编码规范指南

对于提交到Trafodion的C++代码，这些编码规范指南是基本的验收标准，Trafodion代码评审者将使用这些指南来评审代码改动。

此指南表述了必须的或者希望的最佳事件，同时也描述了那些针对不同是实践经验没有偏好的地方。

Trafodion有几个不同的子项目组成，有些在使用一些与Trafodion标准不同的编码规范；例如：a requirement in most areas of the code may be only preferred in others.

也有可能有已经存在的代码违背了标准中的一条或者几条要求。但是标准的目的是随着时间的推移更正这些，因此鼓励大家当在解决一个缺陷或者开发一个新功能的时候更正看到的那些不复合规范的地方。然而不推荐只是针对编码规范的改动，因为这会儿会给评审者增加额外的负担。

头文件

在头文件中尽量减少#include指令的使用。 Forward declare classes and structs when the only use within the header file is a pointer or reference.

While includes should be kept to a minimum, if something is used in a header file and cannot be forward declared, it must be explicitly included. All files, headers and implementation, should include everything they need to be self-sufficient. They should never assume something will be pre-included. In other words, the contents of a header file should compile cleanly by itself. To help ensure this, all implementation files should include their respective header file first.

Header files should not contain declarations for public items that are only used by the implementation file. Public items that are require in the implementation file but not the header file should be declared in the implementation file.

The preference is NOT to create header files that consist only of includes of other header files.

Declare as little as possible in the header file and keep as much of the actual implementation private as is reasonable. For instance, don’t include in a header file declarations of types, enums, and functions that are only referenced by the implementation file.

Including standard header files

The preference is for C++ style includes over C style includes for standard header files.

//Preferred

#include <cstdio>

//Accepted

#include <stdio.h>

Include Guards

All header files must use include guards. The name of the include guard #define should be the filename in all uppercase, with underscore used in place of periods.

For instance, if the header file is named ServerInterface_ODBC.h, the header file should begin and end as follows:

#ifndef SERVERINTERFACE_ODBC_H
#define SERVERINTERFACE_ODBC_H
...
#endif /* SERVERINTERFACE_ODBC_H */

Comments following the #endif indicating the include guard are preferred.

Variable Declaration and Naming Standards

Trafodion uses a combination of Pascal and Camel case. Pascal case means that the first letter in each word in an identifier is capitalized. Camel case is similar except the first letter is in lower case. For both, underscores are NOT used to separate words. The general rule is that identifiers with local scope start with a lower case letter and identifiers with global scope start with an upper case letter.

//Pascal case
class AuthenticationMessage; 

//Camel case (aka lower Camel case or camelCase)
int logonCount; 

Class Names

Class names should be Pascal case and should describe the object contents (not what it does), with as little abbreviation as possible. When names include acronyms, the acronyms should be in all upper case.

Acceptable examples:

//Preferred

class SQLSessionContext; // an object that contains the context for a SQL session
class PrivilegeList; // a list of privileges

Poor examples:

class OutputInfo; // Doesn’t describe class contents, no context
class ReadTableDef; // Describes what class does, not contents
class Cmdline_Args; // Prefer Pascal case, no underscore, for class names

Class Member Variable Names

Private member data variables should be suffixed with an underscore and should use Camel case. When names include acronyms, the acronyms should be in all upper or all lower case, dependent on which case the first letter should be:

Example:

class Employee
{
public:
   Employee ();
private:
   std::string firstName_; 
   std::string lastName_; 
   uint16_t    departmentNumber_;
   std::string departmentName_;
   uint32_t    irsSSN_;
}

Function Names

Class member functions and static file functions should use Camel case. External non-class functions should use Pascal case. Except for constructors, destructors and operators, the function name should include a verb that describes the action the function is performing.

Good examples:

//Class member functions
int32_t getSalary() const;
int32_t setAuthID();
int32_t closeAllCursors();

Bad examples:

// Is it setting break enabled, returning it or ???
int32_t SQLCLI_BreakEnabled();

Enums

Enum types should use Pascal case and describe the class of enums. If the enum is declared outside of a class, the type name should include an indication of the scope of the enums.

Enums themselves should be declared as all upper case. The names may begin with a common prefix or be independent, depending on the usage.

When enums represent an arbitrary set of return values (i.e. error codes, state codes, etc.), avoid the values -1, 0, and 1 if using weakly typed enums, to reduce the chance of matches with Booleans or uninitialized variables.

The preference is to declare enums as strongly typed (enum class EnumName {...};)

Boolean Variables

Boolean variables names should include a verb, state, and optionally a noun (object whose state is in question) indicating the nature of the Boolean. Any combination is acceptable, however verbState is the most common.

Good examples:

bool isValid;           // verbState
bool isValidTable;      // verbStateNoun
bool tableIsDroppable;  // nounVerbState
bool hasData;           // verbState

Bad examples:

bool valid;
bool tableState;
bool empty;

Functions that return a Boolean should also have names of the form verbState or verbStateNoun if the functions return state information. (This naming standard does not apply to functions returning Boolean as indication of success or failure.)

Good examples:

bool isValidHbaseName();
bool isHostNameExcluded();
bool canUseCbServer();

Bad examples:

// Don’t use get for Boolean accessors
bool getUDRAccessModeViolation(); 

// Don’t use integer return for Boolean functions
short existsInHBase();

// Function name implies it is sending settings to the compiler, but it is
// actually only returning an indication that settings should be sent.  
// A better name would be shouldSendSettingsToCompiler().
bool sendSettingsToCompiler();

Parts of Trafodion code use one of two Boolean typedefs, NABoolean and ComBoolean, declared as follows:

typedef int   Int32;
typedef Int32 NABoolean;
typedef NABoolean ComBoolean;
const NABoolean  TRUE = (1 == 1);
const NABoolean  FALSE = (0 == 1);

Exercise care when mixing usage of bool and NABoolean/ComBoolean types, as the latter are not guaranteed to only contain values of TRUE and FALSE. The use of non-standard Boolean types is gradually being phased out.

Constants

All constant names should be all uppercase, regardless of how the constant is declared. That is, enums, defines and variables with the const modifier should be named in all uppercase.

Defines, enums, and const are all permitted and used throughout Trafodion, although most code in Trafodion uses enum for numerical constants and defines for character constants. For new code, the use of const char or string is preferred for character constants instead of defines.

Namespace Names

The preference is for namespaces to be all lowercase, with preference to single words (note the exception to the rule that a name with global scope should start with an upper case). If a namespace must be dual-worded, use underscores. If mixed case names are used, Pascal case is preferred.

Examples:

//Preferred

namespace compiler 

//Accepted

namespace Compiler

Indentation and Formatting

Indentation

TAB characters are not permitted in source files except for text files (e.g. makefiles) that require them.

Trafodion code uses several indenting depths, including 2, 3, 4, and 8 spaces. Most common is 2 and 3. Use the style found in existing code, and when writing new code, use either 2, 3, or 4, and remain consistent.

A variety of control block indentation styles are used throughout Trafodion, most commonly Allman, Whitesmith, Stroustrup, and GNU. Follow the predominant style when making small to medium changes to existing code. For new code, the Allman style is preferred.

//Allman
   if (x > 5)
   {
      error = doThis(x);
      if (error != 0)
      {
         return false;
      }
   }
   else
   {
      doThat(x);
   }

//Whitesmith
   if (x > 5)
      {
      error = doThis(x);
      if (error != 0)
         {
         return false;
         }
      }
   else
      {
      doThat(x);
      }

//Stroustrup 
   if (x > 5) {
      error = doThis(x);
      if (error != 0) {
         return false;
      }
   }
   else {
      doThat(x);
   }

Note that the Stroustrup and the similar K&R formats were popularized by usage in books where conservation of line count was a goal.

//GNU
  if (x > 5)
    {
      error = doThis(x);
      if (error != 0)
        {
          return false;
        }
    }
  else
    {
      doThat(x);
    }

Comments

Comment Style

C++ style comments are preferred, but C comments are acceptable as well.

Some code uses Doxygen style comments, but this is not required.

When/Where Comments Should Be Used

Every file should have a comment at the beginning describing the purpose of the file.

In header files where classes are declared, there should be a comment describing the class, including purpose and usage. Also describe anything out of the ordinary, such as the use of multiple inheritance.

Within implementation files, in addition to the comment at the beginning of the file, add comments for any global or static variables defined in the file, and how the variable is handled in a multi-threaded environment (if applicable).

Also, for each function defined, describe the purpose and intent of the function. For each parameter, list whether it is input or output (or both), how it is used, and any range restrictions imposed. For functions not returning void, describe the possible return values.

Within the body of the function, there is no need to write comments that document the obvious. But if there is any complexity to the logic, at a minimum document the intent, and consider documenting the details (assumptions, limits, unexpected side effects from function calls, etc.)

If a feature is only partially implemented, add a comment indicating at a high level what work remains. Prefix the comment with //TODO.

//TODO Code is currently not thread safe.  Need to protect allocation of ...

Error Handling

Asserts

Trafodion uses asserts in the "debug" build. Use asserts freely, but ensure they do not contain any side effects as the code is not present in the "release" build.

Error Return/Retrieval

Avoid the use of integer return codes for success and error codes. Instead, use bool for simple succeeded/failed and enum types for returns with multiple conditions.

Exception Handling

Trafodion code mixes usage of exceptions and error returns. When using try/catch blocks, keep the scope as small as possible. Ensure all exceptions thrown are handled, potentially in main() if nowhere else.

General Guidelines

Casting

Avoid using C-style casts. Use C++-style casts instead, as they are often safer and easier to search for.

//Preferred

int x = static_cast<int>(shortVariable);

MyType *myVar = reinterpret_cast<MyType *>(voidPtr);

Don’t blindly cast to remove a compiler error or warning. Ensure the cast is safe.

Use const casting sparingly. Often the use of mutable or changing a function to be const correct is a better solution.

Types

Use standard types defined in <cstdint> or <stdint.h>. Note, Trafodion defines and uses many non-standard types (e.g. Int32, Lng32), but this usage is being phased out.

Use types with explicit sizes (e.g. int32_t, int64_t) when size is a factor in the code, such as an external API, or column in a table. Where size is not a factor (counters, indexes) a non-sized type such as int or long may be used. However, in general, size_t and ssize_t are preferred for variables where fixed size is not required.