Impala C++ Code follows a modified version of the Google C++ Style Guide at:

https://google.github.io/styleguide/cppguide.html

There are some key differences which are documented here.

Header Files

  1. Separate inline headers: We allow putting inline functions in separate files, using the suffix .inline.h. This can speed up compile times by reducing the volume of code to be compiled and reducing dependencies between headers.
  2. Header include guards: for new code, we prefer #pragma once because it is cleaner than the classic #include guards

Variable Naming

We use UPPER_CASE for constants instead of kConstantName.

Scoping

  1. namespaces the using namespace directive is allowed in .cc files in limited cases where it greatly reduces code volume.
    1. Pros: reduces code volume, less churn in "using namespace::class" directives.
    2. Cons: pollutes the namespace, causes conflicts, makes it more difficult to determine the type of an object

Commenting

  • TODO Comments we typically do not include the name in the TODO comment. Where possible it's best to fix the TODO immediately (if it is a small task) or create a JIRA to track a more substantial improvement.

Formatting

The .clang-format file checked into the Impala repository should be used to format whitespace (see Contributing to Impala for more info). The .clang-format file is the source of truth for whitespace formatting, except when its output significantly diverges from practices in the existing codebase or from common sense. In those cases .clang-format should be updated. We aim to gradually adapt the codebase to the output of clang-format Therefore we only recommend using it on lines that have non-whitespace changes. This can be accomplished with the git-clang-format tool.


Some key differences from the Google C++ style are:

  1. Line Length We use 90 character line lengths
  2. Function Declaration we line wrap differently than Google, typically packing more parameters per line, e.g.:
// Google Recommends:
ReturnType LongClassName::ReallyReallyReallyLongFunctionName(
    Type par_name1,  // 4 space indent
    Type par_name2,
    Type par_name3) {
  DoSomething();  // 2 space indent
  ...
}
// we use:
ReturnType LongClassName::ReallyReallyReallyLongFunctionName(
    Type par_name1, Type par_name2, Type par_name3) { // 4 space indent
  DoSomething();  // 2 space indent
  ...
}


    3. Conditionals we format conditionals as follows

// Google Recommends:
if (x == kFoo) return new Foo();
if (condition)
  DoSomething();  // 2 space indent.
if (condition) {
  DoSomething();  // 2 space indent.
}
// we only use:
if (x == kFoo) return new Foo();  // If the whole line fits into the 90 character limit
if (condition) {
  DoSomething();  // Otherwise, 2 space indent.
}

Re-formatting

You can use clang-format to automatically reformat only the parts of the code you touched. See https://clang.llvm.org/docs/ClangFormat.html#script-for-patch-reformatting for a general clang-format reference. Impala's clang-format file is located at ${IMPALA_HOME}/.clang-format. The clang-* command uses the format file. 

To reformat the code around the lines that your last commit touched, you can run:

git diff -U0 --no-color HEAD^ | "${IMPALA_TOOLCHAIN_PACKAGES_HOME}/llvm-${IMPALA_LLVM_VERSION}/share/clang/clang-format-diff.py" -binary "${IMPALA_TOOLCHAIN_PACKAGES_HOME}/llvm-${IMPALA_LLVM_VERSION}/bin/clang-format" -p1 -i


The -i option means in-place replacement. Run it without -i option if you'd like to see difference before applying the effect of the command.

If the above approach does not work or you may want to use a different version, you can use a system version of clang-tidy. On Ubuntu,

sudo apt-get install clang-format-3.9
git diff -U0 --no-color HEAD^ | clang-format-diff-3.9 -p1 -i



Tip

clang-format is not perfect and sometimes reformats things in strange ways, or will reformat large areas of code. If clang-format doesn't match the existing style of the surrounding code, please consider manually formatting that part of the code - mixing large white

Third-Party Libraries

  • Boost - we use a different set of Boost libraries. Reducing # of dependencies is encouraged and adding dependencies to new libraries should be carefully evaluated.

Tidy Code

You can check that your code is tidy with clang-tidy:

git diff asf-gerrit/master | "${IMPALA_TOOLCHAIN_PACKAGES_HOME}/llvm-${IMPALA_LLVM_VERSION}/share/clang/clang-tidy-diff.py" -clang-tidy-binary "${IMPALA_TOOLCHAIN_PACKAGES_HOME}/llvm-${IMPALA_LLVM_VERSION}/bin/clang-tidy" -p 1




  • No labels

2 Comments

  1. Is this up to date? I'm not sure I have seen examples of:

     

    • "use of anonymous namespaces are not recommended" (I have seen a few)
    • "the using namespace directive is encouraged" (I have only seen this in .cc files)
    • "we do not mark single argument constructors explicit"
    • "we do not implement or delete copy and assignment constructors" - I have seen us use the macros that delete these for you
    • "we extensively use operator overloading"
    • "we use cast formats such as int x = (int) y;" - I have seen (and use) the C++ casts

     

    1. OK, removed some of that stuff.