Repository navigation
Model target producer bindings and instrument standings - #9323
Closed
gunbai-bot[bot] wants to merge 0 commit into
Closed
gunbai-bot[bot] wants to merge 0 commit into
gunbai-bot[bot] wants to merge 0 commit into
Conversation
gunbai-bot
Bot
changed the title
gunbai-bot Bot
pushed a commit
that referenced
this pull request
Aug 26, 2026
… written PanickedBeforeVerdict and NotAttemptedAfterAbort landed on the floor's disposition vocabulary after #9323 wrote gunbc.build_target's two wildcard-free policy folds, so the merge of main into this branch is where they meet and refuse to compile. That is the fold behaving exactly as its own comment says it must: a new arm fails to compile rather than being inherited from whichever arm a wildcard named. Both are UNSATISFIED for the aggregate. NotAttemptedAfterAbort is the one that would have been damaging to absorb -- a required set treating 'never attempted' as satisfied reports itself green over a run that stopped early, the empty-observation narrow arriving as a green. The Blaze export gives NotAttemptedAfterAbort its exact upstream member, HALTED_BEFORE_TESTING, rather than the tempting INCOMPLETE that would have collapsed it into a pile upstream already distinguishes. PanickedBeforeVerdict has no member and joins FAILED beside RuntimeErroredBeforeVerdict.
gunbai-bot Bot
pushed a commit
that referenced
this pull request
Aug 26, 2026
…er INCOMPLETE #9343 landed on main fixing the same two ClaimDisposition arms this branch fixed independently, so the two met in build_target.dag. Both halves resolved deliberately rather than by taking a side whole. THE IMPORT: main's build_target.dag has no gunbc.target_binding import because #9323 has not merged, so main's TestStanding carries three arms and this branch's carries seven. Taking main's side would have deleted the four producer arms and every consumer of them. HEAD's side survives. THE EXPORT, AND THIS IS A DELIBERATE DIVERGENCE FROM #9343 RATHER THAN A MERGE ARTEFACT. The aggregate policy is byte-identical between us -- both arms unsatisfied, same causes. The Blaze export differs on one line: #9343 renders NotAttemptedAfterAbort as INCOMPLETE, this keeps HALTED_BEFORE_TESTING. The reason is upstream. NotAttemptedAfterAbort is constructed from NotAttempted { halted_by }, and Bazel's HALTED_BEFORE_TESTING means exactly 'the test did not run because the build was halted'. It is an exact member for the fact, so the export loses nothing; INCOMPLETE collapses it into the pile that module's own comment already describes as lossy, beside HostToolUnresolvedBeforeVerdict and RouteGapBeforeVerdict. Nothing reads the export back into a decision, so this is a fidelity choice and not a correctness one -- a one-line revert if the reviewer prefers main's reading.
briansrls
pushed a commit
that referenced
this pull request
Aug 27, 2026
* Model target producer bindings * Type instrument refusal causes * Reject noncanonical target index labels * Refuse noncanonical target lookup labels * Home label canonicality in label authority * Name canonicality for label authority * Keep label annotation at module grain * Seal target binding index construction * wip: target dispatch seam * fix: single-arm coproducts need a record body * Delete the claim_executor route atomically, and name the unpinned-tree limit The old --heads-reading-differential mode is removed in the same change that makes the instrument addressable, so there is never an interval where two routes to one producer both work. Its parse-wall figures survive as an explicitly unmodeled host line: a replacement that silently drops a capability is not a replacement. The producer executes against an unpinned live working tree and the carriers now say so. InvocationOutcome stays an OUTCOME rather than a receipt, and both the model and the host record that no digest field can close the gap -- the tree can go A -> B -> A while the producer reads a mixed population and both hashes agree. * Classify the two ClaimDisposition arms main added after this fold was written PanickedBeforeVerdict and NotAttemptedAfterAbort landed on the floor's disposition vocabulary after #9323 wrote gunbc.build_target's two wildcard-free policy folds, so the merge of main into this branch is where they meet and refuse to compile. That is the fold behaving exactly as its own comment says it must: a new arm fails to compile rather than being inherited from whichever arm a wildcard named. Both are UNSATISFIED for the aggregate. NotAttemptedAfterAbort is the one that would have been damaging to absorb -- a required set treating 'never attempted' as satisfied reports itself green over a run that stopped early, the empty-observation narrow arriving as a green. The Blaze export gives NotAttemptedAfterAbort its exact upstream member, HALTED_BEFORE_TESTING, rather than the tempting INCOMPLETE that would have collapsed it into a pile upstream already distinguishes. PanickedBeforeVerdict has no member and joins FAILED beside RuntimeErroredBeforeVerdict. * Register the host seam as seed-retained and install the regenerated CLI mirror Two regen obligations the CLI model change created. target_invocation_host.rs is a committed src/v1/stage0/src/*.rs the emitter does not produce, so regen refused it as a mirror that stopped being emitted. It is declared as a SeedRetainedIntrinsicRegistration with has_pub_mod: false, matching required_regen_host and declaration_index -- it is wired by #[path] mod inside cli_run.rs and contributes no pub mod line to the emitted lib.rs. The registration moves three generated artifacts alongside it, as every prior registration has. gunbc_cli_dispatch_surface.rs is the emitted mirror of the module that gained the operand axis, installed from the regen candidate rather than hand-written: the two new one-arm enums, CliOperandRow, the operands field on every subcommand, the CliInvokesBoundTargetProducer arm and the test verb row. * cargo fmt * Write the HALTED_BEFORE_TESTING argument onto the arm, with its receipt The export arm for NotAttemptedAfterAbort has now been decided twice, the other way: gunbc#9343 chose INCOMPLETE on main while this branch chose HALTED_BEFORE_TESTING, and the two met in a merge. Without the argument written down, the next reader diffing against that history finds an unexplained divergence and 'fixes' it back. The receipt is checked rather than recalled: extdeps.bazel.test_status carries TestStatusHaltedBeforeTesting as wire number 8, BLAZE_HALTED_BEFORE_TESTING, and the disposition is built from NotAttempted { halted_by }. Same fact, exact member. Module-scope comment, above the declaration -- the annotation grain rule refuses one inside the match body. * The main merge landed #9343's arms BESIDE mine, not onto them Both ClaimDisposition folds in gunbc.build_target carried two PanickedBeforeVerdict arms and two NotAttemptedAfterAbort arms after the merge, and the export's second NotAttemptedAfterAbort carried #9343's INCOMPLETE. Whichever way the compiler resolves a duplicate, one of the two readings was wrong: first-match-wins reds the witness, last-wins makes my change dead code and silently keeps the other value. HOW I MISSED IT, because the check I ran looked sufficient and was not. After the merge I grepped the diff against main and saw my HALTED_BEFORE_TESTING arm present, and concluded the resolution was right. Presence is not exclusivity. A conflict git reports gets resolved carefully; a hunk that applies CLEANLY beside an existing one does not announce itself, and that is the direction this class always arrives from. Found by cool-swift-307 reading my head, not by me and not by CI -- no checks had run. Both folds are now one arm per member: 13 arms against ClaimDisposition's 13 members, verified by counting rather than by eye, and every other match in the touched modules swept for the same shape. bazel_target_model_witness_test moved with the code rather than being left asserting the value the code no longer produces. Its run/never-ran argument is UNCHANGED and survives intact -- INCOMPLETE is in the never-ran family, so the split it tests was never wrong, only vaguer than upstream allows. What the arm records now is which never-ran member is exact, with the wire number and both PRs named. * Retain label canonicality refusal causes * Carry the typed canonicality cause through the invocation route bright-moth-189 deleted `label_is_canonical` in favour of the typed `label_canonicality` / `LabelCanonicalityCause` pair, so the registry now retains WHICH way a label was non-canonical. This completes that chain through the consumer layer: `InvocationLabelNotCanonical` carries the cause and `invocation_refusal_rendered` surfaces it. Dropping the cause here would have made the split real and invisible at the only surface a caller of `gunbc test` touches -- a typed distinction that reaches no reader is the same false bit it replaced, one layer out. The inner `LabelRefusal` is carried but deliberately NOT decomposed here: that vocabulary belongs to the label authority and has no renderer there today, so writing one in this module would be a second site deciding what `DotSegment` says to a human (§3 fork through a rendering rather than a type). This names the CLASS of repair, which is what distinguishes the arms. Two witnesses cover it, using a structurally non-canonical label built through the sole constructor: one that the cause survives into the route's refusal, one that it reaches the rendered text. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: gunbc-ci-auto-heal <gunbc-ci-auto-heal@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
gunbai-bot
Bot
force-pushed
the
session/bright-moth-189
branch
from
August 27, 2026 00:57
f037b60 to
39688a5
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
TargetBindingas a peer relation toBuildTarget, with a unique label-keyed index and typed bound/missing/unknown lookup outcomesTestStandingwith producer-specific behavioral-receipt and heads-reading differential observations without adding argv or execution details to the target model//:requiredand Blaze-export boundaries instead of collapsing their observations into generic pass/fail statusesTest plan
claim_batchexecution overtarget_binding_witness_test.dag— 7/7 PASS, including the canonical-known-target vs forged-binding-target rendered-key collisionrendered_key_collision_currently_misbinds— PASS, confirming the silent alias before the repairgit diff --check— PASScargo fmt --all --check— PASSDashboard node:
adhoc-df227240-974