Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

src/fromager/bootstrapper.py (1)

383-392: ⚠️ Potential issue | 🟠 Major

Fix parent_req to reference the actual parent, not the current package.

At line 386, self.why[-1][1] retrieves the current package being bootstrapped (which was pushed onto the stack by _track_why), not its parent. This causes two problems:

• For top-level packages: self.why contains only the current package, so parent_req is incorrectly set to the package itself rather than None.
• For transitive dependencies: parent_req points to the current package instead of the actual parent.

The correct fix uses self.why[-2] to access the parent:

Suggested fix

             if force_prebuilt:
                 # Resolve prebuilt wheel URL (source_url is the sdist URL)
-                parent_req = self.why[-1][1] if self.why else None
+                parent_req = self.why[-2][1] if len(self.why) > 1 else None
                 wheel_url, fallback_version = self._resolver.resolve(

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/fromager/bootstrapper.py` around lines 383 - 392, The code sets
parent_req using self.why[-1][1], which refers to the current package pushed by
_track_why rather than its parent; update the logic inside the
pbi.pre_built/force_prebuilt branch so parent_req is None when self.why has only
one entry and otherwise uses the previous frame's request (i.e.
self.why[-2][1]); ensure this change is applied where parent_req is passed into
self._resolver.resolve(req=..., req_type=..., parent_req=parent_req,
pre_built=True).

src/fromager/bootstrapper.py (1)

494-500: Consider adding a guard to verify failure was recorded before suppressing.

This exception handling assumes that any exception raised by bootstrap() has already been recorded in failed_packages. However, if an exception occurs early in bootstrap() (e.g., in _add_to_graph() at line 272, which is before the main try block), it would be swallowed without being recorded.

The same concern applies to the identical pattern at lines 686-692.

🛡️ Optional guard pattern

         for dep in self._sort_requirements(install_dependencies):
             with req_ctxvar_context(dep):
+                failures_before = len(self.failed_packages)
                 try:
                     self.bootstrap(req=dep, req_type=RequirementType.INSTALL)
                 except Exception:
-                    if not self.test_mode:
+                    if not self.test_mode or len(self.failed_packages) == failures_before:
                         raise
                     # Failure already recorded in failed_packages; continue.
             self.progressbar.update()

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/fromager/bootstrapper.py` around lines 494 - 500, The exception handler
in bootstrapper.bootstrap that currently swallows all exceptions when
self.test_mode is true should first verify the failure was recorded in
self.failed_packages (or whatever structure records failures) before
suppressing; change the except Exception: blocks (the one wrapping
self.bootstrap(req=dep, req_type=RequirementType.INSTALL) and the identical
block at the later lines) to except Exception as e: and if self.test_mode and
dep in self.failed_packages: continue else: raise (this ensures exceptions
thrown early in bootstrap (e.g., in _add_to_graph) are not silently swallowed).

src/fromager/commands/bootstrap.py (1)

185-192: Pattern aligns with _bootstrap_impl() dependency loops.

The exception-handling logic here mirrors the approach used in _bootstrap_impl() for install and build dependencies. The assumption that failures are "already recorded in bt.failed_packages" holds for resolution and build errors caught within bootstrap().

However, note that a past review flagged a subtle edge case: if an exception occurs in _add_to_graph() (before the main try block in bootstrap()), it would be swallowed here without being recorded. Consider adding a guard to verify the failure was actually recorded before suppressing, as suggested in prior reviews.

🛡️ Optional guard to verify failure was recorded

         for req in resolved_reqs:
             token = requirement_ctxvar.set(req)
+            failures_before = len(bt.failed_packages)
             try:
                 bt.bootstrap(req, requirements_file.RequirementType.TOP_LEVEL)
             except Exception:
-                if not test_mode:
+                if not test_mode or len(bt.failed_packages) == failures_before:
                     raise
                 # In test mode, failure is already recorded in
                 # bt.failed_packages; continue to next requirement.
             progressbar.update()
             requirement_ctxvar.reset(token)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/fromager/commands/bootstrap.py` around lines 185 - 192, The current
except block after calling bt.bootstrap(req, ...) can swallow exceptions that
weren't recorded (e.g., those raised in _add_to_graph()); update the handler to
capture the exception (except Exception as e:) and, when test_mode is True,
verify the failure was actually recorded in bt.failed_packages for the current
requirement (use the same key/identifier bt uses for failures) before
suppressing; if the requirement is not present in bt.failed_packages, re-raise
the exception (or propagate e) so unexpected errors aren’t silently dropped.
Ensure this logic is applied around the bt.bootstrap(req, ...) call and
references bt.failed_packages and the current req identifier.