DEVELOPING.md

Developing

Getting started

In order to test and develop in this repo you will need the following dependencies installed:

• docker
• make

After cloning do the following:

1. run make bootstrap to download go mod dependencies, create the /.tmp dir, and download helper utilities.
2. run make to run linting, tests, and other verifications to make certain everything is working alright.

Checkout make help to see what other actions you can take.

The main make tasks for common static analysis and testing are lint, lint-fix, unit, integration, and cli.

Levels of testing

• unit: The default level of test which is distributed throughout the repo are unit tests. Any _test.go file that does not reside somewhere within the /test directory is a unit test. Other forms of testing should be organized in the /test directory. These tests should focus on correctness of functionality in depth. % Test coverage metrics only considers unit tests and no other forms of testing.

• integration: located within test/integration, these tests focus on the behavior surfaced by the common library entrypoints from the syft package and make light assertions about the results surfaced. Additionally, these tests tend to make diversity assertions for enum-like objects, ensuring that as enum values are added to a definition that integration tests will automatically fail if no test attempts to use that enum value. For more details see the "Data diversity and freshness assertions" section below.

• cli: located with in test/cli, these are tests that test the correctness of application behavior from a snapshot build. This should be used in cases where a unit or integration test will not do or if you are looking for in-depth testing of code in the cmd/ package (such as testing the proper behavior of application configuration, CLI switches, and glue code before syft library calls).

• acceptance: located within test/compare and test/install, these are smoke-like tests that ensure that application
  packaging and installation works as expected. For example, during release we provide RPM packages as a download artifact. We also have an accompanying RPM acceptance test that installs the RPM from a snapshot build and ensures the output of a syft invocation matches canned expected output. New acceptance tests should be added for each release artifact and architecture supported (when possible).

Data diversity and freshness assertions

It is important that tests against the codebase are flexible enough to begin failing when they do not cover "enough" of the objects under test. "Cover" in this case does not mean that some percentage of the code has been executed during testing, but instead that there is enough diversity of data input reflected in testing relative to the definitions available.

For instance, consider an enum-like value like so:

type Language string

const (
  Java            Language = "java"
  JavaScript      Language = "javascript"
  Python          Language = "python"
  Ruby            Language = "ruby"
  Go              Language = "go"
)

Say we have a test that exercises all the languages defined today:

func TestCatalogPackages(t *testing.T) {
  testTable := []struct {
    // ... the set of test cases that test all languages
  }
  for _, test := range cases {
    t.Run(test.name, func (t *testing.T) {
      // use inputFixturePath and assert that syft.CatalogPackages() returns the set of expected Package objects
      // ...
    })
  }
}

Where each test case has a inputFixturePath that would result with packages from each language. This test is brittle since it does not assert that all languages were exercised directly and future modifications (such as adding a new language) won't be covered by any test cases.

To address this the enum-like object should have a definition of all objects that can be used in testing:

type Language string

// const( Java Language = ..., ... )

var AllLanguages = []Language{
	Java,
	JavaScript,
	Python,
	Ruby,
	Go,
	Rust,
}

Allowing testing to automatically fail when adding a new language:

func TestCatalogPackages(t *testing.T) {
  testTable := []struct {
  	// ... the set of test cases that (hopefully) covers all languages
  }

  // new stuff...
  observedLanguages := strset.New()
  
  for _, test := range cases {
    t.Run(test.name, func (t *testing.T) {
      // use inputFixturePath and assert that syft.CatalogPackages() returns the set of expected Package objects
    	// ...
    	
    	// new stuff...
    	for _, actualPkg := range actual {
        observedLanguages.Add(string(actualPkg.Language))
    	}
    	
    })
  }

   // new stuff...
  for _, expectedLanguage := range pkg.AllLanguages {
    if 	!observedLanguages.Contains(expectedLanguage) {
      t.Errorf("failed to test language=%q", expectedLanguage)	
    }
  }
}

This is a better test since it will fail when someone adds a new language but fails to write a test case that should exercise that new language. This method is ideal for integration-level testing, where testing correctness in depth is not needed (that is what unit tests are for) but instead testing in breadth to ensure that units are well integrated.

A similar case can be made for data freshness; if the quality of the results will be diminished if the input data is not kept up to date then a test should be written (when possible) to assert any input data is not stale.

An example of this is the static list of licenses that is stored in internal/spdxlicense for use by the SPDX presenters. This list is updated and published periodically by an external group and syft can grab and update this list by running go generate ./... from the root of the repo.

An integration test has been written to grabs the latest license list version externally and compares that version with the version generated in the codebase. If they differ, the test fails, indicating to someone that there is an action needed to update it.

_The key takeaway is to try and write tests that fail when data assumptions change and not just when code changes._

Snapshot tests

The format objects make a lot of use of "snapshot" testing, where you save the expected output bytes from a call into the git repository and during testing make a comparison of the actual bytes from the subject under test with the golden copy saved in the repo. The "golden" files are stored in the test-fixtures/snapshot directory relative to the go package under test and should always be updated by invoking go test on the specific test file with a specific CLI update flag provided.

Many of the Format tests make use of this approach, where the raw SBOM report is saved in the repo and the test compares that SBOM with what is generated from the latest presenter code. For instance, at the time of this writing the CycloneDX presenter snapshots can be updated by running:

go test ./internal/formats -update-cyclonedx

These flags are defined at the top of the test files that have tests that use the snapshot files.

Snapshot testing is only as good as the manual verification of the golden snapshot file saved to the repo! Be careful and diligent when updating these files.

Architecture

TODO: outline:

• analysis creates a static SBOM which can be encoded and decoded.
• format objects, should strive to not add or enrich data in encoding that could otherwise be done during analysis
• pkg.Catalogers
• file catalogers
• source.Source
• file.Resolvers
• logger abstraction
• events / bus abstraction