We follow the Google Java Style Guide.
The build will fail if the source code is not formatted according to the google java style.
The main goal is to avoid extensive reformatting caused by different IDEs having different opinion about how things should be formatted by establishing.
Running
./gradlew spotlessApply
reformats all the files that need reformatting.
Running
./gradlew spotlessCheck
runs formatting verify task only.
To completely delegate code style formatting to the machine, there is a pre-commit hook setup to verify formatting before committing. It can be activated with this command:
git config core.hooksPath .githooks
As additional convenience for IntelliJ users, we provide .editorconfig
file. IntelliJ will automatically use it to adjust its code formatting settings.
It does not support all required rules, so you still have to run
spotlessApply
from time to time.
The build uses checkstyle to verify some parts of the Google Java Style Guide that cannot be handled by auto-formatting.
To run these checks locally:
./gradlew checkstyleMain checkstyleTest
We leverage static imports for many common types of operations. However, not all static methods or constants are necessarily good candidates for a static import. The following list is a very rough guideline of what are commonly accepted static imports:
- Test assertions (JUnit and AssertJ)
- Mocking/stubbing in tests (with Mockito)
- Collections helpers (such as
singletonList()
andCollectors.toList()
) - ByteBuddy
ElementMatchers
(for building instrumentation modules) - Immutable constants (where clearly named)
- Singleton instances (especially where clearly named and hopefully immutable)
tracer()
methods that expose tracer singleton instances- Semantic convention attribute keys used in tests
Some of these are enforced by checkstyle rules:
- look for
RegexpSinglelineJava
incheckstyle.xml
- use
@SuppressWarnings("checkstyle:RegexpSinglelineJava")
to suppress the checkstyle warning
The following order is preferred:
- Static fields (final before non-final)
- Instance fields (final before non-final)
- Constructors
- Methods
- Nested classes
If methods call each other, it's nice if the calling method is ordered (somewhere) above the method that it calls. So, for one example, a private method would be ordered (somewhere) below the non-private methods that use it.
In static utility classes (where all members are static), the private constructor (used to prevent construction) should be ordered after methods instead of before methods.
Public classes should be declared final
where possible.
Methods should only be declared final
if they are in non-final public classes.
Fields should be declared final
where possible.
Method parameters and local variables should never be declared final
.
[Note: this section is aspirational, as opposed to a reflection of the current codebase]
All parameters and fields which can be null
should be annotated with @Nullable
(specifically javax.annotation.Nullable
, which is included by the
otel.java-conventions
gradle plugin as a compileOnly
dependency).
There is no need to use @NonNull
, as this is the default, which should be declared in a
package-info.java
file on the root package of each module, e.g.
@DefaultQualifier(
value = NonNull.class,
locations = {TypeUseLocation.FIELD, TypeUseLocation.PARAMETER, TypeUseLocation.RETURN})
package io.opentelemetry.instrumentation.api;
import org.checkerframework.checker.nullness.qual.NonNull;
import org.checkerframework.framework.qual.DefaultQualifier;
Public APIs should still defensively check for null
parameters, even if the parameter is not
annotated with @Nullable
. Internal APIs do not need to defensively check for null
parameters.
To help enforce @Nullable
annotation usage, the otel.nullaway-conventions
gradle plugin
should be used in all modules to perform basic nullable usage validation:
plugins {
id("otel.nullaway-conventions")
}
Following the reasoning from Writing a Java library with better experience (slide 12),
usage of java.util.Optional
is kept at a minimum in this project.
It is ok to use Optional
in places where it does not leak into public API signatures.
Also, avoid Optional
usage on the hot path (instrumentation code), unless the instrumented library
itself uses it.
Avoid allocations whenever possible on the hot path (instrumentation code).
This includes Iterator
allocations from collections; note that
for(SomeType t : plainJavaArray)
does not allocate an iterator object.
Non-allocating stream api usage on the hot path is acceptable but may not
fit the surrounding code style; this is a judgement call. Note that
some stream apis make it much more difficult to allocate efficiently
(e.g., collect
with presized sink data structures involves
convoluted Supplier
code, or lambdas passed to forEach
might be
capturing/allocating lambdas).