Repository navigation
Fix 10 failing tests (structure_mapper confidence, hardcoded path, e2e slot assertions, cpp StateStore call) - #40
Open
shentonyan wants to merge 3 commits into
Conversation
…ed seed path in test_publish - sglang_trace_analyze/structure_mapper._map_one: when a call stack is present and the layer is resolved, 'confidence' was never assigned, raising UnboundLocalError. Default to 'high' (matches test_build_mapping_with_call_stack). - harness_evolve/tests/test_publish.py: resolve the dcu_kernel_auto_opt harness_default seed relative to the test file instead of a machine-specific absolute path, so the tests run on any checkout. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
scoreboard.list_claims() is documented as a snapshot of every GPU, free or held, so a released slot still appears as a row with status='free'. The four e2e tests asserted that no row exists for the slot (or that the list is empty) and therefore failed after a successful release. Filter on status == 'held'. test_remote_job_flow's during-run check had the same flaw and always passed; it now also requires status == 'held'. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
StateStore.init_or_resume(task_id) no longer takes task_type (run.json does not store it; requirements.json is the source of truth). Two cpp framework tests still passed it and raised TypeError. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Fixes 10 tests that fail on a fresh checkout of
master(b3f6505). Three commits, each independent:sglang_trace_analyze/orchestrator/structure_mapper.py: in_map_one, when a call stack is present and the layer is resolved,confidencewas never assigned, so the function raisedUnboundLocalError. It now defaults to"high", which is whattest_build_mapping_with_call_stackalready expects. Same commit:harness_evolve/tests/test_publish.pyresolves theharness_defaultseed relative to the test file instead of a machine-specific absolute path (3 tests).tests/e2e/*(4 tests):scoreboard.list_claims()returns one row per GPU, free or held. The tests assumed a released slot disappears from the list. They now only count rows withstatus == "held".test_remote_job_flow's during-run check had the same flaw and always passed; it now also requiresstatus == "held".gen_infer_framework_cpp/tests(2 tests): drop thetask_typeargument fromStateStore.init_or_resume, which no longer accepts it.Test results
Fresh clone, CI-style install,
METAINFER_DISABLE_NETWORK=1 PYTHONDONTWRITEBYTECODE=1 python -m pytest:The 6 that remain are unrelated to this change and are described in the linked issue. On 3.10 one timing-sensitive test in
test_supervisor.pyalso failed intermittently when I ran several suites in parallel; it passes on its own.No production code other than the
structure_mapper.pyfix is touched.Related issue: #39