
The format strings in PETSc ASCII output routines, such as PetscPrintf(), take a %" PetscInt_FMT " for all PETSc variables of type PetscInt, not a %d.
All arguments of type PetscReal to PETSc ASCII output routines, such as PetscPrintf, must be cast to double, for example,
```
PetscPrintf(PETSC_COMM_WORLD, "Norm %g\n", (double)norm);
```

Formatted Comments#

PETSc uses formatted comments and the Sowing packages [Gro95b] [Gro95a] to generate documentation (manual pages) and the Fortran interfaces. Documentation for Sowing and the formatting may be found at http://wgropp.cs.illinois.edu/projects/software/sowing/; in particular, see the documentation for doctext. Currently, doctext produces Markdown files ending in .md, which Sphinx later processes.

/*@

a formatted comment of a function that will be used for documentation and a Fortran interface.
/*@C

a formatted comment of a function that will be used only for documentation, not to generate a Fortran interface. Certain constructs and usages do not yet support automatically generating a Fortran interface. In general, such labeled C functions should have a custom Fortran interface provided.
/*E

a formatted comment of an enum used for documentation only. Note that each of these needs to be listed in lib/petsc/conf/bfort-petsc.txt as a native and defined in the corresponding include/petsc/finclude/petscxxx.h Fortran include file and the values set as parameters in the file src/SECTION/f90-mod/petscSUBSECTION.h, for example, src/vec/f90-mod/petscis.h.
/*S

a formatted comment for a data type such as KSP. Each of these needs to be listed in lib/petsc/conf/bfort-petsc.txt as a nativeptr.
/*J

a formatted comment for a string type such as KSPType.
/*MC

a formatted comment of a CPP macro or enum value for documentation.

The Fortran interface files supplied manually by the developer go into the two directories ftn-custom and f90-custom, while those generated by Sowing go into ftn-auto directories in the $PETSC_ARCH/src`` directory tree.

Each include file that contains formatted comments needs to have a line of the form

/* SUBMANSEC = submansec (for example Sys) */

preceded by and followed by a blank line. For source code, this information is found in the makefile in that source code’s directory in the format

MANSEC   = DM
SUBMANSEC= DMPlex

Manual Page Format#

Each function, typedef, class, macro, enum, and so on in the public API should include the following data, correctly formatted (see codes section) to generate complete manual pages and (possibly) Fortran interfaces with Sowing. All entries below should be separated by blank lines. Except where noted, add a newline after the section headings.

The item’s name, followed by a dash and brief (one-sentence) description.
If documenting a function implemented with a preprocessor macro (e.g., PetscOptionsBegin()), an explicit Synopsis: section noting the required header and the function signature.
If documenting a function, a description of the function’s “collectivity”.
- Not Collective if the function need not be called on multiple (or possibly all) MPI processes
- Collective if the function is a collective operation.
- Logically Collective; yyy must contain common value] if the function is collective but does not require any actual synchronization (e.g., setting class parameters uniformly). Any argument yyy, which must have the same value on all ranks of the MPI communicator should be noted here.
If the function is not supported in Fortran, then after the collective information, on the same line, one should provide ; No Fortran support.
If documenting a function with input parameters, a list of input parameter descriptions in an Input Parameter(s): section.
If documenting a function with output parameters, a list of output parameter descriptions in an Output Parameter(s): section.
If any input or output parameters are function pointers, they should be documented in the style
```
Calling sequence of `func()`:
. arg - the integer argument description
```
If documenting a function that interacts with the options database, a list of options database keys in an Options Database Key(s): section.
Level: (no newline) followed by beginner, intermediate, advanced, developer, or deprecated. This should be listed before the various Note(s): sections.
(Optional) a Note(s): section containing in-depth discussion, technical caveats, special cases, and so on. If it is ambiguous whether returned pointers/objects need to be freed/destroyed by the user or not, this information should be mentioned here.
(If applicable) a Fortran Note(s): section detailing any relevant differences in calling or using the item from Fortran.
(If applicable) a Developer Note(s): section detailing any relevant information about the code for developers, for example, why a particular algorithm was implemented.
(If applicable) references should be indicated inline with {cite}`Bibtex-key` where Bibtex-key is in the file doc/petsc.bib, as in the manual page for PCFIELDSPLIT.
.seealso: (no newline, no spaces to the left of this text), followed by a list of related manual pages. These manual pages should usually also point back to this manual page in their seealso: sections. This is the final entry in the comment. There should be no blank line after the .seealso: items.
All PETSc functions that appear in a manual page (except the one in the header at the top) should end with a () and be enclosed in single back tick marks. All PETSc enum types and macros etc, should also be enclosed in single back tick marks. This includes each item listed in the .seealso: lines.

Spelling and Capitalization#

Proper nouns, including Unix, Linux, X Windows, and Microsoft Windows, should be fully written and capitalized. This includes all operating systems. The Apple computer operating system is written as macOS.
Company names and product names should be capitalized.
Company names and terms that are traditionally all capitalized, for example, MATLAB, NVIDIA, and CUDA, should be all capitalized.
ARM is a family of processor designs, while Arm is the company that licenses them.
Unix should not be all capitalized.
Microsoft Windows should always be written out with two words. That is, it should not be shortened to Windows or MS Win etc.
CMake should be capitalized as shown.
BLAS and LAPACK are written in full capitalization.
Open MPI is written as two words.

References#

[Gro95a]

W Gropp. Users manual for doctext: producing documentation from source code. Technical Report ANL/MCS-TM-206, Argonne National Laboratory, 1995.

[Gro95b]

W Gropp. Users manual for bfort: producing Fortran interfaces to C source code. Technical Report ANL/MCS-TM-208, Argonne National Laboratory, 1995.