SUSY Run 2

[ChrisJChang]

This adds the code for the SUSY Run 2 project, which includes many new analyses, and ColliderBit code.

Do not be scared by the very large number of changed files. This PR includes many spelling fixes in comments and removed trailing whitespace.

I have added a few reviewers, but this may change. Please wait to start your review until I say (I may make some more small improvements first)

Some tests for this PR that have been run (locally):

• CBS runs on a single analysis
• ColliderBit_CMSSM.yaml works
• Linux build

Some tests that should be run (feel free to add to):

• Mac build
• Github CI runners (currently failing on master)

[Pengxuan-Zhu-Phys]

I did a local pass on PR #574 from an isolated checkout of SUSYRun2_mastermerge and I do not think this is ready for blanket approval yet.

The main PR-specific issue I found is in ColliderBit/src/analyses/Analysis_ATLAS_13TeV_2LEPsoft_139invfb.cpp: the low-pT muon preselection loop iterates over event->electrons() instead of event->muons(). That duplicates electron tracks and drops the intended muon contribution in the preselected track collection. I fixed this locally (Line 214) as:

- for (const HEPUtils::Particle* mu : event->electrons())
+ for (const HEPUtils::Particle* mu : event->muons())

[Pengxuan-Zhu-Phys]

Detailed Review Comments for PR #574

NOTE: This comment is completed with AI assistance, basically make a summary of file changes
Here is a collection of the detailed review comments for the local fixes and portability issues I found while testing the SUSYRun2_mastermerge branch on macOS with the clang-based preset.
CmakeUserPresets.json

For context, the macOS build issues I reported were reproduced locally on the following machine/configuration:

• Hardware: Apple M4 Pro
• Architecture: arm64 (Apple Silicon)
• OS: macOS 26.3.1 (build 25D2128)
• Xcode SDK: MacOSX 26.2
• SDKROOT: /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
• CMake: 4.1.1
• C compiler: Homebrew clang 19.1.7
• C++ compiler: Homebrew clang++ 19.1.7
• Fortran compiler: GNU Fortran 15.1.0 (Homebrew GCC 15.1.0)
• Python: 3.10.12

The comments below are written so they can be pasted directly into GitHub review comments. For each file, I have included:

1. What the issue is.
2. Why it matters.
3. A ready-to-paste English review comment.

1. ColliderBit/src/analyses/Analysis_ATLAS_13TeV_2LEPsoft_139invfb.cpp

What changed locally

- for (const HEPUtils::Particle* mu : event->electrons())
+ for (const HEPUtils::Particle* mu : event->muons())

Why this matters

This is a correctness bug in the analysis code. The low-pT muon contribution to preselectedTracks is being read from the electron collection, which duplicates electron tracks and omits the intended soft-muon tracks.

Suggested review comment

This loop is filling the low-pT muon track contribution from event->electrons() rather than event->muons(). As written, electron tracks are duplicated and soft muons are omitted from preselectedTracks, which changes the track collection used by the analysis. This should iterate over event->muons() instead.

2. ColliderBit/include/gambit/ColliderBit/analyses/AnalysisMacros.hpp

What changed locally

- _cutflows.addCutflow(SR, {"Preselection", ##__VA_ARGS__, "Final"});
+ _cutflows.addCutflow(SR, {"Preselection" __VA_OPT__(,) __VA_ARGS__, "Final"});

- ADD_CUTFLOW(NAME, ##__VA_ARGS__)
+ ADD_CUTFLOW(NAME __VA_OPT__(,) __VA_ARGS__)

- DEFINE_SIGNAL_REGION(name,## __VA_ARGS__)
+ DEFINE_SIGNAL_REGION(name __VA_OPT__(,) __VA_ARGS__)

Why this matters

The original macro form relies on the GNU , ##__VA_ARGS__ extension. Under the macOS clang toolchain used for this branch, the build fails with -pedantic-errors because clang rejects that extension as -Wgnu-zero-variadic-macro-arguments.

This means the branch is not build-clean on macOS/clang even when the overall preset and toolchain are set up correctly.

Suggested review comment

This macro relies on the GNU , ##__VA_ARGS__ extension. Under the macOS clang toolchain used for this branch, -pedantic-errors rejects that form (-Wgnu-zero-variadic-macro-arguments), which blocks the build in ColliderBit. Please switch these variadic macros to a standard form such as __VA_OPT__(,) __VA_ARGS__ so the analysis macros expand cleanly on clang as well as GCC.

3. cmake/contrib.cmake

What changed locally

- CMAKE_COMMAND ${CMAKE_COMMAND} ..
- CMAKE_ARGS -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} -DCMAKE_CXX_FLAGS=${HEPMC_CXX_FLAGS} -DHEPMC3_ENABLE_ROOTIO=${HEPMC3_ROOTIO} -DCMAKE_INSTALL_PREFIX=${HEPMC_PATH}/local -DCMAKE_INSTALL_LIBDIR=${HEPMC_PATH}/local/lib -DHEPMC3_ENABLE_PYTHON=OFF -DHEPMC3_ENABLE_SEARCH=ON -DHEPMC3_BUILD_STATIC_LIBS=OFF -DHEPMC3_CXX_STANDARD=${HEPMC3_STD}
+ CMAKE_COMMAND ${CMAKE_COMMAND}
+ CMAKE_ARGS -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} -DCMAKE_CXX_FLAGS=${HEPMC_CXX_FLAGS} -DCMAKE_EXE_LINKER_FLAGS=${CMAKE_EXE_LINKER_FLAGS} -DCMAKE_SHARED_LINKER_FLAGS=${CMAKE_SHARED_LINKER_FLAGS} -DCMAKE_MODULE_LINKER_FLAGS=${CMAKE_MODULE_LINKER_FLAGS} -DCMAKE_POLICY_VERSION_MINIMUM=3.5 -DHEPMC3_ENABLE_ROOTIO=${HEPMC3_ROOTIO} -DCMAKE_INSTALL_PREFIX=${HEPMC_PATH}/local -DCMAKE_INSTALL_LIBDIR=${HEPMC_PATH}/local/lib -DHEPMC3_ENABLE_PYTHON=OFF -DHEPMC3_ENABLE_SEARCH=ON -DHEPMC3_BUILD_STATIC_LIBS=OFF -DHEPMC3_CXX_STANDARD=${HEPMC3_STD}

+ if(OpenMP_omp_LIBRARY)
+   get_filename_component(YODA_OPENMP_LIBDIR "${OpenMP_omp_LIBRARY}" DIRECTORY)
+   set(YODA_CONFIG_LDFLAGS "${YODA_CONFIG_LDFLAGS} -L${YODA_OPENMP_LIBDIR} -Wl,-rpath,${YODA_OPENMP_LIBDIR}")
+ endif()

- CONFIGURE_COMMAND ./configure CXX=${CMAKE_CXX_COMPILER} CXXFLAGS=${FJCONTRIB_CXX_FLAGS} --fastjet-config=${fastjet_dir}/fastjet-config --prefix=${fastjet_dir}/local
+ CONFIGURE_COMMAND ./configure CXX=${CMAKE_CXX_COMPILER} CXXFLAGS=${FJCONTRIB_CXX_FLAGS} LDFLAGS=${FJCONTRIB_LINKER_FLAGS} --fastjet-config=${fastjet_dir}/fastjet-config --prefix=${fastjet_dir}/local

Why this matters

The ExternalProject configuration here is incomplete for a macOS clang build:

• YODA needed the OpenMP library path propagated explicitly on Darwin.
• HepMC3 needed parent linker flags propagated into the sub-build.
• HepMC3 also needed a compatible CMake policy floor.
• The contrib-side fjcontrib configure step also needed the linker flags propagated.

Without these, the branch does not build cleanly on macOS even when the top-level configuration is otherwise correct.

Suggested review comment

The ExternalProject configuration here is not robust enough for the macOS clang toolchain. In local builds, YODA needed the OpenMP linker path propagated explicitly on Darwin, and HepMC3 also needed the parent linker flags and a compatible CMake policy minimum forwarded into the sub-build. The contrib-side fjcontrib configure step likewise needed the linker flags propagated. Without that, the branch does not build cleanly on macOS even when the top-level preset is otherwise correct.

4. cmake/backends.cmake

What changed locally

+ set(fastjet_patch_dir "${PROJECT_SOURCE_DIR}/Backends/patches/${name}/${ver}")
- PATCH_COMMAND ""
+ PATCH_COMMAND patch -p1 < ${fastjet_patch_dir}/patch_${name}_${ver}.dif

- set(FJCONTRIB_CXX_FLAGS "${FJ_CXX_FLAGS} -std=c++17 -I${dir}/RecursiveTools")
+ set(FJCONTRIB_CXX_FLAGS "${FJ_CXX_FLAGS} -iquote${dir}/RecursiveTools")

+ if(OpenMP_omp_LIBRARY)
+   get_filename_component(FJCONTRIB_OPENMP_LIBDIR "${OpenMP_omp_LIBRARY}" DIRECTORY)
+   set(FJCONTRIB_CXX_FLAGS "${FJCONTRIB_CXX_FLAGS} -L${FJCONTRIB_OPENMP_LIBDIR} -Wl,-rpath,${FJCONTRIB_OPENMP_LIBDIR}")
+ endif()
+ if(${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+   set(FJCONTRIB_CXX_FLAGS "${FJCONTRIB_CXX_FLAGS} -Wl,-headerpad_max_install_names")
+ endif()
+ set(FJCONTRIB_LD_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${NO_FIXUP_CHAINS}")

- CONFIGURE_COMMAND ./configure CXX=${CMAKE_CXX_COMPILER} CXXFLAGS=${FJCONTRIB_CXX_FLAGS} --fastjet-config=${fastjet_dir}/fastjet-config --prefix=${fastjet_dir}/local
+ CONFIGURE_COMMAND ./configure CXX=${CMAKE_CXX_COMPILER} CXXFLAGS=${FJCONTRIB_CXX_FLAGS} LDFLAGS=${FJCONTRIB_LD_FLAGS} --fastjet-config=${fastjet_dir}/fastjet-config --prefix=${fastjet_dir}/local

Why this matters

The backend build logic here was not sufficient for the macOS clang build:

• fastjet 3.4.0 needed a real patch step.
• fjcontrib could not be built with the current flag handling.
• Forcing -std=c++17 was incompatible with the bundled fastjet sources because of older code patterns.
• The RecursiveTools include path also needed more careful handling with libc++.
• Darwin needed additional linker handling for the shared-install step.

As written, these backend steps left the branch broken on macOS.

Suggested review comment

The backend build logic here is not sufficient for the macOS clang configuration used by this branch. In local testing, fastjet 3.4.0 required an actual patch step, and fjcontrib could not be built with the current flag handling: forcing -std=c++17 is incompatible with the bundled fastjet sources, the RecursiveTools include path needs more careful handling with libc++, and Darwin also needs the appropriate linker flags for the shared-install step. As written, this backend setup leaves the macOS build broken.

5. Backends/include/gambit/Backends/backend_types/Pythia_8_312/abstract_Logger.h

What changed locally

+ #include <map>

- virtual ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> begin() =0;
- virtual ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> end() =0;
- virtual ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> begin() const =0;
- virtual ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> end() const =0;
+ virtual ::std::map<std::string, int>::iterator begin() =0;
+ virtual ::std::map<std::string, int>::iterator end() =0;
+ virtual ::std::map<std::string, int>::const_iterator begin() const =0;
+ virtual ::std::map<std::string, int>::const_iterator end() const =0;

Why this matters

The wrapper interface uses libstdc++ implementation details such as std::_Rb_tree_iterator. Those names are not part of the C++ standard library API and fail under libc++ on macOS.

Using the public iterator types from the actual container fixes the portability problem.

Suggested review comment

This wrapper interface uses implementation-specific iterator types such as std::_Rb_tree_iterator, which are not portable and fail under libc++ on macOS. The wrapper should use the public iterator type from the actual container, e.g. std::map<std::string, int>::iterator, instead of depending on libstdc++ internals. As written, this blocks the backend build with the clang/libc++ toolchain.

6. Backends/include/gambit/Backends/backend_types/Pythia_8_312/wrapper_Logger_decl.h

What changed locally

+ #include <map>

- ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> begin();
- ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> end();
- ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> begin() const;
- ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> end() const;
+ ::std::map<std::string, int>::iterator begin();
+ ::std::map<std::string, int>::iterator end();
+ ::std::map<std::string, int>::const_iterator begin() const;
+ ::std::map<std::string, int>::const_iterator end() const;

Why this matters

The declarations repeat the same non-portable dependency on internal libstdc++ tree iterator types. That is not valid across standard library implementations and breaks the macOS clang/libc++ build.

Suggested review comment

The declaration here hard-codes std::_Rb_tree_iterator / std::_Rb_tree_const_iterator, which are libstdc++ implementation details rather than standard library API. That makes the generated backend wrapper fail on macOS with libc++. Please use the corresponding std::map<std::string, int>::iterator and const_iterator types instead.

7. Backends/include/gambit/Backends/backend_types/Pythia_8_312/wrapper_Logger_def.h

What changed locally

- inline ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> Logger::begin()
+ inline ::std::map<std::string, int>::iterator Logger::begin()

- inline ::std::_Rb_tree_iterator<std::pair<const std::basic_string<char>, int>> Logger::end()
+ inline ::std::map<std::string, int>::iterator Logger::end()

- inline ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> Logger::begin() const
+ inline ::std::map<std::string, int>::const_iterator Logger::begin() const

- inline ::std::_Rb_tree_const_iterator<std::pair<const std::basic_string<char>, int>> Logger::end() const
+ inline ::std::map<std::string, int>::const_iterator Logger::end() const

Why this matters

The method definitions repeat the same libc++ portability problem from the abstract interface and declaration. The backend wrapper must use public container iterator types consistently rather than relying on implementation-specific tree iterator types.

Suggested review comment

The method definitions here again depend on std::_Rb_tree_iterator and related internal types. That is not portable across standard library implementations and fails with the macOS clang/libc++ build. The definitions need to follow the public container iterator types used in the declarations instead of relying on libstdc++ internals.

[ChrisJChang]

Adding a simple comment to avoid forgetting, but Anders suggested that the BOSS gb.py script's dictionary at the bottom of the file could potentially solve this Rb_Tree_iterator bug, because it takes expanded types generated by castxml, and converts them back to portable types.

[anderkve]

Yeah, and just to add a quick note before I forget this too: The reason why these strange typenames are there in the first place is that these sorts of complicated-looking expanded types are what we get out of CastXML (which does the C++ code parsing) when the original code has some container type like std::map or similar. (And it seems to be to some extent system/library-dependent exactly what these complicated type names are.) So if we simply fix the BOSS-generated code after the fact, the next time we regenerate these files with BOSS, the complicated-looking types will be back. So that's why we have the dictionary at the bottom of gb.py, which we use to do a find and replace during the code generation. So that dictionary just needs to be expanded as we encounter more such cases in the various backends we use.

[ChrisJChang]

So making the suggested change for the variadic arguments in the Macros only works for C++20 and above.

• _cutflows.addCutflow(SR, {"Preselection", ##VA_ARGS, "Final"});

• _cutflows.addCutflow(SR, {"Preselection" VA_OPT(,) VA_ARGS, "Final"});

We often force C++17 in order to build with some backends, e.g. I think rivet forces it to be C++17.

So whilst I support adding C++20 as an option (and I am not looking into that), I am not sure we can force requiring C++20 or above in order to build ColliderBit given that we have other codes requiring C++17 (or sometimes even C++14)

[Pengxuan-Zhu-Phys]

I agree that we should not require C++20 just to resolve this particular macro issue, especially given that GAMBIT/ColliderBit still needs to remain compatible with backends that effectively force C++17, and in some cases even older standards.

It seems the final GPT score was 6/7.

[agbuckley]

Rivet should require at least C++17. We should also be compatible with higher versions, as far as I'm aware: they can be specified via the CXXFLAGS variable, IIRC. Maybe not worth the bump to C++20 for just this issue, but you want to and Rivet is a block on that, it's not intentional and we can/will fix it!

[ChrisJChang]

Since we haven't tried to let GAMBIT build with standards above C++17 yet, it might be that some of the forcing to C++17 can be removed in the case you are building with standards above this. I will experiment with this, but might not end up putting it in this PR.

[ahye]

I'm not quite sure what you are proposing as a solution to the macro problem @ChrisJChang ? We need some sort of fix for the macs. Are you saying we should have an option to compile with C++20?

[ChrisJChang]

I have pushed changes that should hopefully remove this problem entirely. I got rid of the macros that used variadic arguments, in favour of a templated function. This also removes the separate "_NOCUTS" version of the define signal region macro that was being used to fix another macro problem that was previously found.

Now the signal regions can be defined with:

defineSignalRegion("SR-name", "cut1", "cut2", etc);
defineSignalRegions("SR-name-", N, "cut1", "cut2", etc);

As opposed to the previous:

DEFINE_SIGNAL_REGION("SR-name", "cut1", "cut2", etc)
DEFINE_SIGNAL_REGION_NOCUTS("SR-name")
DEFINE_SIGNAL_REGIONS("SR-name-", N, "cut1", "cut2", etc)

(where N is the number of duplicate SRs to create with the same cuts, just appending an index to the SR name)

It would be good if @Pengxuan-Zhu-Phys , you could test whether this fix works for your Mac.

[Pengxuan-Zhu-Phys]

I have pushed changes that should hopefully remove this problem entirely. I got rid of the macros that used variadic arguments, in favour of a templated function. This also removes the separate "_NOCUTS" version of the define signal region macro that was being used to fix another macro problem that was previously found.

Now the signal regions can be defined with:

defineSignalRegion("SR-name", "cut1", "cut2", etc);
defineSignalRegions("SR-name-", N, "cut1", "cut2", etc);

As opposed to the previous:

DEFINE_SIGNAL_REGION("SR-name", "cut1", "cut2", etc)
DEFINE_SIGNAL_REGION_NOCUTS("SR-name")
DEFINE_SIGNAL_REGIONS("SR-name-", N, "cut1", "cut2", etc)

(where N is the number of duplicate SRs to create with the same cuts, just appending an index to the SR name)

It would be good if @Pengxuan-Zhu-Phys , you could test whether this fix works for your Mac.

Thanks a lot for making this change — the move from variadic macros to templated functions looks much cleaner, and it’s great to see the separate _NOCUTS macro gone as well.

I’m afraid I haven’t been able to complete a full build test on my Mac yet, but the reason now seems unrelated to your macro fix.

At the moment I’m blocked by backend downloads: packages such as YODA cannot be fetched from HepForge during the build. I also tried accessing/downloading them manually, but the HepForge side seemed unresponsive from my end. So right now I can’t get to the stage where I can properly test whether the compilation issue is fully resolved on macOS.

Once I can get the missing backends downloaded, I’ll test this again on my Mac and report back.

[Pengxuan-Zhu-Phys]

The changes of

defineSignalRegion("SR-name", "cut1", "cut2", etc);
defineSignalRegions("SR-name-", N, "cut1", "cut2", etc);

works well.

There is another small issue:
For the fastjet / METSignificance path, the macOS failure was mainly a build-system issue: METSignificance could be built before fastjet was ready, causing missing fastjet/PseudoJet.hh. Adding an explicit dependency (METSignificance -> fastjet) resolves that race. I also had to clean up fjcontrib configure/link flags for clang+OpenMP on macOS; after that, fjcontrib config/build succeeded.

Cheers