feat(hermes): propose fixes for SonarQube findings on a schedule
Triage has only ever entered on a failure. Static analysis is the opposite
shape - a standing backlog that never fails a build and so never asks anyone
for attention. On this instance that backlog is 139 open findings on Ariadne
alone, each already naming its file, its line, its rule and what is wrong.
That is better-located evidence than the console text the code-repair flow
normally mines, and it was being thrown away.
This is a second way into the same flow, not a second flow. A scheduled sweep
picks one finding and hands it to the existing proposal path, which is
unchanged: Hermes returns a patch as data, Ariadne validates it against the
file it names, pushes a branch, opens a pull request nobody merges. A finding
arriving from outside the build is not a reason to relax the gates that make a
proposal worth reading, so it does not.
Three deliberate limits. Security hotspots are never fetched: SonarQube models
them as needing human review, the quality gate here fails on exactly that
condition, and an automation that resolved them would be marking them
reviewed without review - defeating the control rather than satisfying it.
Findings already marked won't-fix carry a judgement someone made, and
reopening it produces pull requests that argue with a person. And the sweep
proposes one fix per run by default, because 139 pull requests nobody reads
would make the review gate theatre.
Selection is by SonarQube's own effort estimate rather than severity: effort
is the closest available proxy for the one-anchor change the patch validator
can actually check, so a trivial CRITICAL beats an involved MINOR. An
unparseable estimate is treated as ineligible, not as free.
Off by default. Triage reacts to a failure someone already cares about; this
opens pull requests nobody asked for, and that is a decision an operator makes
deliberately rather than inherits on upgrade.
Branch naming now sanitizes its token, since it arrives from a finding key as
well as a build number and a ref is one of the few places where an unexpected
character stops being cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 22:08:42 -03:00
|
|
|
"""Tests for the scheduled SonarQube quality sweep."""
|
|
|
|
|
|
|
|
|
|
from __future__ import annotations
|
|
|
|
|
|
|
|
|
|
from types import SimpleNamespace
|
|
|
|
|
|
|
|
|
|
import pytest
|
|
|
|
|
|
|
|
|
|
from ariadne.services import hermes_sonar_sweep as module
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
REPO_CFG = {
|
|
|
|
|
"owner": "bstein",
|
|
|
|
|
"repo": "ariadne",
|
|
|
|
|
"allowed_path_prefixes": ["ariadne/"],
|
|
|
|
|
"allowed_suffixes": [".py"],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _issue(**overrides):
|
|
|
|
|
issue = {
|
|
|
|
|
"key": "AZ-1",
|
|
|
|
|
"path": "ariadne/services/hermes_code_defects.py",
|
|
|
|
|
"line": 160,
|
|
|
|
|
"rule": "python:S1172",
|
|
|
|
|
"severity": "MAJOR",
|
|
|
|
|
"type": "CODE_SMELL",
|
|
|
|
|
"effort": "5min",
|
|
|
|
|
"message": "Remove the unused function parameter.",
|
|
|
|
|
}
|
|
|
|
|
issue.update(overrides)
|
|
|
|
|
return issue
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _config(**overrides):
|
|
|
|
|
values = {
|
|
|
|
|
"hermes_sonar_projects": {"ariadne": "ariadne"},
|
|
|
|
|
"hermes_sonar_url": "http://sonar:9000",
|
|
|
|
|
"hermes_sonar_token": "t",
|
|
|
|
|
"hermes_sonar_types": ["CODE_SMELL"],
|
|
|
|
|
"hermes_sonar_severities": [],
|
|
|
|
|
"hermes_sonar_max_per_sweep": 1,
|
|
|
|
|
"hermes_sonar_max_effort_minutes": 20,
|
|
|
|
|
"hermes_sonar_timeout_seconds": 5,
|
|
|
|
|
}
|
|
|
|
|
values.update(overrides)
|
|
|
|
|
return SimpleNamespace(**values)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class _Storage:
|
|
|
|
|
def __init__(self):
|
|
|
|
|
self.events = []
|
|
|
|
|
|
|
|
|
|
def record_event(self, event_type, detail):
|
|
|
|
|
self.events.append((event_type, detail))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def wiring(monkeypatch):
|
|
|
|
|
"""Stub the repo resolution, the client and the proposal flow."""
|
|
|
|
|
|
|
|
|
|
calls = {"proposals": [], "issues": [_issue()], "error": None, "result": {"status": "pr_opened"}}
|
|
|
|
|
monkeypatch.setattr(module.hermes_code_flow, "code_config", lambda config: {})
|
|
|
|
|
monkeypatch.setattr(module.hermes_code_flow, "resolve_repo_config", lambda job, cfg: REPO_CFG)
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module.hermes_sonar_client, "fetch_issues",
|
|
|
|
|
lambda cfg, project: (calls["issues"], calls["error"]),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def _propose(storage, **kwargs):
|
|
|
|
|
calls["proposals"].append(kwargs)
|
|
|
|
|
return calls["result"]
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(module.hermes_code_flow, "propose_code_fix", _propose)
|
feat(hermes): one open proposal per rule, and link every issue to its run
One SonarQube rule is usually one root cause spread across many files. S2208
appears in three Ariadne modules and the cognitive-complexity rule in dozens,
and a sweep with no memory of what it already proposed would open a
near-identical pull request for every instance. Thirty of those get read as
none, which costs more than proposing nothing.
The sweep now skips any rule that already has an open proposal for that
project. The rules under review are read back from the open pull requests'
own titles rather than from a stored index: the pull requests are the thing
that actually exists, an index could disagree with them, and disagreeing is
the one failure mode that matters here. Once the open one is dealt with, the
next instance of that rule becomes eligible again.
This is not the root-cause collapse - it does not make one pull request fix
every instance of a rule, it just stops proposing the same rule repeatedly.
The collapse needs multi-file patch sets, which the frozen patch contract
cannot express yet.
Fails open like every other duplicate check here: an unreadable list yields no
known rules, so a lookup failure costs one extra proposal rather than
silently dropping a whole rule.
Issues now link their run id into the Hermes console, matching what pull
requests already do. Both artifacts claim a model made the call; both should
let a reader open the page where that call is visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-07 00:39:05 -03:00
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module.hermes_code_repair, "open_proposal_incidents",
|
|
|
|
|
lambda cfg: (calls.get("open_incidents", []), None),
|
|
|
|
|
)
|
feat(hermes): propose fixes for SonarQube findings on a schedule
Triage has only ever entered on a failure. Static analysis is the opposite
shape - a standing backlog that never fails a build and so never asks anyone
for attention. On this instance that backlog is 139 open findings on Ariadne
alone, each already naming its file, its line, its rule and what is wrong.
That is better-located evidence than the console text the code-repair flow
normally mines, and it was being thrown away.
This is a second way into the same flow, not a second flow. A scheduled sweep
picks one finding and hands it to the existing proposal path, which is
unchanged: Hermes returns a patch as data, Ariadne validates it against the
file it names, pushes a branch, opens a pull request nobody merges. A finding
arriving from outside the build is not a reason to relax the gates that make a
proposal worth reading, so it does not.
Three deliberate limits. Security hotspots are never fetched: SonarQube models
them as needing human review, the quality gate here fails on exactly that
condition, and an automation that resolved them would be marking them
reviewed without review - defeating the control rather than satisfying it.
Findings already marked won't-fix carry a judgement someone made, and
reopening it produces pull requests that argue with a person. And the sweep
proposes one fix per run by default, because 139 pull requests nobody reads
would make the review gate theatre.
Selection is by SonarQube's own effort estimate rather than severity: effort
is the closest available proxy for the one-anchor change the patch validator
can actually check, so a trivial CRITICAL beats an involved MINOR. An
unparseable estimate is treated as ineligible, not as free.
Off by default. Triage reacts to a failure someone already cares about; this
opens pull requests nobody asked for, and that is a decision an operator makes
deliberately rather than inherits on upgrade.
Branch naming now sanitizes its token, since it arrives from a finding key as
well as a build number and a ref is one of the few places where an unexpected
character stops being cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 22:08:42 -03:00
|
|
|
return calls
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
|
|
|
("effort", "expected"),
|
|
|
|
|
[
|
|
|
|
|
("5min", 5),
|
|
|
|
|
("1h", 60),
|
|
|
|
|
("1h30min", 90),
|
|
|
|
|
("2d", 960),
|
|
|
|
|
("", None),
|
|
|
|
|
("soon", None),
|
|
|
|
|
("30", None),
|
|
|
|
|
(None, None),
|
|
|
|
|
],
|
|
|
|
|
)
|
|
|
|
|
def test_the_effort_estimate_is_parsed_into_minutes(effort, expected) -> None:
|
|
|
|
|
assert module.effort_minutes(effort) == expected
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_cheapest_finding_is_chosen_first() -> None:
|
|
|
|
|
"""Effort is the closest proxy for a change the validator can check."""
|
|
|
|
|
|
|
|
|
|
issues = [
|
|
|
|
|
_issue(key="a", effort="15min"),
|
|
|
|
|
_issue(key="b", effort="5min"),
|
|
|
|
|
_issue(key="c", effort="10min"),
|
|
|
|
|
]
|
|
|
|
|
|
|
|
|
|
assert module.select_issue(issues, REPO_CFG, _config())["key"] == "b"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_severity_breaks_a_tie_on_effort() -> None:
|
|
|
|
|
issues = [_issue(key="a", severity="MINOR"), _issue(key="b", severity="CRITICAL")]
|
|
|
|
|
|
|
|
|
|
assert module.select_issue(issues, REPO_CFG, _config())["key"] == "b"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_key_breaks_a_full_tie_so_the_pick_is_stable() -> None:
|
|
|
|
|
"""A sweep that proposes nothing today must retry the same finding."""
|
|
|
|
|
|
|
|
|
|
issues = [_issue(key="zz"), _issue(key="aa")]
|
|
|
|
|
|
|
|
|
|
assert module.select_issue(issues, REPO_CFG, _config())["key"] == "aa"
|
|
|
|
|
assert module.select_issue(list(reversed(issues)), REPO_CFG, _config())["key"] == "aa"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_an_expensive_finding_is_left_to_a_person() -> None:
|
|
|
|
|
issues = [_issue(effort="4h")]
|
|
|
|
|
|
|
|
|
|
assert module.select_issue(issues, REPO_CFG, _config()) is None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_an_unparseable_effort_is_not_treated_as_free() -> None:
|
|
|
|
|
"""An unreadable estimate is not evidence that the work is small."""
|
|
|
|
|
|
|
|
|
|
assert module.select_issue([_issue(effort="dunno")], REPO_CFG, _config()) is None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
|
|
|
"path",
|
|
|
|
|
["docs/readme.md", "ariadne/thing.txt", "scripts/tool.py", ""],
|
|
|
|
|
)
|
|
|
|
|
def test_a_finding_outside_the_write_allowlist_is_never_chosen(path) -> None:
|
|
|
|
|
assert module.select_issue([_issue(path=path)], REPO_CFG, _config()) is None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_sweep_opens_one_proposal_for_the_chosen_finding(wiring) -> None:
|
|
|
|
|
storage = _Storage()
|
|
|
|
|
|
|
|
|
|
result = module.sweep(storage, _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result == {"proposed": 1, "skipped": [], "projects": 1}
|
|
|
|
|
proposal = wiring["proposals"][0]
|
feat(hermes): one open proposal per rule, and link every issue to its run
One SonarQube rule is usually one root cause spread across many files. S2208
appears in three Ariadne modules and the cognitive-complexity rule in dozens,
and a sweep with no memory of what it already proposed would open a
near-identical pull request for every instance. Thirty of those get read as
none, which costs more than proposing nothing.
The sweep now skips any rule that already has an open proposal for that
project. The rules under review are read back from the open pull requests'
own titles rather than from a stored index: the pull requests are the thing
that actually exists, an index could disagree with them, and disagreeing is
the one failure mode that matters here. Once the open one is dealt with, the
next instance of that rule becomes eligible again.
This is not the root-cause collapse - it does not make one pull request fix
every instance of a rule, it just stops proposing the same rule repeatedly.
The collapse needs multi-file patch sets, which the frozen patch contract
cannot express yet.
Fails open like every other duplicate check here: an unreadable list yields no
known rules, so a lookup failure costs one extra proposal rather than
silently dropping a whole rule.
Issues now link their run id into the Hermes console, matching what pull
requests already do. Both artifacts claim a model made the call; both should
let a reader open the page where that call is visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-07 00:39:05 -03:00
|
|
|
assert proposal["incident_id"] == "sonar/ariadne/python:S1172/AZ-1"
|
feat(hermes): propose fixes for SonarQube findings on a schedule
Triage has only ever entered on a failure. Static analysis is the opposite
shape - a standing backlog that never fails a build and so never asks anyone
for attention. On this instance that backlog is 139 open findings on Ariadne
alone, each already naming its file, its line, its rule and what is wrong.
That is better-located evidence than the console text the code-repair flow
normally mines, and it was being thrown away.
This is a second way into the same flow, not a second flow. A scheduled sweep
picks one finding and hands it to the existing proposal path, which is
unchanged: Hermes returns a patch as data, Ariadne validates it against the
file it names, pushes a branch, opens a pull request nobody merges. A finding
arriving from outside the build is not a reason to relax the gates that make a
proposal worth reading, so it does not.
Three deliberate limits. Security hotspots are never fetched: SonarQube models
them as needing human review, the quality gate here fails on exactly that
condition, and an automation that resolved them would be marking them
reviewed without review - defeating the control rather than satisfying it.
Findings already marked won't-fix carry a judgement someone made, and
reopening it produces pull requests that argue with a person. And the sweep
proposes one fix per run by default, because 139 pull requests nobody reads
would make the review gate theatre.
Selection is by SonarQube's own effort estimate rather than severity: effort
is the closest available proxy for the one-anchor change the patch validator
can actually check, so a trivial CRITICAL beats an involved MINOR. An
unparseable estimate is treated as ineligible, not as free.
Off by default. Triage reacts to a failure someone already cares about; this
opens pull requests nobody asked for, and that is a decision an operator makes
deliberately rather than inherits on upgrade.
Branch naming now sanitizes its token, since it arrives from a finding key as
well as a build number and a ref is one of the few places where an unexpected
character stops being cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 22:08:42 -03:00
|
|
|
assert proposal["job"] == "ariadne"
|
|
|
|
|
assert proposal["build_number"] == "sonar-AZ-1"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_bundle_offers_only_the_file_the_finding_names() -> None:
|
|
|
|
|
"""The finding states the file; ranking console text alongside adds noise."""
|
|
|
|
|
|
|
|
|
|
bundle = module.bundle_for("ariadne", _issue())
|
|
|
|
|
|
|
|
|
|
assert bundle["sonarqube"] == {"project": "ariadne", "issues": [_issue()]}
|
|
|
|
|
assert bundle["jenkins"]["console_failures"] == []
|
|
|
|
|
assert bundle["jenkins"]["console_tail"] == ""
|
feat(hermes): one open proposal per rule, and link every issue to its run
One SonarQube rule is usually one root cause spread across many files. S2208
appears in three Ariadne modules and the cognitive-complexity rule in dozens,
and a sweep with no memory of what it already proposed would open a
near-identical pull request for every instance. Thirty of those get read as
none, which costs more than proposing nothing.
The sweep now skips any rule that already has an open proposal for that
project. The rules under review are read back from the open pull requests'
own titles rather than from a stored index: the pull requests are the thing
that actually exists, an index could disagree with them, and disagreeing is
the one failure mode that matters here. Once the open one is dealt with, the
next instance of that rule becomes eligible again.
This is not the root-cause collapse - it does not make one pull request fix
every instance of a rule, it just stops proposing the same rule repeatedly.
The collapse needs multi-file patch sets, which the frozen patch contract
cannot express yet.
Fails open like every other duplicate check here: an unreadable list yields no
known rules, so a lookup failure costs one extra proposal rather than
silently dropping a whole rule.
Issues now link their run id into the Hermes console, matching what pull
requests already do. Both artifacts claim a model made the call; both should
let a reader open the page where that call is visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-07 00:39:05 -03:00
|
|
|
assert bundle["incident_id"] == "sonar/ariadne/python:S1172/AZ-1"
|
feat(hermes): propose fixes for SonarQube findings on a schedule
Triage has only ever entered on a failure. Static analysis is the opposite
shape - a standing backlog that never fails a build and so never asks anyone
for attention. On this instance that backlog is 139 open findings on Ariadne
alone, each already naming its file, its line, its rule and what is wrong.
That is better-located evidence than the console text the code-repair flow
normally mines, and it was being thrown away.
This is a second way into the same flow, not a second flow. A scheduled sweep
picks one finding and hands it to the existing proposal path, which is
unchanged: Hermes returns a patch as data, Ariadne validates it against the
file it names, pushes a branch, opens a pull request nobody merges. A finding
arriving from outside the build is not a reason to relax the gates that make a
proposal worth reading, so it does not.
Three deliberate limits. Security hotspots are never fetched: SonarQube models
them as needing human review, the quality gate here fails on exactly that
condition, and an automation that resolved them would be marking them
reviewed without review - defeating the control rather than satisfying it.
Findings already marked won't-fix carry a judgement someone made, and
reopening it produces pull requests that argue with a person. And the sweep
proposes one fix per run by default, because 139 pull requests nobody reads
would make the review gate theatre.
Selection is by SonarQube's own effort estimate rather than severity: effort
is the closest available proxy for the one-anchor change the patch validator
can actually check, so a trivial CRITICAL beats an involved MINOR. An
unparseable estimate is treated as ineligible, not as free.
Off by default. Triage reacts to a failure someone already cares about; this
opens pull requests nobody asked for, and that is a decision an operator makes
deliberately rather than inherits on upgrade.
Branch naming now sanitizes its token, since it arrives from a finding key as
well as a build number and a ref is one of the few places where an unexpected
character stops being cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 22:08:42 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_sweep_budget_stops_after_its_quota(wiring) -> None:
|
|
|
|
|
"""A backlog must not become a pull request queue nobody drains."""
|
|
|
|
|
|
|
|
|
|
config = _config(hermes_sonar_projects={"a": "a", "b": "b", "c": "c"})
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), config, {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 1
|
|
|
|
|
assert len(wiring["proposals"]) == 1
|
|
|
|
|
assert any("budget spent" in item for item in result["skipped"])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_declined_proposal_is_reported_and_does_not_spend_budget(wiring) -> None:
|
|
|
|
|
wiring["result"] = {"status": "human_required", "reason": "open_proposal_limit_reached"}
|
|
|
|
|
config = _config(hermes_sonar_projects={"a": "a", "b": "b"})
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), config, {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 0
|
|
|
|
|
assert len(wiring["proposals"]) == 2
|
|
|
|
|
assert "a: open_proposal_limit_reached" in result["skipped"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_no_configured_projects_makes_no_call(wiring) -> None:
|
|
|
|
|
result = module.sweep(_Storage(), _config(hermes_sonar_projects={}), {})
|
|
|
|
|
|
|
|
|
|
assert result == {"proposed": 0, "skipped": ["no_projects_configured"], "projects": 0}
|
|
|
|
|
assert wiring["proposals"] == []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_project_naming_no_repository_is_skipped(monkeypatch, wiring) -> None:
|
|
|
|
|
monkeypatch.setattr(module.hermes_code_flow, "resolve_repo_config", lambda job, cfg: None)
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 0
|
|
|
|
|
assert "maps to no repository" in result["skipped"][0]
|
|
|
|
|
assert wiring["proposals"] == []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_fetch_error_is_reported_not_raised(wiring) -> None:
|
|
|
|
|
wiring["error"] = "sonar fetch http 503"
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 0
|
|
|
|
|
assert result["skipped"] == ["ariadne: sonar fetch http 503"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_project_with_nothing_mechanical_is_reported(wiring) -> None:
|
|
|
|
|
wiring["issues"] = [_issue(effort="8h")]
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
feat(hermes): one open proposal per rule, and link every issue to its run
One SonarQube rule is usually one root cause spread across many files. S2208
appears in three Ariadne modules and the cognitive-complexity rule in dozens,
and a sweep with no memory of what it already proposed would open a
near-identical pull request for every instance. Thirty of those get read as
none, which costs more than proposing nothing.
The sweep now skips any rule that already has an open proposal for that
project. The rules under review are read back from the open pull requests'
own titles rather than from a stored index: the pull requests are the thing
that actually exists, an index could disagree with them, and disagreeing is
the one failure mode that matters here. Once the open one is dealt with, the
next instance of that rule becomes eligible again.
This is not the root-cause collapse - it does not make one pull request fix
every instance of a rule, it just stops proposing the same rule repeatedly.
The collapse needs multi-file patch sets, which the frozen patch contract
cannot express yet.
Fails open like every other duplicate check here: an unreadable list yields no
known rules, so a lookup failure costs one extra proposal rather than
silently dropping a whole rule.
Issues now link their run id into the Hermes console, matching what pull
requests already do. Both artifacts claim a model made the call; both should
let a reader open the page where that call is visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-07 00:39:05 -03:00
|
|
|
assert result["skipped"] == ["ariadne: no new mechanically fixable finding"]
|
feat(hermes): propose fixes for SonarQube findings on a schedule
Triage has only ever entered on a failure. Static analysis is the opposite
shape - a standing backlog that never fails a build and so never asks anyone
for attention. On this instance that backlog is 139 open findings on Ariadne
alone, each already naming its file, its line, its rule and what is wrong.
That is better-located evidence than the console text the code-repair flow
normally mines, and it was being thrown away.
This is a second way into the same flow, not a second flow. A scheduled sweep
picks one finding and hands it to the existing proposal path, which is
unchanged: Hermes returns a patch as data, Ariadne validates it against the
file it names, pushes a branch, opens a pull request nobody merges. A finding
arriving from outside the build is not a reason to relax the gates that make a
proposal worth reading, so it does not.
Three deliberate limits. Security hotspots are never fetched: SonarQube models
them as needing human review, the quality gate here fails on exactly that
condition, and an automation that resolved them would be marking them
reviewed without review - defeating the control rather than satisfying it.
Findings already marked won't-fix carry a judgement someone made, and
reopening it produces pull requests that argue with a person. And the sweep
proposes one fix per run by default, because 139 pull requests nobody reads
would make the review gate theatre.
Selection is by SonarQube's own effort estimate rather than severity: effort
is the closest available proxy for the one-anchor change the patch validator
can actually check, so a trivial CRITICAL beats an involved MINOR. An
unparseable estimate is treated as ineligible, not as free.
Off by default. Triage reacts to a failure someone already cares about; this
opens pull requests nobody asked for, and that is a decision an operator makes
deliberately rather than inherits on upgrade.
Branch naming now sanitizes its token, since it arrives from a finding key as
well as a build number and a ref is one of the few places where an unexpected
character stops being cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 22:08:42 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_sweep_never_takes_down_the_scheduler(monkeypatch) -> None:
|
|
|
|
|
"""Triage shares this scheduler; a background sweep must not stop it."""
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module.hermes_code_flow, "code_config",
|
|
|
|
|
lambda config: (_ for _ in ()).throw(RuntimeError("boom")),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 0
|
|
|
|
|
assert "sweep_failed" in result["skipped"][0]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_project_with_no_job_maps_to_itself() -> None:
|
|
|
|
|
assert module._projects(_config(hermes_sonar_projects={"ariadne": ""})) == {
|
|
|
|
|
"ariadne": "ariadne"
|
|
|
|
|
}
|
|
|
|
|
assert module._projects(SimpleNamespace(hermes_sonar_projects="nope")) == {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_client_config_is_built_from_settings() -> None:
|
|
|
|
|
cfg = module.client_config(_config())
|
|
|
|
|
|
|
|
|
|
assert cfg["sonar_base_url"] == "http://sonar:9000"
|
|
|
|
|
assert cfg["sonar_token"] == "t"
|
|
|
|
|
assert cfg["sonar_types"] == ["CODE_SMELL"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
|
|
|
("attr", "func", "bad", "default"),
|
|
|
|
|
[
|
|
|
|
|
("hermes_sonar_max_per_sweep", "_max_per_sweep", "junk", module.DEFAULT_MAX_PER_SWEEP),
|
|
|
|
|
(
|
|
|
|
|
"hermes_sonar_max_effort_minutes",
|
|
|
|
|
"_max_effort_minutes",
|
|
|
|
|
"junk",
|
|
|
|
|
module.DEFAULT_MAX_EFFORT_MINUTES,
|
|
|
|
|
),
|
|
|
|
|
],
|
|
|
|
|
)
|
|
|
|
|
def test_an_unparseable_bound_falls_back_to_its_default(attr, func, bad, default) -> None:
|
|
|
|
|
assert getattr(module, func)(_config(**{attr: bad})) == default
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_bounds_are_floored() -> None:
|
|
|
|
|
assert module._max_per_sweep(_config(hermes_sonar_max_per_sweep=-4)) == 0
|
|
|
|
|
assert module._max_effort_minutes(_config(hermes_sonar_max_effort_minutes=0)) == 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_sweep_is_disabled_until_an_operator_turns_it_on(monkeypatch) -> None:
|
|
|
|
|
"""It opens pull requests nobody asked for; that is a deliberate choice."""
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(module, "settings", SimpleNamespace(hermes_sonar_enabled=False))
|
|
|
|
|
|
|
|
|
|
assert module.run_hermes_sonar_sweep(_Storage()) == {"status": "disabled"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_default_deployment_leaves_it_off() -> None:
|
|
|
|
|
"""A fresh deployment must not start opening pull requests on its own."""
|
|
|
|
|
|
|
|
|
|
from ariadne.settings import settings as real_settings
|
|
|
|
|
|
|
|
|
|
assert real_settings.hermes_sonar_enabled is False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_scheduled_entry_point_runs_the_sweep_when_enabled(monkeypatch) -> None:
|
|
|
|
|
seen = {}
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module,
|
|
|
|
|
"settings",
|
|
|
|
|
SimpleNamespace(
|
|
|
|
|
hermes_sonar_enabled=True,
|
|
|
|
|
hermes_api_url="http://hermes:8642",
|
|
|
|
|
hermes_api_key="k",
|
|
|
|
|
hermes_run_timeout_seconds=420.0,
|
|
|
|
|
),
|
|
|
|
|
)
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module, "sweep", lambda storage, config, cfg: seen.update(cfg=cfg) or {"proposed": 0}
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert module.run_hermes_sonar_sweep(_Storage()) == {"proposed": 0}
|
|
|
|
|
assert seen["cfg"] == {
|
|
|
|
|
"base_url": "http://hermes:8642",
|
|
|
|
|
"api_key": "k",
|
|
|
|
|
"total_timeout_seconds": 420.0,
|
|
|
|
|
}
|
2026-08-06 22:16:36 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_sweep_prompt_never_claims_a_build_failed() -> None:
|
|
|
|
|
"""A green build framed as a failing test is wrong in the first line read."""
|
|
|
|
|
|
|
|
|
|
from ariadne.services import hermes_code_prompt
|
|
|
|
|
|
|
|
|
|
cfg = {"owner": "bstein", "repo": "ariadne", "base_branch": "master"}
|
|
|
|
|
prompt = hermes_code_prompt.build_patch_prompt(
|
|
|
|
|
"sonar/ariadne/AZ-1",
|
|
|
|
|
module.bundle_for("ariadne", _issue()),
|
|
|
|
|
{**cfg, "allowed_path_prefixes": ["ariadne/"], "allowed_suffixes": [".py"]},
|
|
|
|
|
{"ariadne/services/hermes_code_defects.py": "x"},
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert prompt.startswith("A static-analysis finding, not a build failure.")
|
|
|
|
|
assert "Failing test evidence bundle" not in prompt
|
|
|
|
|
assert "Static analysis evidence bundle:" in prompt
|
|
|
|
|
assert "The build reported these specific defects" not in prompt
|
|
|
|
|
# The triage skill investigates a failed build; there isn't one.
|
|
|
|
|
assert "$triage-titan-test-failures" not in prompt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_build_driven_prompt_is_unchanged() -> None:
|
|
|
|
|
from ariadne.services import hermes_code_prompt
|
|
|
|
|
|
|
|
|
|
bundle = {"jenkins": {"console_tail": "AssertionError", "console_failures": [], "failed_tests": []}}
|
|
|
|
|
prompt = hermes_code_prompt.build_patch_prompt(
|
|
|
|
|
"ariadne/9", bundle, {"owner": "bstein", "repo": "ariadne"}, {"ariadne/a.py": "x"}
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert prompt.startswith("Use $triage-titan-test-failures.")
|
|
|
|
|
assert "Failing test evidence bundle:" in prompt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
|
|
|
("bundle", "expected"),
|
|
|
|
|
[
|
|
|
|
|
({"sonarqube": {"issues": [{"path": "a.py"}]}, "jenkins": {}}, True),
|
|
|
|
|
({"sonarqube": {"issues": []}, "jenkins": {}}, False),
|
|
|
|
|
({"jenkins": {"console_tail": "boom"}}, False),
|
|
|
|
|
# Both present: a real failure is the more urgent framing.
|
|
|
|
|
(
|
|
|
|
|
{"sonarqube": {"issues": [{"path": "a.py"}]}, "jenkins": {"console_tail": "boom"}},
|
|
|
|
|
False,
|
|
|
|
|
),
|
|
|
|
|
("not-a-dict", False),
|
|
|
|
|
],
|
|
|
|
|
)
|
|
|
|
|
def test_a_bundles_origin_decides_the_framing(bundle, expected) -> None:
|
|
|
|
|
from ariadne.services import hermes_code_prompt
|
|
|
|
|
|
|
|
|
|
assert hermes_code_prompt.is_quality_sweep(bundle) is expected
|
feat(hermes): one open proposal per rule, and link every issue to its run
One SonarQube rule is usually one root cause spread across many files. S2208
appears in three Ariadne modules and the cognitive-complexity rule in dozens,
and a sweep with no memory of what it already proposed would open a
near-identical pull request for every instance. Thirty of those get read as
none, which costs more than proposing nothing.
The sweep now skips any rule that already has an open proposal for that
project. The rules under review are read back from the open pull requests'
own titles rather than from a stored index: the pull requests are the thing
that actually exists, an index could disagree with them, and disagreeing is
the one failure mode that matters here. Once the open one is dealt with, the
next instance of that rule becomes eligible again.
This is not the root-cause collapse - it does not make one pull request fix
every instance of a rule, it just stops proposing the same rule repeatedly.
The collapse needs multi-file patch sets, which the frozen patch contract
cannot express yet.
Fails open like every other duplicate check here: an unreadable list yields no
known rules, so a lookup failure costs one extra proposal rather than
silently dropping a whole rule.
Issues now link their run id into the Hermes console, matching what pull
requests already do. Both artifacts claim a model made the call; both should
let a reader open the page where that call is visible.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-07 00:39:05 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_a_rule_already_under_review_is_not_proposed_again(wiring) -> None:
|
|
|
|
|
"""One rule is one root cause; thirty near-identical PRs get read as none."""
|
|
|
|
|
|
|
|
|
|
wiring["open_incidents"] = ["sonar/ariadne/python:S1172/AZ-OTHER"]
|
|
|
|
|
wiring["issues"] = [
|
|
|
|
|
_issue(key="a", rule="python:S1172"),
|
|
|
|
|
_issue(key="b", rule="python:S2208", effort="10min"),
|
|
|
|
|
]
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 1
|
|
|
|
|
assert wiring["proposals"][0]["incident_id"] == "sonar/ariadne/python:S2208/b"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_every_rule_under_review_means_nothing_new_to_propose(wiring) -> None:
|
|
|
|
|
wiring["open_incidents"] = ["sonar/ariadne/python:S1172/AZ-OTHER"]
|
|
|
|
|
wiring["issues"] = [_issue(rule="python:S1172")]
|
|
|
|
|
|
|
|
|
|
result = module.sweep(_Storage(), _config(), {})
|
|
|
|
|
|
|
|
|
|
assert result["proposed"] == 0
|
|
|
|
|
assert result["skipped"] == ["ariadne: no new mechanically fixable finding"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_another_projects_open_proposal_does_not_block_this_one(wiring) -> None:
|
|
|
|
|
wiring["open_incidents"] = ["sonar/metis/python:S1172/AZ-X", "ariadne/408"]
|
|
|
|
|
|
|
|
|
|
assert module.sweep(_Storage(), _config(), {})["proposed"] == 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_the_rules_under_review_are_read_from_the_open_proposals(monkeypatch) -> None:
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module.hermes_code_repair, "open_proposal_incidents",
|
|
|
|
|
lambda cfg: (
|
|
|
|
|
[
|
|
|
|
|
"sonar/ariadne/python:S2208/AZ1",
|
|
|
|
|
"sonar/ariadne/python:S3776/AZ2",
|
|
|
|
|
"sonar/other/python:S1172/AZ3",
|
|
|
|
|
"ariadne/408",
|
|
|
|
|
"sonar/ariadne/malformed",
|
|
|
|
|
],
|
|
|
|
|
None,
|
|
|
|
|
),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert module.proposed_rules(REPO_CFG, "ariadne") == {"python:S2208", "python:S3776"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_an_unreadable_proposal_list_costs_a_duplicate_not_a_silent_rule(monkeypatch) -> None:
|
|
|
|
|
"""Failing open here means one extra PR; failing closed loses a whole rule."""
|
|
|
|
|
|
|
|
|
|
monkeypatch.setattr(
|
|
|
|
|
module.hermes_code_repair, "open_proposal_incidents", lambda cfg: ([], "http 500")
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert module.proposed_rules(REPO_CFG, "ariadne") == set()
|