syft/DEVELOPING.md

Developing

Getting started

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

• Golang
• docker
• make
• Python (>= 3.9)

Docker settings for getting started

Make sure you've updated your docker settings so the default docker socket path is available.

Go to:

docker -> settings -> advanced

Make sure:

Allow the default Docker socket to be used

is checked.

Also double check that the docker context being used is the default context. If it is not, run:

docker context use default

After cloning the following step can help you get setup:

1. run make bootstrap to download go mod dependencies, create the /.tmp dir, and download helper utilities.
2. run make to view the selection of developer commands in the Makefile
3. run make build to build the release snapshot binaries and packages
4. for an even quicker start you can run go run cmd/syft/main.go to print the syft help.
   • this command go run cmd/syft/main.go alpine:latest will compile and run syft against alpine:latest
5. view the README or syft help output for more output options

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

See make help for all the current make tasks.

Internal Artifactory Settings

Not always applicable

Some companies have Artifactory setup internally as a solution for sourcing secure dependencies. If you're seeing an issue where the unit tests won't run because of the below error then this section might be relevant for your use case.

[ERROR] [ERROR] Some problems were encountered while processing the POMs

If you're dealing with an issue where the unit tests will not pull/build certain java fixtures check some of these settings:

• a settings.xml file should be available to help you communicate with your internal artifactory deployment
• this can be moved to syft/pkg/cataloger/java/test-fixtures/java-builds/example-jenkins-plugin/ to help build the unit test-fixtures
• you'll also want to modify the build-example-jenkins-plugin.sh to use settings.xml

For more information on this setup and troubleshooting see issue 1895

Architecture

At a high level, this is the package structure of syft:

./cmd/syft/
│   ├── cli/
│   │   ├── cli.go          // where all commands are wired up
│   │   ├── commands/       // all command implementations
│   │   ├── options/        // all command flags and configuration options
│   │   └── ui/             // all handlers for events that are shown on the UI
│   └── main.go             // entrypoint for the application
└── syft/                   // the "core" syft library
    ├── format/             // contains code to encode or decode to and from SBOM formats
    ├── pkg/                // contains code to catalog packages from a source
    ├── sbom/               // contains the definition of an SBOM
    └── source/             // contains code to create a source object for some input type (e.g. container image, directory, etc)

Syft's core library is implemented in the syft package and subpackages, where the major packages are:

• the syft/source package produces a source.Source object that can be used to catalog a directory, container, and other source types.
• the syft package contains a single function that can take a source.Source object and catalog it, producing an sbom.SBOM object
• the syft/format package contains the ability to encode and decode SBOMs to and from different SBOM formats (such as SPDX and CycloneDX)

The cmd pacakge at the highest level execution flow wires up spf13/cobra commands for execution in the main application:

sequenceDiagram
    participant main as cmd/syft/main
    participant cli as cli.New()
    participant root as root.Execute()
    participant cmd as <command>.Execute()

    main->>+cli: 

    Note right of cli: wire ALL CLI commands
    Note right of cli: add flags for ALL commands

    cli-->>-main:  root command 

    main->>+root: 
    root->>+cmd: 
    cmd-->>-root: (error)  

    root-->>-main: (error) 

    Note right of cmd: Execute SINGLE command from USER

The packages command uses the core library to generate an SBOM for the given user input:

sequenceDiagram
    participant source as source.New(ubuntu:latest)
    participant sbom as sbom.SBOM
    participant catalog as syft.CatalogPackages(src)
    participant encoder as syft.Encode(sbom, format)

    Note right of source: use "ubuntu:latest" as SBOM input

    source-->>+sbom: add source to SBOM struct
    source-->>+catalog: pass src to generate catalog
    catalog-->-sbom: add cataloging results onto SBOM
    sbom-->>encoder: pass SBOM and format desiered to syft encoder
    encoder-->>source: return bytes that are the SBOM of the original input 

    Note right of catalog: cataloger configuration is done based on src

Additionally, here is a gist of using syft as a library to generate a SBOM for a docker image.

pkg.Package object

The pkg.Package object is a core data structure that represents a software package. Fields like name and version probably don't need a detailed explanation, but some of the other fields are worth a quick overview:

• FoundBy: the name of the cataloger that discovered this package (e.g. python-pip-cataloger).
• Locations: these are the set of paths and layer ids that were parsed to discover this package (e.g. python-pip-cataloger).
• Language: the language of the package (e.g. python).
• Type: this is a high-level categorization of the ecosystem the package resides in. For instance, even if the package is a egg, wheel, or requirements.txt reference, it is still logically a "python" package. Not all package types align with a language (e.g. rpm) but it is common.
• Metadata: specialized data for specific location(s) parsed. We should try and raise up as much raw information that seems useful. As a rule of thumb the object here should be as flat as possible and use the raw names and values from the underlying source material parsed.

When pkg.Package is serialized an additional MetadataType is shown. This is a label that helps consumers understand the datashape of the Metadata field.

By convention the MetadataType value should follow these rules of thumb:

• Only use lowercase letters, numbers, and hyphens. Use hyphens to separate words.
• Try to anchor the name in the ecosystem, language, or packaging tooling it belongs to. For a package manager for a language ecosystem the language, framework or runtime should be used as a prefix. For instance pubspec-lock is an OK name, but dart-pubspec-lock is better. For an OS package manager this is not necessary (e.g. apk-db-entry is a good name, but alpine-apk-db-entry is not since alpine and the a in apk is redundant).
• Be as specific as possible to what the data represents. For instance ruby-gem is NOT a good MetadataType value, but ruby-gemspec is. Why? Ruby gem information can come from a gemspec file or a Gemfile.lock, which are very different. The latter name provides more context as to what to expect.
• Should describe WHAT the data is, NOT HOW it's used. For instance r-description-installed-file is NOT a good MetadataType value since it's trying to convey that we use the DESCRIPTION file in the R ecosystem to detect installed packages. Instead simply describe what the DESCRIPTION file is itself without context of how it's used: r-description.
• Use the lock suffix to distinct between manifest files that loosely describe package version requirements vs files that strongly specify one and only one version of a package ("lock" files). These should only be used with respect to package managers that have the guide and lock distinction, but would not be appropriate otherwise (e.g. rpm does not have a guide vs lock, so lock should NOT be used to describe a db entry).
• Use the archive suffix to indicate a package archive (e.g. rpm file, apk file, etc) that describes the contents of the package. For example an RPM file that was cataloged would have a rpm-archive metadata type (not to be confused with an RPM DB record entry which would be rpm-db-entry).
• Use the entry suffix to indicate information about a package that was found as a single entry within file that has multiple package entries. If the entry was found within a DB or a flat-file store for an OS package manager, you should use db-entry.
• Should NOT contain the phrase package, though exceptions are allowed (say if the canonical name literally has the phrase package in it).
• Should NOT contain have a file suffix unless the canonical name has the term "file", such as a pipfile or gemfile. An example of a bad name for this rule isruby-gemspec-file; a better name would be ruby-gemspec.
• Should NOT contain the exact filename+extensions. For instance pipfile.lock shouldn't really be in the name, instead try and describe what the file is: python-pipfile-lock (but shouldn't this be python-pip-lock you might ask? No, since the pip package manger is not related to the pipfile project).
• Should NOT contain the phrase metadata, unless the canonical name has this term.
• Should represent a single use case. For example, trying to describe Hackage metadata with a single HackageMetadata struct (and thus MetadataType) is not allowed since it represents 3 mutually exclusive use cases: representing a stack.yaml, stack.lock, or cabal.project file. Instead, each of these should have their own struct types and MetadataType values.

There are other cases that are not covered by these rules... and that's ok! The goal is to provide a consistent naming scheme that is easy to understand and use when it's applicable. If the rules do not exactly apply in your situation then just use your best judgement (or amend these rules as needed whe new common cases come up).

What if the underlying parsed data represents multiple files? There are two approaches to this:

• use the primary file to represent all the data. For instance, though the dpkg-cataloger looks at multiple files to get all information about a package, it's the status file that gets represented.
• nest each individual file's data under the Metadata field. For instance, the java-archive-cataloger may find information from on or all of the files: pom.xml, pom.properties, and MANIFEST.MF. However, the metadata is simply `java-metadata' with each possibility as a nested optional field.

Syft Catalogers

Catalogers are the way in which syft is able to identify and construct packages given a set a targeted list of files. For example, a cataloger can ask syft for all package-lock.json files in order to parse and raise up javascript packages (see how file globs and file parser functions are used for a quick example).

From a high level catalogers have the following properties:

• They are independent from one another. The java cataloger has no idea of the processes, assumptions, or results of the python cataloger, for example.

• They do not know what source is being analyzed. Are we analyzing a local directory? an image? if so, the squashed representation or all layers? The catalogers do not know the answers to these questions. Only that there is an interface to query for file paths and contents from an underlying "source" being scanned.

• Packages created by the cataloger should not be mutated after they are created. There is one exception made for adding CPEs to a package after the cataloging phase, but that will most likely be moved back into the cataloger in the future.

Cataloger names should be unique and named with the following rules of thumb in mind:

• Must end with -cataloger
• Use lowercase letters, numbers, and hyphens only
• Use hyphens to separate words
• Catalogers for language ecosystems should start with the language name (e.g. python- for a cataloger that raises up python packages)
• Distinct between when the cataloger is searching for evidence of installed packages vs declared packages. For example, there are currently two different gemspec-based catalogers, the ruby-gemspec-cataloger and ruby-installed-gemspec-cataloger, where the latter requires that the gemspec is found within a specifications directory (which means it was installed, not just at the root of a source repo).

Building a new Cataloger

Catalogers must fulfill the pkg.Cataloger interface in order to add packages to the SBOM. All catalogers should be added to:

• the global list of catalogers
• at least one source-specific list, today the two lists are directory catalogers and image catalogers

For reference, catalogers are invoked within syft one after the other, and can be invoked in parallel.

generic.NewCataloger is an abstraction syft used to make writing common components easier (see the apkdb cataloger for example usage). It takes the following information as input:

• A catalogerName to identify the cataloger uniquely among all other catalogers.
• Pairs of file globs as well as parser functions to parse those files. These parser functions return a slice of pkg.Package as well as a slice of artifact.Relationship to describe how the returned packages are related. See this the apkdb cataloger parser function as an example.

Identified packages share a common pkg.Package struct so be sure that when the new cataloger is constructing a new package it is using the Package struct. If you want to return more information than what is available on the pkg.Package struct then you can do so in the pkg.Package.Metadata section of the struct, which is unique for each pkg.Type. See the pkg package for examples of the different metadata types that are supported today. These are plugged into the MetadataType and Metadata fields in the above struct. MetadataType informs which type is being used. Metadata is an interface converted to that type.

Finally, here is an example of where the package construction is done within the apk cataloger:

• Calling the APK package constructor from the parser function
• The APK package constructor itself

Interested in building a new cataloger? Checkout the list of issues with the new-cataloger label! If you have questions about implementing a cataloger feel free to file an issue or reach out to us on slack!

Searching for files

All catalogers are provided an instance of the file.Resolver to interface with the image and search for files. The implementations for these abstractions leverage stereoscope in order to perform searching. Here is a rough outline how that works:

1. a stereoscope file.Index is searched based on the input given (a path, glob, or MIME type). The index is relatively fast to search, but requires results to be filtered down to the files that exist in the specific layer(s) of interest. This is done automatically by the filetree.Searcher abstraction. This abstraction will fallback to searching directly against the raw filetree.FileTree if the index does not contain the file(s) of interest. Note: the filetree.Searcher is used by the file.Resolver abstraction.
2. Once the set of files are returned from the filetree.Searcher the results are filtered down further to return the most unique file results. For example, you may have requested for files by a glob that returns multiple results. These results are filtered down to deduplicate by real files, so if a result contains two references to the same file, say one accessed via symlink and one accessed via the real path, then the real path reference is returned and the symlink reference is filtered out. If both were accessed by symlink then the first (by lexical order) is returned. This is done automatically by the file.Resolver abstraction.
3. By the time results reach the pkg.Cataloger you are guaranteed to have a set of unique files that exist in the layer(s) of interest (relative to what the resolver supports).

Testing

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 cmd/syft/internal/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.