Bug 16051

Note: it would be nice to add this check to the Yocto check-layer test set.

Idea: add a QA check outside of checklayer then add it to CHECKLAYER_REQUIRED_TESTS
This way users can still benefit from the QA check without running checklayer.

Idea: Use "layer-specific" QA checks to enforce it on oe-core (and most likely meta-openembedded) but not by default.

The trivial implementation of this enhancement seem to have triggered underlying problems: Corentin has posted a RFC series about this:
[RFC PATCH 0/4] Clean AUTOREV usage, add revision check for every SCM URI
https://lore.kernel.org/openembedded-core/20260304104342.869457-1-corentin.guillevic@smile.fr/T/#u

(Unassigning myself since a do not realistically have the time to work on this)

Hi, I would like to work on this issue as part of my Yocto contribution work. I have been reviewing the discussion around the missing SRCREV QA check, AUTOREV handling, and the related RFC patch series from Corentin Guillevic. I am currently reproducing the issue locally and analyzing the interactions between SRCREV expansion, SCM URIs, and fetcher behavior. If there are no objections, I would like to work on proposing a fix for this bug.

Hi Sai,

Thank you for working on this :)

I'm assigning the ticket to you to make it official.

Hi all,

After reading through all the mailing list feedback carefully and 
spending time studying the bitbake fetcher code, I think I now have 
a much clearer picture of what went wrong with my approach. I wanted 
to summarize my understanding before attempting v6.

The core problem with v5:

The bigger issue is that intercepting at the vardepvalue level in 
base.bbclass was the wrong approach entirely. when I suppress the 
hash computation for recipes with missing SRCREV I am also 
accidentally breaking something important bitbake uses that hash 
to know when to re-run do_fetch. So if a branch or tag changes 
upstream do_fetch would silently not re-run which is worse than 
the original problem.

Other issues I missed:

- Using d.getVar(candidate, False) misses MACHINE/DISTRO overrides. 
  Paul is right that override resolution is the bitbake parser's job, 
  not something OE-core should be reimplementing.

- Hardcoding git/gitsm/hg/svn was fragile. bitbake already has 
  supports_srcrev() on each fetcher class for exactly this purpose.

- The parse-time warning was wrong. src-uri-bad is purely a QA-time 
  check I should not have added parse-time warnings.

- The problem is broader than just missing SRCREV as a SRCREV set to 
  a branch or tag name instead of a commit hash causes the same 
  reproducibility issues.

- The tag= skip was wrong since the selftest recipe should use 
  INSANE_SKIP instead of skipping tag= globally.

- The CHECKLAYER_REQUIRED_TESTS decision needs proper discussion 
  not a unilateral call in the cover letter.

Where I think the fix should go:

Richard suggested fixing this in the bitbake fetchers themselves 
using supports_srcrev(). I am not sure if that means changes to the 
BitBake repository itself, or just using supports_srcrev() within 
OE-core. Could someone clarify the right direction?

My current thinking for OE-core is a simple QA check in 
insane.bbclass using oe.qa.handle_error() with supports_srcrev() 
to identify which URIs need checking, along with fixing the 
gitunpackoffline selftest with INSANE_SKIP. But I want to make sure 
this aligns with what the maintainers have in mind before writing 
more code.

Thanks again for the thorough review it was genuinely helpful.

Sai Sneha

Sai, Please also post your last comment/sumary to the email thread 
if you haven't do already.

Ok Sure, Randy

Following up on suggestions to fix this in the BitBake 
fetchers using supports_srcrev() and I have tried ipmlementing that 
approach locally. here is what I did and how it works.

the fix has two parts:

part 1 — BitBake fix (prevent parse crash):
The crash happens in setup_revisions() when srcrev_internal_helper() 
raises a FetchError for missing SRCREV. This propagates up through 
expandWithRefs() as an ExpansionError which halts parsing for all 
recipes. The fix catches FetchError in setup_revisions() and sets 
a missing_srcrev flag on the FetchData object:

  def setup_revisions(self, d):
      try:
          self.revision = srcrev_internal_helper(self, d, self.name)
      except FetchError:
          self.revision = None
          self.missing_srcrev = True

git.py and hg.py also needed a guard to prevent calling 
latest_revision() when missing_srcrev is set, since that would 
trigger a live network fetch under BB_NO_NETWORK=1.

part 2 — OE-core QA check (insane.bbclass):
a simple test_missing_srcrev() in do_recipe_qa uses supports_srcrev() 
to identify which URIs need checking:

  def test_missing_srcrev(pn, d):
      import bb.fetch2
      try:
          fetcher = bb.fetch2.Fetch(
              (d.getVar('SRC_URI') or '').split(), d)
          for url in fetcher.ud:
              ud = fetcher.ud[url]
              if ud.method.supports_srcrev() and \
                 getattr(ud, 'missing_srcrev', False):
                  oe.qa.handle_error("missing-srcrev",
                      "%s: SRCREV not set for SCM URI %s" % (pn, url), d)
      except bb.fetch2.FetchError:
          pass

Test results:

  $ bitbake -p 2>&1 | grep -E "(WARNING|ERROR|SRCREV)"
  WARNING: You are running bitbake under WSLv2...
  Summary: There was 1 WARNING message.
  (no parse crash, all recipes parse cleanly)

  $ bitbake -c recipe_qa -f test-fetcher 2>&1 | grep -E "(WARNING|ERROR|missing-srcrev)"
  WARNING: test-fetcher-1.0-r0 do_recipe_qa: QA Issue: test-fetcher: 
  SRCREV not set for SCM URI git://github.com/example/project.git [missing-srcrev]

  $ bitbake -c recipe_qa test-srcrev 2>&1 | grep -E "(WARNING|ERROR|missing-srcrev)"
  ERROR: test-srcrev-1.0-r0 do_recipe_qa: QA Issue: test-srcrev: 
  SRCREV not set for SCM URI git://github.com/example/project.git [missing-srcrev]

selftest results:
  test_git_unpack_nonetwork      PASSED
  test_git_unpack_nonetwork_fail PASSED
  test_git_patchtool             PASSED
  test_git_patchtool2            PASSED
  test_just_parse                PASSED
  test_warnings_errors           PASSED
  test_invalid_recipe_src_uri    PASSED

Changes are minimal:
  BitBake:  3 files, 8 insertions, 3 deletions
  OE-core:  1 file,  16 insertions, 1 deletion

Does this approach align with what was suggested? I am happy to 
refine it further based on feedback.

Link to the work in progress patch:
[PATCH v5 0/4] Add missing SRCREV check for SCM URIs - Sai Sneha
https://lore.kernel.org/all/20260602062940.3107241-1-saisneha196@gmail.com/

Hi all,

I will not be able to continue working on this bug and am 
unassigning myself so someone else can pick it up.

Thank you all for the guidance.

Sai Sneha