gh-209: implement mgclass by zsirap · Pull Request #302 · cloe-org/cloelib

zsirap · 2025-10-23T01:11:38Z

🚀 Pull Request Checklist

✅ Summary

closes #290

🔄 Changes

allows the user to call additional parameters for the different models in MGCLASS
allows to import separate CLASS and MGCLASS in the same python script

🛠 How to Test

📝 Documentation

This PR updates documentation

🏗 Related Issues

Resolves #290

📸 Screenshots (if applicable)

📌 Additional Notes

✅ PR Checklist for Developers

I have titled this PR before merging as "gh-#:", where "#" represents the task it closes
I have run locally pre-commit using pre-commit run --all-files
I have tested my changes locally
No new warnings or errors introduced
I have updated documentation (if applicable)
My changes do not introduce breaking changes (i.e: the package still gets installed)
I have added unit tests (if applicable)
I have consistently updated the GitHub information for the project, including milestones, task types, and other relevant details.

✅ PR Checklist for Reviewers

zsirap · 2025-10-23T01:19:00Z

I am still getting style errors that I don't know the origin (though locally the style test and mglass test pass) if you can help on that @gcanasherrera

update: i see other PRs have some similar issues ... later i might recheck again to see if i manage to solve mine .. for time being i labeled the PR as draft but it can be already reviewed at your or any other reviewer ease

gcanasherrera · 2025-10-23T11:13:18Z

Make sure this title of this PR is properly named

zsirap · 2025-10-29T02:27:20Z

not a draft anymore ... it only fails now because MGCLASS is an external code and is not automatically installed ... otherwise it passes locally ...

gcanasherrera · 2025-10-29T10:16:13Z

Hi @zsirap,

Before I jump into the reviewing and into addressing the issue on how to install MGCLASS into the CI environment, could you please make sure that you are following properly the requests specified in the PR Checklist for Developers?

For instance, you have clicked as done "I have titled this PR before merging as "gh-#:", where "#" represents the task it closes", however you have not done it.

Also, make sure you have a proper picture that you would like to share in your GitHub profile so that the all-contributors bot gives you visibility accordantly (if you want to keep the default GitHub picture, too, it's also fine).

zsirap · 2025-10-29T10:53:33Z

For instance, you have clicked as done "I have titled this PR before merging as "gh-#:", where "#" represents the task it closes", however you have not done it.

ah i thought you meant by this previous comment changing the title to make it more suitable ...

Make sure this title of this PR is properly named

my bad .. ok done now i hope ...

Before I jump into the reviewing

no rush .. at your ease @gcanasherrera ..

if you want to keep the default GitHub picture, too, it's also fine

for the moment ... maybe later

gcanasherrera · 2025-10-29T20:49:47Z

Hi @zsirap,

Thanks for this. As you're one of the main users of MGCLASS, I was wondering: why don't you make it pip installable so it is easier for us to install, also in the CI?
CC- @matmartinelli

gcanasherrera

Please, adapt the names to MGCLASS instead of CLASS only everywhere

gcanasherrera · 2026-05-04T12:00:13Z

You're assigned to this one @santiagocasas, you're more than a CLASS person than me.

gcanasherrera · 2026-05-04T16:53:12Z

@zsirap (CC- @santiagocasas):

note how class is now interfacing in the main branch, where it requires an optional linear module to be passed to the non-linear module. This is to allow for cloelike interface. Please, follow the same approach in mgclass

santiagocasas · 2026-05-08T12:38:15Z

@gcanasherrera yes, I am taking a look into this. But regarding your comment on the linear module being passed, I think we should make mgclass and hi_class more modular. We are currently duplicating a lot of boilerplate code for background for instance, and this is not only dangerous but difficult to maintain (cloelib/cosmology/mgclass_cosmology.py).

Let me think about this for a day.

santiagocasas · 2026-05-08T12:38:33Z

thanks @zsirap for your work on this so far

santiagocasas

all in all, quite close to finishing, but a few things need revision. Especially Protocol adherence and the physical error (double-check!) in Omega_cb and so on. See below! thanks!

santiagocasas · 2026-05-12T14:39:08Z

        run: |
          mamba run -n $ENV_NAME python -m pip install --upgrade pip
          mamba run -n $ENV_NAME python -m pip install ".[camb,classy,euclidemu2,baccoemu]"
+      - name: Install external prereq using pip


please don't install mgclassy like that. Use the same format as for the other external codes and solve the conflict, so that all codes are there. There is no need to specify the version, this is installed via the pyproject.toml and this is set there already: mgclassy = ["mgclassy==2.9.4.5"]

I agree but for a reason not yet known, mgclass is not installing if i use the same format as the others (I might give it a try again to see if it works again but for the moment if there is urgency we can merge it like that)

a general comment mentioned before and related to this and subsequent comments is that there is a new version of MGCLASS, not released yet, that will be built above a newer or the latest version of CLASS ... but here we started to implement MGCLASS in CLOE as it is now based on an older version of CLASS because also we wanted to be quick and pragmatic and we wanted it to not install above CLASS like other patches do ... and so in trying to do so, one thing led to another such as had to make a pip installable version of what we have now ...

I will comment on this below

santiagocasas · 2026-05-12T14:40:59Z

+        self.Omega_cdm0 = Omega_cdm0
+        self.Omega_k0 = Omega_k0
+        self.As = As
+        self.ns = ns


please rebase to latest version of main and add here the new requirement in the Background protocol, the parameter alpha_s. See here:

cloelib/cloelib/cosmology/class_cosmology.py

Line 39 in c33e526

alpha_s: float = 0.0,

santiagocasas · 2026-05-12T14:44:27Z

+                [
+                    (self.results.Omega_b() + self.results.Omega0_cdm())
+                    * (1 + z) ** 3.0
+                    / (self.hubble_parameter(z) / self.H0) ** 0.5


why the square root of E(z)? , it should be (E(z))^2.
If possible, simply use the exposed variables as CLASS does: return self.results.Om_b(zs) + self.results.Om_cdm(zs).
Also, avoid for-loops. These functions are usually broadcastable.

yes you are right about Ez ... it must be that I wanted to put E(z) value here by mistake ... fixed

Also, avoid for-loops. These functions are usually broadcastable.

yes but they are not broadcastable in this version of CLASS

santiagocasas · 2026-05-12T14:50:57Z

+    return {"Linear": mgclass_lin, "NonLinear": mgclass_non}
+
+
+@pytest.mark.parametrize("key", ["Linear"])


change to @pytest.mark.parametrize("key", ["Linear", "NonLinear"]) so you also test the NonLinear adherence to the protocol

fixed and fixed everything related to this change

santiagocasas · 2026-05-12T14:54:41Z

+        )
+
+        return D_z_k
+


for MGCLASSNonLinearPerturbations to adhere to the Perturbations protocol, it needs to have a growth_rate, see here:

cloelib/cloelib/cosmology/class_cosmology.py

Line 578 in c33e526

def growth_rate(self) -> np.ndarray:

Tests were passing because you were not testing protocol adherence in NonLinear. See my comment in tests/test_mgclass_cosmology.py

fixed as I replied to a previous comment

santiagocasas · 2026-05-12T14:58:26Z

+
+    # add neutrinos:
+    mgclass_background_instance.interface_args["MGCLASSparams"]["N_ncdm"] = 1
+    mgclass_background_instance.interface_args["MGCLASSparams"]["m_ncdm"] = 0.2


using this here, while leaving these other parameters as:
mgclass_background_instance.mnu == 0.0 mgclass_background_instance.N_mnu == 0
is inconsistent, in case someone uses another parametrization. Tests might fail or pass for the wrong reason

Maybe but I think I followed the same as it is implemented in the plain CLASS test .. so should we change here and there ?

gcanasherrera · 2026-05-19T07:38:32Z

@zsirap do you have any time scale for addressing these comments?

zsirap · 2026-05-19T08:00:26Z

@zsirap do you have any time scale for addressing these comments?

was in a workshop ... will start addressing them from today ... thanks for reminding me and thanks to @santiagocasas for the extra review

…call non linear, Ez bug etc ...

santiagocasas · 2026-06-02T12:10:40Z

@gcanasherrera @zsirap very busy week unfortunately, but I will look into this in between traveling, hopefully by end of the week!

santiagocasas · 2026-07-23T12:33:05Z

@zsirap thanks for your patience with this and apologies for the late reply.

I have a few comments still for the PR, some of them can be solved now, some of them I have patched myself and some require a larger coordination with the other classy-related EBS codes.

MGCLASSY GCC Installation Problem
mgclassy==2.9.4.5 is distributed as a source package and must compile its C code during installation.
The failure occurs with Conda’s GCC 15.2.0 because GCC 15 defaults to GNU C23. MGCLASSY contains legacy declarations such as:
extern int evolver_ndf15();
int (*generic_evolver)();
In older C standards, an empty parameter list means “parameters unspecified.” In C23, it means “takes no parameters.” MGCLASSY later calls these functions with arguments, while the headers declare them with full prototypes. GCC therefore reports errors such as:
conflicting types for 'evolver_ndf15'
assignment from incompatible function pointer type
too many arguments to function 'generic_evolver'
This is not a missing GSL or Python dependency problem. The same source builds with system GCC 9, or with GCC 15 when compiled using GNU C17:
CFLAGS="-std=gnu17"

santiagocasas · 2026-07-23T12:35:28Z

@zsirap I have patched the .github/workflows/ci.yaml so that:
one can install mgclassy from the pyproject.toml dependencies as in the other code, all within the same environment

@zsirap what is the repo for mgclassy? I would recommend that in the future or current mgclassy, this is fixed by:

1. Replace empty parameter lists with explicit function prototypes.
2. Compile the project with -std=gnu17 or another supported pre-C23 standard.
3. Ensure the Makefile respects externally supplied CC, CFLAGS, and similar variables instead of overriding them.

if you point me to it, I might even make a PR myself there, once I have some time.

santiagocasas · 2026-07-23T12:38:37Z

@zsirap there was an annoying print coming directly from the C implementation of MGCLASS. This is not good for codes that do CI, testing or when we have to run cloelib/like in production environments.
I have also patched it in another commit.
The annoying banner was not from pytest or CLOE. It came from unconditional printf() calls in MGCLASSY’s C source:
if (pba->mg_ansatz != GR) {
printf("Modified Gravity extensions enabled...");
...
}
Since this release has no verbosity parameter for that output, CLOE suppresses it around each MGCLASS compute() call.
The wrapper:

Saves the current stdout file descriptor.
Redirects stdout to /dev/null.
Runs self.results.compute().
Flushes native C streams using:
ctypes.CDLL(None).fflush(None)
Restores the original stdout descriptor.
This is implemented in:
cloelib/cosmology/mgclass_cosmology.py

In the future, @zsirap , please implement a verbosity parameter for MGCLASSY.

santiagocasas

@zsirap please check that my patches work on your end. Install the cloe-org-environment and install mgclassy with the pip install of cloelib.
If this works, make sure the tests pass and that you agree with my patches.
One more thing, tests currently use: mg_E11=0 and mg_E22=0, you might want to test also some MG functionality.
Once this is checked, ready to merge!
Thanks for your work!

santiagocasas

Tests are passing locally and on the CI with the new patches. After the general update of cloe-org-environments, this can be merged,

santiagocasas · 2026-07-31T08:48:49Z

@gcanasherrera I think you can now safely remove your changes suggestions or click on approve. I will merge after the update to the cloe-org-environments.

zsirap · 2026-07-31T10:05:47Z

@zsirap please check that my patches work on your end. Install the cloe-org-environment and install mgclassy with the pip install of cloelib. If this works, make sure the tests pass and that you agree with my patches. One more thing, tests currently use: mg_E11=0 and mg_E22=0, you might want to test also some MG functionality. Once this is checked, ready to merge! Thanks for your work!

Thank you actually @santiagocasas !! great fix and effort to bring this PR to completion ... Wanted but haven't had time (on the move the past week) to thoroughly check or reply to some of your comments or try to implement others you suggested ... hopefully in the coming days (from my side only since summer break is starting so everything off for the next month) ...

But let's not delay further the merge of this PR and merge it as you suggested ... Anything I want to reply, comment or implement, i can open an issue for that later ...

Thank you again :)

santiagocasas · 2026-07-31T11:15:08Z

@gcanasherrera since #566 was merged and passed, I did one final commit here to double check. This should pass and can be merged now.

zsirap and others added 10 commits October 13, 2025 14:30

allowing_MG_cosmology_using_MGCLASS

678d2fc

update mgclass_cosmology to comply with new class_cosmology

c91f538

style fix mgclass_cosmology.py

afc4ca1

style fix mgclass_cosmology.py

8e7e0c7

style fix mgclass_cosmology.py

728412a

style fix mgclass_cosmology

a34ae4f

format mgclass_cosmology

35a9f1a

Merge remote-tracking branch 'origin/main' into 290_mgclass_BZ

6a63194

Update pyproject.toml to add mgclass

210fe8d

Update README.md to add mgclass info

3727285

zsirap requested a review from gcanasherrera October 23, 2025 01:11

gcanasherrera assigned zsirap Oct 23, 2025

gcanasherrera added background-perturbations Tasks related to Background and Perturbations th-1 labels Oct 23, 2025

gcanasherrera added this to the v0.11 milestone Oct 23, 2025

zsirap changed the title ~~290 mgclass bz~~ 290 mgclass integration Oct 23, 2025

zsirap marked this pull request as ready for review October 29, 2025 02:24

zsirap changed the title ~~290 mgclass integration~~ 290 implement mgclass Oct 29, 2025

zsirap changed the title ~~290 implement mgclass~~ gh-implement mgclass Oct 29, 2025

zsirap changed the title ~~gh-implement mgclass~~ gh-209: implement mgclass Oct 29, 2025

gcanasherrera requested changes Oct 29, 2025

View reviewed changes

ziwangithub and others added 3 commits October 30, 2025 03:52

style fix

811b3a6

fix style test_mgclass_cosmology.py

a1d78ff

Merge remote-tracking branch 'origin/main' into 290_mgclass_BZ

4d844b6

gcanasherrera marked this pull request as ready for review May 4, 2026 11:59

gcanasherrera requested a review from santiagocasas May 4, 2026 11:59

gcanasherrera added the priority For PM purposes label May 11, 2026

santiagocasas requested changes May 12, 2026

View reviewed changes

zsirap added 4 commits May 25, 2026 13:21

fix conflicts with main

45b7ca4

update with main + solve review issues: add alpha_s, a fix to how we …

617b22b

…call non linear, Ez bug etc ...

fix pre-commit errors

173b9ec

Merge remote-tracking branch 'origin/main' into 290_mgclass_BZ

f3d3b94

santiagocasas added 2 commits July 23, 2026 14:39

fix mgclassy installation problems with gcc>9, added std=gnu17 flag

49149a6

wrap and patch the printing messages coming from C for mgclassy

054b6fb

santiagocasas reviewed Jul 23, 2026

View reviewed changes

santiagocasas added 2 commits July 31, 2026 04:44

Fix MGClass GCC 15 CI build

5ef86c5

Ensure MGCLASS uses GNU17 compiler wrapper

5177171

santiagocasas approved these changes Jul 31, 2026

View reviewed changes

Use released environment 2026.2

2bc2bff

gcanasherrera merged commit 0373cef into main Jul 31, 2026
8 checks passed

gcanasherrera deleted the 290_mgclass_BZ branch July 31, 2026 13:03

		return {"Linear": mgclass_lin, "NonLinear": mgclass_non}


		@pytest.mark.parametrize("key", ["Linear"])

Uh oh!

Conversation

zsirap commented Oct 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🚀 Pull Request Checklist

✅ Summary

🔄 Changes

🛠 How to Test

📝 Documentation

🏗 Related Issues

📸 Screenshots (if applicable)

📌 Additional Notes

✅ PR Checklist for Developers

✅ PR Checklist for Reviewers

Uh oh!

zsirap commented Oct 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gcanasherrera commented Oct 23, 2025

Uh oh!

zsirap commented Oct 29, 2025

Uh oh!

gcanasherrera commented Oct 29, 2025

Uh oh!

zsirap commented Oct 29, 2025

Uh oh!

gcanasherrera commented Oct 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gcanasherrera left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

gcanasherrera commented May 4, 2026

Uh oh!

gcanasherrera commented May 4, 2026

Uh oh!

santiagocasas commented May 8, 2026

Uh oh!

santiagocasas commented May 8, 2026

Uh oh!

santiagocasas left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gcanasherrera commented May 19, 2026

zsirap commented Oct 23, 2025 •

edited

Loading

zsirap commented Oct 23, 2025 •

edited

Loading

gcanasherrera commented Oct 29, 2025 •

edited

Loading

santiagocasas commented Jul 23, 2026 •

edited

Loading