STYLE.md

C++ coding style

• The Envoy source code is formatted using clang-format. Thus all white space, etc. issues are taken care of automatically. The Travis tests will automatically check the code format and fail. There are make targets that can both check the format (check_format) as well as fix the code format for you (fix_format).
• Beyond code formatting, for the most part Envoy uses the Google C++ style guidelines. The following section covers the major areas where we deviate from the Google guidelines.

Deviations from Google C++ style guidelines

• Exceptions are allowed and encouraged where appropriate. When using exceptions, do not add additional error handing that cannot possibly happen in the case an exception is thrown.
• References are always preferred over pointers when the reference cannot be null. This includes both const and non-const references.
• Function names should all use camel case starting with a lower case letter (e.g., doFoo()).
• Struct/Class member variables have a _ postfix (e.g., int foo_;).
• Enum values using PascalCase (e.g., RoundRobin).
• 100 columns is the line limit.
• Use your GitHub name in TODO comments, e.g. TODO(foobar): blah.
• Smart pointers are type aliased:
  • typedef std::unique_ptr<Foo> FooPtr;
  • typedef std::shared_ptr<Bar> BarSharedPtr;
  • typedef std::shared_ptr<const Blah> BlahConstSharedPtr;
  • Regular pointers (e.g. int* foo) should not be type aliased.
• If move semantics are intended, prefer specifying function arguments with &&. E.g., void onHeaders(Http::HeaderMapPtr&& headers, ...). The rationale for this is that it forces the caller to specify std::move(...) or pass a temporary and makes the intention at the callsite clear. Otherwise, it's difficult to tell if a const reference is actually being passed to the called function. This is true even for std::unique_ptr.
• Prefer unique_ptr over shared_ptr wherever possible. unique_ptr makes ownership in production code easier to reason about. Note that this creates some test oddities where production code requires a unique_ptr but the test must still have access to the memory the production code is using (mock or otherwise). In these cases it is acceptable to allocate raw memory in a test and return it to the production code with the expectation that the production code will hold it in a unique_ptr and free it. Envoy uses the factory pattern quite a bit for these cases. (Search the code for "factory").
• The Google C++ style guide points out that non-PoD static and global variables are forbidden. This includes types such as std::string. We encourage the use of the advice in the C++ FAQ on the static initialization fiasco for how to best handle this.
• API-level comments should follow normal Doxygen conventions. Use @param to describe parameters and @return <return-type> for return values. Internal comments for methods and member variables may be regular C++ // comments or Doxygen at developer discretion. Where possible, methods should have meaningful documentation on expected input and state preconditions.
• Header guards should use #pragma once.
• All code should be inside a top-level Envoy namespace. There are some exceptions such as main() functions. When code cannot be placed inside the Envoy namespace there should be a comment of the form // NOLINT(namespace-envoy) at the top of the file.
• If a method that must be defined outside the test directory is intended to be called only from test code then it should have a name that ends in ForTest() such as aMethodForTest(). In most cases tests can and should be structured so this is not necessary.
• Tests default to StrictMock so will fail if hitting unexpected warnings. Feel free to use NiceMock for mocks whose behavior is not the focus of a test.
• There are probably a few other things missing from this list. We will add them as they are brought to our attention.

Error handling

A few general notes on our error handling philosophy:

• All error code returns should be checked.
• At a very high level, our philosophy is that errors that are likely to happen should be gracefully handled. Examples of likely errors include any type of network error, disk IO error, bad data returned by an API call, bad data read from runtime files, etc. Errors that are unlikely to happen should lead to process death, under the assumption that the additional burden of defensive coding and testing is not an effective use of time for an error that should not happen given proper system setup. Examples of these types of errors include not being able to open the shared memory region, an invalid initial JSON config read from disk, system calls that should not fail assuming correct parameters (which should be validated via tests), etc. Examples of system calls that should not fail when passed valid parameters include most usages of setsockopt(), getsockopt(), the kernel returning a valid sockaddr after a successful call to accept(), pthread_create(), pthread_join(), etc.
• OOM events (both memory and FDs) are considered fatal crashing errors. An OOM error should never silently be ignored and should crash the process either via the C++ allocation error exception, an explicit RELEASE_ASSERT following a third party library call, or an obvious crash on a subsequent line via null pointer dereference. This rule is again based on the philosophy that the engineering costs of properly handling these cases is not worth it. Time is better spent designing proper system controls that shed load if resource usage becomes too high, etc.
• The "less is more" error handling philosophy described in the previous two points is primarily based on the fact that restarts are designed to be fast, reliable and cheap.
• Although we strongly recommend that any type of startup error leads to a fatal error, since this is almost always a result of faulty configuration which should be caught during a canary process, there may be cases in which we want some classes of startup errors to be non-fatal. For example, if a misconfigured option is not necessary for server operation. Although this is discouraged, we will discuss these on a case by case basis case basis during code review (an example of this is the --admin-address-path option). If degraded mode error handling is implemented, we require that there is complete test coverage for the degraded case. Additionally, the user should be aware of the degraded state minimally via an error log of level warn or greater and via the increment of a stat.
• The error handling philosophy described herein is based on the assumption that Envoy is deployed using industry best practices (primarily canary). Major and obvious errors should always be caught in canary. If a low rate error leads to periodic crash cycling when deployed to production, the error rate should allow for rollback without large customer impact.
• Tip: If the thought of adding the extra test coverage, logging, and stats to handle an error and continue seems ridiculous because "this should never happen", it's a very good indication that the appropriate behavior is to terminate the process and not handle the error. When in doubt, please discuss.
• Per above it's acceptable to turn failures into crash semantics via RELEASE_ASSERT(condition) or PANIC(message) if there is no other sensible behavior, e.g. in OOM (memory/FD) scenarios. Only RELEASE_ASSERT(condition) should be used to validate conditions that might be imposed by the external environment. ASSERT(condition) should be used to document (and check in debug-only builds) program invariants. Use ASSERT liberally, but do not use it for things that will crash in an obvious way in a subsequent line. E.g., do not do ASSERT(foo != nullptr); foo->doSomething();. Note that there is a gray line between external environment failures and program invariant violations. For example, memory corruption due to a security issue (a bug, deliberate buffer overflow etc.) might manifest as a violation of program invariants or as a detectable condition in the external environment (e.g. some library returning a highly unexpected error code or buffer contents). Unfortunately no rule can cleanly cover when to use RELEASE_ASSERT vs. ASSERT. In general we view ASSERT as the common case and RELEASE_ASSERT as the uncommon case, but experience and judgment may dictate a particular approach depending on the situation.

Hermetic and deterministic tests

Tests should be hermetic, i.e. have all dependencies explicitly captured and not depend on the local environment. In general, there should be no non-local network access. In addition:

• Port numbers should not be hardcoded. Tests should bind to port zero and then discover the bound port when needed. This avoids flakes due to conflicting ports and allows tests to be executed concurrently by Bazel. See test/integration/integration_test.h and test/common/network/listener_impl_test.cc for examples of tests that do this.

• Paths should be constructed using:

  • The methods in TestEnvironment for C++ tests.
  • With ${TEST_TMPDIR} (for writable temporary space) or ${TEST_RUNDIR} for read-only access to test inputs in shell tests.
  • With {{ test_tmpdir }}, {{ test_rundir }} and {{ test_udsdir }} respectively for JSON templates. {{ test_udsdir }} is provided for pathname based Unix Domain Sockets, which must fit within a 108 character limit on Linux, a property that might not hold for {{ test_tmpdir }}.

Tests should be deterministic. They should not rely on randomness or details such as the current time. Instead, mocks such as MockRandomGenerator and Mock*TimeSource should be used.

Google style guides for other languages

• Python
• Bash
• Bazel