feat: validate now accepts components with external bom

[mmarseu]

The What of this PR is straight-forward: It makes the validate command accept components als valid which have an external reference of type bom, even if those components do not fulfil the requirements of our custom schema.
The program now outputs a warning to inform the user that the externally referenced BOM cannot be checked.

The How merits a little explanation, I guess, to help with the review. Because the schema diffs look much more confusing than they are.

The custom schemas now wrap all of our custom requirements on components in a big if-then-else construct, where the if tests whether the component has an external reference to a BOM.

• If it does not, the else clause applies all requirements as before. (see note on version below)
• If it does, the then clause applies a schema which is always expected to fail (by requiring an arbitrarily named property). I call this a marker requirement.

This marker requirement can be detected in the validation routine. When jsonschema produces an error for this particular marker, the tool only writes a warning but otherwise ignores the error.

Note on version: If the removal of the version requirements trips you up when you see it all the way at the top of the diff, don't worry. I have only moved it into the else clause, because an externally referenced BOM would be where the version is declared.

Misc changes:

• Adds a dependency on pytest-subtests, a small pytest plugin which enables support for unittest's subtest feature when running the tests through pytest. The alternative would be to either not use subtests (but I think the feature is very useful in this case to run tests against all spec versions at once) or to rewrite the tests to native pytest which has a comparable feature.

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
validator
validate.py	110	6	94%	110–111, 113, 147, 163, 198
TOTAL	2008	104	94%

Tests	Skipped	Failures	Errors	Time
367	2 💤	0 ❌	0 🔥	12.535s ⏱️

[italvi]

@mmarseu I faced two issues while testing: If e.g. version is empty, an error for the component is still thrown, because the if is after the try-catch, where you throw the warning, contradicting the statement from the warning.

Further: if the url is not a valid bom-link as described on the cdx website, respectively specification, you ignore it. So I could write some gibberish, even just an empty url and still would ignore the validation. Though this hopefully is only an issue if providing an externalReference manually.

[mmarseu]

@mmarseu I faced two issues while testing: If e.g. version is empty, an error for the component is still thrown, because the if is after the try-catch, where you throw the warning, contradicting the statement from the warning.

I don't really see the problem with this particular behavior. If you have an empty version property on a component that references an external SBOM, what information is that supposed to convey? I don't see valid use-cases for this, so the error makes sense.

But your argument brought me to a larger issue I hadn't considered. Obviously, the user shouldn't be required to include all of the mandatory fields in a component which references an external SBOM. That is the entire point of this PR.
But what if they chose to include some information nevertheless (e.g., a version as in your example)? Either we prohibit this (not my preferred solution) or we ensure that any properties which are present still conform to our requirements (e.g., not be empty).

AFAIK, this applies to:

• name (minLength 1)
• version (minLength 1)
• copyright (minLength 1) or licenses (minItems 1)

Further: if the url is not a valid bom-link as described on the cdx website, respectively specification, you ignore it. So I could write some gibberish, even just an empty url and still would ignore the validation. Though this hopefully is only an issue if providing an externalReference manually.

That is by design. Bom-linking isn't the only way to reference an external SBOM. As far as I am concerned, a URL is also fine. The stock CDX spec already ensures that the URL is either a valid IRI-reference or a bom-link. Unfortunately, it is actually not easy to find a string that isn't a valid IRI-reference, so it's hard to verify that this works. Try :foo as a URL. You'll see that this throws an error.
The empty string is a special case and I agree that it should be invalid. I'll make the change.

[italvi]

lgtm, just rebase it please, then I will approve it 😁

[italvi]

now lgtm