Merge aloha_obj_wmerged + add standalone_mg7 to the check language comparison#4
Closed
oliviermattelaer wants to merge 482 commits into
Closed
Merge aloha_obj_wmerged + add standalone_mg7 to the check language comparison#4oliviermattelaer wants to merge 482 commits into
oliviermattelaer wants to merge 482 commits into
Conversation
Simpler collect events
…iagram generation (#248) * Support tuple charges in merged flavor conservation --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: oliviermattelaer <33414646+oliviermattelaer@users.noreply.github.com>
Fix wrong charm quark couplings in FD gauge Z-_quark interaction ----- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
…#250) Adding support of FD gauge for the standalone_cpp output mode Agent-Logs-Url: https://github.com/mg5amcnlo/mg5amcnlo/sessions/8f746958-03e9-476b-a89a-e88616d674f4
Move get_color.f to Source/MODEL with select case Agent-Logs-Url: https://github.com/mg5amcnlo/mg5amcnlo/sessions/bb15ed83-1365-439d-9724-0bcca2933c7f
Agent-Logs-Url: https://github.com/mg5amcnlo/mg5amcnlo/sessions/789c1924-e92f-4a0a-b767-73d4173dc29f Co-authored-by: oliviermattelaer <33414646+oliviermattelaer@users.noreply.github.com>
Agent-Logs-Url: https://github.com/mg5amcnlo/mg5amcnlo/sessions/5462e87d-c646-4d4d-b88d-da295c70d123 Co-authored-by: oliviermattelaer <33414646+oliviermattelaer@users.noreply.github.com>
Agent-Logs-Url: https://github.com/mg5amcnlo/mg5amcnlo/sessions/c9c2f653-615a-4f19-9e7c-12ec8d9e054b Co-authored-by: oliviermattelaer <33414646+oliviermattelaer@users.noreply.github.com>
EVA (effective vector approximation) collision tests for w+ w- > t t~ used a bare `output`, now defaulting to mg7. Pin to `output madevent`. Verified both pass. No mg7 equivalents: EVA is a madevent-only beam mode for now (consistent with test_eva_collision). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
`pip install ./madspace` failed on the CI runner (ubuntu-22.04 / old pip) while preparing metadata: scikit-build-core's vendored pyproject_metadata calls packaging.utils.canonicalize_name(..., validate=True) and references packaging.utils.InvalidName, both introduced in packaging 22.0. The build- isolation env was falling back to the older system packaging, raising "canonicalize_name() got an unexpected keyword argument 'validate'" / "module 'packaging.utils' has no attribute 'InvalidName'". Add packaging>=22 to [build-system].requires so build isolation always installs a compatible packaging. Verified the source build still succeeds and madspace imports. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bare `output` now defaults to mg7; these tests all drive MadEventCmdShell / launch / generate_events and need a MadEvent directory. Pin to `output madevent`: test_customised_madevent_via_run_card, test_eft_running, test_complex_mass_scheme, test_width_scan, test_e_e_collision, test_add_time_of_flight, test_w_production_with_ms_decay, test_w_production_with_PA_decay(+inline), test_generation_heft, test_polarization_top_decay, test_rivet_from_file. Verified: 11/12 pass locally. test_rivet_from_file gets through output + event generation and only fails at the local `rivet` executable step (rivet not installed here) -- an environment issue, not the pin; it uses rivet from the heptools cache in CI. No mg7 equivalents in this batch: these are run-card / width / EVA-ish / MadSpin / rivet / decay-chain madevent features (MadSpin and these run modes are not on the mg7 side yet). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… helpers) Add two non-xfail mg7 cross-section trackers (per request), and refactor the shared mg7-run helpers to module level so they are usable from any test class: - test_e_e_collision_mg7 (e+ e- > e+ e-): mg7 has no lepton-beam support -- its run_card.toml only carries a hadron PDF, so generate_events aborts with "PID 11 not found in pdf grid". Asserts the physical 155.9 pb; red tracker. - test_generation_heft_mg7 (HEFT g g > b b~ HIW<=1): mg7 runs but the cross section is ~257x below physical (~1.6e6 vs 4.117e8 pb). Asserts the physical reference; red tracker for the mg7 ggH-coupling normalisation. _mg7_datadir_or_skip / _run_mg7_xsec moved from TestMECmdShell to module-level functions (test passed as first arg) so TestMEfromfile can use them too; the three existing callers updated. Both new tests verified to fail for the right reasons (failures=2, no errors) and self-skip without the mg7 stack. Wired both as (red, tracking) jobs in acceptancetest_mg7.yml. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add test_standalone_mg7_vs_cpp: generates p p > e+ e- QCD=0 and checks that the standalone_mg7 (madmatrix) matrix elements reproduce the standalone_cpp ones. standalone_mg7 ships a UMAMI-based check_sa.exe whose "matrix" mode is by design identical to the Fortran/C++ check drivers; the test compares the per-flavour MEs as sorted multisets within a relative tolerance (the backends print at different precision and may order flavours differently). A massless final state is used so both drivers evaluate the same default 1000 GeV phase-space point (avoiding the energy auto-bump that the Fortran/C++ ./check applies but check_sa.exe does not). Adds a shared _assert_me_lists_close helper and wires the test into the acceptance workflow. Verified passing (~12s). Note: not woven into the existing heavier comparison tests -- test_standalone_cpp_fd_output_consistency (6-particle / 104-flavour / two gauges) becomes >30 min with a third backend, and the MSSM benchmark test_standalone_cpp uses a massive final state where ./check auto-bumps the energy but check_sa.exe does not; a dedicated fast massless comparison avoids both issues. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…tion) The earlier madspace/pyproject.toml packaging>=22 build-require was not enough: the runner ships an ancient pip (pip/_vendor/pep517, ~22.0.2) whose build isolation still leaves scikit-build-core's vendored pyproject_metadata importing an old packaging, so 'pip install ./madspace' keeps failing in metadata preparation (canonicalize_name validate kwarg / packaging.utils.InvalidName, both need packaging>=22). Upgrade pip (and packaging) in checkout_mg5 before the madspace build so a modern pip performs build isolation correctly. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…standalone
Replace standalone_cpp by standalone_mg7 (madmatrix) in the FD/unitary-gauge
consistency test: it now checks that the standalone_mg7 matrix elements agree
with the Fortran standalone ones in FD gauge, in unitary gauge, and across the
two gauges (gauge invariance). get_values drives standalone_mg7's UMAMI
check_sa.exe ('matrix' mode == the Fortran check driver), and the per-flavour
values are compared as sorted multisets via _assert_me_lists_close.
The helper gains an atol: this process's MEs are O(1e-20), i.e. at the
floating-point noise floor where the Fortran and cudacpp backends differ, so an
absolute floor (atol=1e-7) is applied -- matching the leniency the original
cpp-vs-standalone assertAlmostEqual already had there. Verified passing (~6.4 min;
test_standalone_mg7_vs_cpp still green with the backward-compatible signature).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
madspace requires Python >=3.11 but the ubuntu-22.04 runners ship Python 3.10, so checkout_mg5's 'pip install ./madspace' fails with "Package 'madspace' requires a different Python: 3.10.12 not in '>=3.11'". Pin every test-CI runner to ubuntu-24.04 (Python 3.12): replace all runs-on: ubuntu-22.04 and runs-on: ubuntu-latest with ubuntu-24.04 across the acceptance / unittest / parallel / IOtest / aloha / warm_cache workflows, and collapse the os: [ubuntu-22.04, ubuntu-24.04] matrices to [ubuntu-24.04]. The lhapdf/heptools caches are keyed by $ImageOS so they track the runner image (warm_cache now (re)builds the ubuntu-24 caches). madspace.yml (wheel publisher) is left untouched. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
After the standalone_cpp benchmark (g g > go go, 6.4739191), also generate standalone_mg7 and confirm its matrix element matches standalone_cpp for this massive BSM process. The Fortran/C++ ./check auto-bumps the CM energy to 2*total_mass for the heavy gluinos but check_sa.exe does not, so both are evaluated at the same explicit above-threshold energy (5000 GeV). Verified passing (~18s); standalone_mg7 reproduces standalone_cpp (5.7733905 vs 5.7733902). This extends standalone_mg7 coverage to a massive/MSSM process, complementing test_standalone_mg7_vs_cpp (massless p p > e+ e-) and the FD/unitary gauge test_standalone_cpp_fd_output_consistency. The remaining standalone comparison tests (decay-chain output_consistency, density) sit at the numerical-noise floor (matrix elements ~0 at the test phase-space point), where a cross-backend value comparison is not meaningful, so they are intentionally left out. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…failure) MadSpin's MC-integrated partial-width computation (run_onshell BR path) generates a width_calculator via a bare 'output', which now defaults to mg7. The mg7 launcher requires lhapdf, so the onshell decay modes failed in CI (e.g. test_short_madspin_singletop[onshell_decay_chain]) on runners without lhapdf, while passing locally where lhapdf is present. The legacy (full_decay_chain) path uses factorized analytic widths and is unaffected, which is why only the onshell modes broke. Force 'output madevent' for the width_calculator, mirroring the existing pins for the production (madspin_comparator) and the decay matrix element (interface_madspin.py:1391/1395). Verified test_short_madspin_singletop still passes (228s) -- madevent is the structure this width computation expects. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
When a MadSpin mode fails (or produces no decayed LHE), surface the per-mode madspin.log on stdout so the CI output shows why -- otherwise only the log path (on the runner, inaccessible after the job) was reported. Gated by the active -l logging level: always at DEBUG (even on success), only on failure at INFO, and never when quieter. Verified the gating (INFO: success=no/fail=yes; WARNING: never) and the dump formatting in isolation. This makes the recurring test_short_madspin_singletop[onshell_decay_chain] CI failure self-diagnosing on the next run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On ubuntu-24 / Python 3.12, numpy>=1.26's f2py uses the meson backend to compile the Fortran matrix-element wrappers (e.g. all_matrix2py.so for MadSpin/reweight), which needs the meson and ninja executables. They were absent on the runner, so the f2py compile failed with FileNotFoundError: 'meson' (surfacing as the test_short_madspin_singletop[onshell_decay_chain] failure once the MadSpin log was printed). Install meson + ninja alongside numpy into the pip-numpy cache; restore-pip-cache already exposes that dir on PYTHONPATH and its bin/ on PATH, so the executables are found wherever the cache is restored. Verified 'pip install --target' places runnable meson/ninja in bin/. NOTE: requires rebuilding the pip-numpy-ubuntu24 cache (re-run warm_cache with the cache reset / delete the existing pip-numpy-$ImageOS cache) for this to take effect. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the 'github.ref_name == 3.x || test_ci' job guards (8 job-level + 1 always()-wrapped) with main/test_ci plus the workflow_dispatch and schedule event triggers, so the cache-warming jobs actually run on the main branch and on manual/scheduled invocations. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The W width computation (w+ > all all under loop_sm, tree-level) used a bare
'output', now defaulting to mg7. launch then hit the mg7 launcher
('Can't load lhapdf module' -- a width needs no PDF) and produced no madevent
HTML/results.pkl, so load_result('run_01') returned None and the test errored
with TypeError. Pin to 'output madevent'. Verified passing (18s).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bare `output <dir>` now defaults to mg7 (ProcessExporterMadMatrix) after the aloha merge, which produces no Cards/me5_configuration.txt and no MadEvent launcher. decay_comparator.py then builds a MadEventCmd on that directory and runs calculate_decay_widths/launch, failing with FileNotFoundError; test_madevent_ufo_aloha_merged likewise looked for the Fortran DHELAS ALOHA routines that mg7 never emits. Pin all five output calls in decay_comparator (has_same_decay, check_3body, get_2body) and the merged ufo_aloha test to `output madevent`. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
push: has no branch filter, so same-repo PR branches are already covered by the push event; the pull_request run was a duplicate. Add the existing `github.event_name == 'push' || pull_request.head.repo.fork == true` guard (already used by 8 jobs) to the remaining 35 jobs so non-fork PRs skip the redundant run while fork PRs (no push trigger on base repo) still run. Co-Authored-By: Claude Opus 4.8 <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
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
Verification
`check language e+ e- > a a` via the CLI shows all four backends passing; the 9 language/flavor unit tests pass; `check_sa.exe` matrix and perf modes both run.
🤖 Generated with Claude Code