Skip to content

merge: link the BOM subject, stamp Tools metadata, clean up duplicated metadata/empty lists - #513

Open
jimklimov wants to merge 8 commits into
CycloneDX:mainfrom
jimklimov:merge-metadata-and-cleanup
Open

jimklimov wants to merge 8 commits into
CycloneDX:mainfrom
jimklimov:merge-metadata-and-cleanup

Conversation

@jimklimov

Copy link
Copy Markdown
Contributor

What

  • Passes the BOM subject into FlatMerge so the merged document's dependency graph stays linked to it.
  • Stamps Metadata.Tools on the merged document via the library's BomMetadataUpdate().
  • Cleans up a duplicated metadata component and empty lists left behind by the merge, via the
    library's CleanupMetadataComponent()/CleanupEmptyLists().

Why

Supersedes both #340 (bom-self-metadata) and #334 (flatmerge-dedup-squash) — in the current
codebase these turned out to be one coherent change to MergeCommand.cs rather than two separate
concerns, so they're combined here instead of kept as two like-for-like PRs. Closing both in favor of
this PR.

Depends on cyclonedx-dotnet-library#446
(BomMetadataUpdate) and #448
(CleanupMetadataComponent/CleanupEmptyLists). Stacked on #512 (merge-input-files-list) — must
merge first.

MergeCommand now calls CycloneDXUtils.FlatMerge/HierarchicalMerge with
MergeStrategy.Default() when built against a library that has it
(#if NET8_0_OR_GREATER, matching the library's own guard -- CLI is
net10.0-only today but this keeps the two projects' conditional
compilation symmetric and self-documenting), falling back to the
plain overload otherwise. No new CLI flags: strategy toggles are not
yet exposed at the command-line layer, so this only changes default
merge behavior, not the command's surface.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Adds a "rename-entity" command that renames a bom-ref and every
back-reference to it throughout a BOM document, built on the
library's new BomRefWalker-based Bom.RenameRef(old, new) API. Follows
current System.CommandLine conventions
(System.CommandLine.NamingConventionBinder, not the older
System.CommandLine.Invocation namespace) and matches Convert/
MergeCommand's internal (not public) visibility.

Verified end-to-end against a real fixture: renaming a component's
bom-ref correctly rewrites both the dependsOn back-reference and the
dependency's own ref entry, and stamps fresh SerialNumber/Timestamp/
Tools metadata via BomMetadataUpdate/BomMetadataReferThisToolkit.

Wired into Program.cs (alphabetical position, between Merge and Sign),
guarded by #if NET8_0_OR_GREATER to match the library capability it
depends on.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Follows MergeTests.cs conventions: direct handler call, TempDirectory,
Snapshooter with serialNumber/timestamp stripped (plus the tools list,
whose contents are build/environment-specific -- assembly versions and
"testhost" vs. the real CLI name under `dotnet test`). Covers a JSON
and an XML output round-trip of the rewrite-identifier-and-back-refs
case, plus a no-op case when the requested old-ref isn't present.

Full suite: 132 passed / 0 failed (129 pre-existing + 3 new).

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
…usal

merge gains --component-conflict-resolution
<KeepSeparate|Squash_UpgradeScope|Squash_DowngradeScope|Squash_RenameByScope>,
defaulting to the library's MergeStrategy.Default() (Squash_UpgradeScope)
when not specified. This closes a pre-existing gap: the library-side
strategy was never selectable from the CLI at all -- it was always
hardcoded to Default() internally. Verified end-to-end: merging two
BOMs where the same component is "required" in one and "excluded" in
the other with --component-conflict-resolution Squash_RenameByScope
produces two distinct components (lp:scope=Required /
lp:scope=Excluded) with each source's dependsOn correctly pointing at
its own variant.

rename-entity now catches the InvalidOperationException
Bom.RenameRef throws when the requested new-ref collides with an
existing identifier, reporting it as a clean parameter-validation
error instead of an unhandled crash.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
This lets callers pass files listing BOM filenames (one per line,
or 0x00-separated for the --input-files-nul-list variant, e.g. from
`find -print0`) instead of one --input-files argument per BOM, to
avoid OS/shell command-line length or argument-count limits when
merging many BOMs, a real constraint once the number of input files
grows large.

Entries are deduplicated against --input-files and each other as
they are collected. Non-absolute paths are resolved relative to
the current working directory. No library changes needed -- this
feature only builds a longer input-file list before the existing
InputBoms(...) call.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Previously, when a flat merge specified a subject component (--group/
--name/--version), MergeCommand set Metadata.Component directly and
never called the FlatMerge overload that accepts a subject component.
As a result the merged document's dependency graph never linked the
subject to the components each input BOM contributed -- no
"<dependency ref="subject"><dependsOn>...` entry ever appeared,
unlike a hierarchical merge with the same options.

Calling FlatMerge(inputBoms, bomSubject, ...) instead lets the library
build that dependency entry itself (namespacing the subject's bom-ref
and linking it to each input BOM's own metadata component), matching
what hierarchical merge already does. The "pick the first input BOM's
component as a default" fallback still runs, but only when no subject
was requested at all.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Replaces manually setting Version/SerialNumber/Timestamp inline with
Bom.BomMetadataUpdate(true) and Bom.BomMetadataReferThisToolkit().
The manual version never recorded which library/tool produced the
merged document at all -- Metadata.Tools was left exactly as whichever
input BOM happened to contribute one, or absent entirely. The merged
document now carries an entry for this library and the running
program, the same way validate/convert/sign already leave a traceable
record of what processed a document.

Updates MergeTests' snapshot-cleanup regex to also strip the tools
block before comparison, since its contents (assembly name/version)
are build-environment-specific -- same treatment already used for
other commands' tests.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Calls the library's new CleanupMetadataComponent/CleanupEmptyLists on
the merged document. Complements the library-side change (which
already covers this for a merge that requests an explicit BOM
subject): when MergeCommand falls back to picking the first input
BOM's own metadata component as the subject, that same component can
also be present in the merged Components list, producing two entries
with the same bom-ref -- a specification violation the library-level
cleanup couldn't see, since the CLI's fallback selection happens after
the library call returns. CleanupEmptyLists also drops now-empty
top-level lists (e.g. an empty "vulnerabilities": []) that a flat
merge of BOMs with no vulnerabilities previously left in the output.

Signed-off-by: Jim Klimov <jimklimov@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant