limit source-distribution refs to deb and rpm

- apk origins and alpm pkgbases are build recipes, not published source packages, and their names are usually also real binary packages, so a `source-distribution` ref there would point at a binary package. They keep relying on the `upstream` PURL qualifier.
- document why this is an external reference rather than `pedigree.ancestors`, with links to the CycloneDX spec discussion (CycloneDX/specification#612)

Signed-off-by: Alex Goodman <wagoodman@users.noreply.github.com>

syft/format/cyclonedxjson/testdata/snapshot/TestCycloneDxSourcePackageEncoder.golden

@@ -156,12 +156,6 @@
       "name": "gcc-libs",
       "version": "13.2.1-3",
       "purl": "pkg:alpm/arch/gcc-libs@13.2.1-3?arch=x86_64&distro=arch-rolling&upstream=gcc",
-      "externalReferences": [
-        {
-          "url": "pkg:alpm/arch/gcc@13.2.1-3?distro=arch-rolling",
-          "type": "source-distribution"
-        }
-      ],
       "properties": [
         {
           "name": "syft:package:foundBy",
@@ -207,12 +201,6 @@
       "name": "libc-utils",
       "version": "0.7.2-r3",
       "purl": "pkg:apk/alpine/libc-utils@0.7.2-r3?arch=x86_64&distro=alpine-3.16.3&upstream=libc-dev",
-      "externalReferences": [
-        {
-          "url": "pkg:apk/alpine/libc-dev@0.7.2-r3?distro=alpine-3.16.3",
-          "type": "source-distribution"
-        }
-      ],
       "properties": [
         {
           "name": "syft:package:foundBy",

syft/format/cyclonedxxml/testdata/snapshot/TestCycloneDxSourcePackageEncoder.golden

@@ -73,11 +73,6 @@
       <name>gcc-libs</name>
       <version>13.2.1-3</version>
       <purl>pkg:alpm/arch/gcc-libs@13.2.1-3?arch=x86_64&amp;distro=arch-rolling&amp;upstream=gcc</purl>
-      <externalReferences>
-        <reference type="source-distribution">
-          <url>pkg:alpm/arch/gcc@13.2.1-3?distro=arch-rolling</url>
-        </reference>
-      </externalReferences>
       <properties>
         <property name="syft:package:foundBy">alpm-db-cataloger</property>
         <property name="syft:package:type">alpm</property>
@@ -94,11 +89,6 @@
       <name>libc-utils</name>
       <version>0.7.2-r3</version>
       <purl>pkg:apk/alpine/libc-utils@0.7.2-r3?arch=x86_64&amp;distro=alpine-3.16.3&amp;upstream=libc-dev</purl>
-      <externalReferences>
-        <reference type="source-distribution">
-          <url>pkg:apk/alpine/libc-dev@0.7.2-r3?distro=alpine-3.16.3</url>
-        </reference>
-      </externalReferences>
       <properties>
         <property name="syft:package:foundBy">apk-db-cataloger</property>
         <property name="syft:package:type">apk</property>

syft/format/internal/cyclonedxutil/helpers/external_references.go

@@ -89,9 +89,26 @@ func encodeExternalReferences(p pkg.Package) *[]cyclonedx.ExternalReference {
 	return nil
 }
 
-// encodeSourcePackageExternalReference returns a "source-distribution" reference holding the PURL of the source
-// package that an OS package was built from. It is only emitted when the source differs from the binary package
-// (by name or version) and when the package PURL is of the type implied by the metadata.
+// encodeSourcePackageExternalReference returns a "source-distribution" reference holding the PURL of the distro
+// source package that a binary OS package was built from (e.g. libpam-runtime -> pkg:deb/debian/pam@...?arch=source).
+// It is only emitted when the source differs from the binary package (by name or version) and when the package PURL
+// is of the type implied by the metadata. The main consumer is vulnerability matching, since distros tend to report
+// vulnerabilities against the source package rather than each binary package built from it.
+//
+// why an external reference and not pedigree: `component.pedigree.ancestors` looks like the obvious spot for "this
+// binary came from that source", but pedigree describes code lineage (forks, patched variants, and the patches that
+// separate them). A binary built unmodified from its own distro source package is the same code in a different form,
+// not a fork, and CycloneDX components are deliberately agnostic to source vs binary form (see the comments from the
+// CycloneDX maintainers in https://github.com/CycloneDX/specification/issues/612#issuecomment-2958800363). The
+// ancestor slot is better left for the upstream project a distro patched (e.g. Linux-PAM for Debian's pam), which
+// syft does not know today.
+//
+// why "source-distribution": the spec defines it as "the location where the source code distributable can be
+// obtained", and a deb/rpm source package is exactly that distributable (a .dsc and tarballs in the Debian source
+// pool, a .src.rpm in an SRPMS repo). The value is a PURL (an identifier) rather than a download URL, which the
+// schema allows since external reference URLs are URIs of any scheme, and a CycloneDX maintainer confirmed this exact
+// shape is valid in https://github.com/CycloneDX/specification/issues/612#issuecomment-5428057874. A first-class
+// "source" component type was discussed in that issue and deferred to CycloneDX 2.0.
 func encodeSourcePackageExternalReference(p pkg.Package) *cyclonedx.ExternalReference {
 	if p.PURL == "" {
 		return nil
@@ -138,7 +155,7 @@ type sourcePackage struct {
 	purlType string
 	name     string
 	version  string
-	// arch is the ecosystem's source architecture marker, empty when the ecosystem has no such convention
+	// arch is the ecosystem's source architecture marker ("source" for deb, "src" or "nosrc" for rpm)
 	arch string
 }
 
@@ -153,14 +170,17 @@ func (s sourcePackage) keepQualifier(key string) bool {
 }
 
 // sourcePackageOf returns the source package for p, or nil when there is none or it is the same as the binary.
+//
+// only deb and rpm are supported, since those are the ecosystems that publish real source packages with their own
+// PURL identity (arch=source / arch=src). apk (origin) and alpm (pkgbase) are intentionally left out: there the
+// "source" is a build recipe rather than a published source distributable, and its name is usually also the name of
+// a real binary package (e.g. pkg:apk/alpine/libc-dev@0.7.2-r3 is an installable package, not a source archive), so
+// a "source-distribution" reference would point at a binary package. Whether those ecosystems have a meaningful
+// source PURL is still an open question, and until then the `upstream` PURL qualifier continues to carry that info.
 func sourcePackageOf(p pkg.Package) *sourcePackage {
 	var src sourcePackage
 	var binName, binVersion string
 	switch m := p.Metadata.(type) {
-	case pkg.ApkDBEntry:
-		// apk has no source arch convention, so the source PURL carries no arch
-		src = sourcePackage{purlType: pkg.ApkPkg.PackageURLType(), name: m.OriginPackage, version: m.Version}
-		binName, binVersion = m.Package, m.Version
 	case pkg.DpkgDBEntry:
 		src = dpkgSourcePackage(m)
 		binName, binVersion = m.Package, m.Version
@@ -173,10 +193,6 @@ func sourcePackageOf(p pkg.Package) *sourcePackage {
 	case pkg.RpmArchive:
 		src = rpmSourcePackage(m.SourceRpm)
 		binName, binVersion = m.Name, m.Version+"-"+m.Release
-	case pkg.AlpmDBEntry:
-		// alpm has no source arch convention, so the source PURL carries no arch
-		src = sourcePackage{purlType: pkg.AlpmPkg.PackageURLType(), name: m.BasePackage, version: m.Version}
-		binName, binVersion = m.Package, m.Version
 	default:
 		return nil
 	}

syft/format/internal/cyclonedxutil/helpers/external_references_test.go

@@ -159,20 +159,20 @@ func Test_encodeSourcePackage(t *testing.T) {
 		{
 			name: "no PURL",
 			input: pkg.Package{
-				Name:    "some-package",
-				Version: "1.0.0-r1",
-				Type:    pkg.ApkPkg,
-				Metadata: pkg.ApkDBEntry{
-					Package:       "some-package",
-					OriginPackage: "some-origin",
-					Version:       "1.0.0-r1",
-					Architecture:  "x86_64",
+				Name:    "libpam-runtime",
+				Version: "1.5.2-6+deb12u1",
+				Type:    pkg.DebPkg,
+				Metadata: pkg.DpkgDBEntry{
+					Package: "libpam-runtime",
+					Version: "1.5.2-6+deb12u1",
+					Source:  "pam",
 				},
 			},
 			expected: nil,
 		},
 		{
-			name: "from apk",
+			// apk origins are build recipes, not published source packages, so they are not supported
+			name: "from apk is not supported",
 			input: pkg.Package{
 				Name:    "libc-utils",
 				Version: "0.7.2-r3",
@@ -185,24 +185,6 @@ func Test_encodeSourcePackage(t *testing.T) {
 					Architecture:  "x86_64",
 				},
 			},
-			expected: &cyclonedx.ExternalReference{
-				Type: cyclonedx.ERTypeSourceDistribution, URL: "pkg:apk/alpine/libc-dev@0.7.2-r3?distro=alpine-3.16.3",
-			},
-		},
-		{
-			name: "from apk with source = bin",
-			input: pkg.Package{
-				Name:    "some-package",
-				Version: "1.0.0-r1",
-				Type:    pkg.ApkPkg,
-				PURL:    "pkg:apk/alpine/some-package@1.0.0-r1?arch=x86_64&distro=alpine-3.16.3",
-				Metadata: pkg.ApkDBEntry{
-					Package:       "some-package",
-					OriginPackage: "some-package",
-					Version:       "1.0.0-r1",
-					Architecture:  "x86_64",
-				},
-			},
 			expected: nil,
 		},
 		{
@@ -334,7 +316,8 @@ func Test_encodeSourcePackage(t *testing.T) {
 			},
 		},
 		{
-			name: "from alpm",
+			// alpm pkgbases are build recipes, not published source packages, so they are not supported
+			name: "from alpm is not supported",
 			input: pkg.Package{
 				Name:    "gcc-libs",
 				Version: "13.2.1-3",
@@ -347,24 +330,6 @@ func Test_encodeSourcePackage(t *testing.T) {
 					Architecture: "x86_64",
 				},
 			},
-			expected: &cyclonedx.ExternalReference{
-				Type: cyclonedx.ERTypeSourceDistribution, URL: "pkg:alpm/arch/gcc@13.2.1-3?distro=arch-rolling",
-			},
-		},
-		{
-			name: "from alpm with source = bin",
-			input: pkg.Package{
-				Name:    "some-package",
-				Version: "1.0.0",
-				Type:    pkg.AlpmPkg,
-				PURL:    "pkg:alpm/arch/some-package@1.0.0?arch=x86_64&distro=arch-rolling&upstream=some-package",
-				Metadata: pkg.AlpmDBEntry{
-					Package:      "some-package",
-					Version:      "1.0.0",
-					BasePackage:  "some-package",
-					Architecture: "x86_64",
-				},
-			},
 			expected: nil,
 		},
 		{

syft/format/internal/testutil/source_package_input.go

@@ -43,7 +43,7 @@ func newSourcePackageCatalog() *pkg.Collection {
 
 	catalog := pkg.NewCollection()
 
-	// apk: the origin differs from the package name, and shares its version
+	// apk: the origin differs from the package name, but apk is not supported so no source reference is expected
 	catalog.Add(pkg.Package{
 		Name:      "libc-utils",
 		Version:   "0.7.2-r3",
@@ -59,7 +59,7 @@ func newSourcePackageCatalog() *pkg.Collection {
 		},
 	})
 
-	// apk: negative control -- the package is its own origin, so no source reference is expected
+	// apk: negative control -- the package is its own origin
 	catalog.Add(pkg.Package{
 		Name:      "busybox",
 		Version:   "1.35.0-r17",
@@ -126,7 +126,7 @@ func newSourcePackageCatalog() *pkg.Collection {
 		},
 	})
 
-	// alpm: the base package differs from the package name, and shares its version
+	// alpm: the base package differs from the package name, but alpm is not supported so no source reference is expected
 	catalog.Add(pkg.Package{
 		Name:      "gcc-libs",
 		Version:   "13.2.1-3",