[cburgard]

This Pull request:

Adds HS3 export of RooParamHistFunc (RooFit core), which so far had no HS3 mapping at all.

A RooParamHistFunc with relative parameters (paramRelative = true, the default, e.g. in rf709_BarlowBeeston) takes in bin i the value p_i times the content w_i of bin i of its histogram. That is what a HistFactory sample does, so it is written like one:

{"name": "func", "type": "product", "factors": ["func_nominal", "func_scale"]}
{"name": "func_nominal", "type": "step", "data": {"axes": [...], "contents": [...]}}
{"name": "func_scale", "type": "step", "variables": ["x"], "parameters": ["func_gamma_bin_0", ...], "axes": [...]}

i.e. a product of the dataset-like step (RooHistFunc) and the parameter-list step (the HistFactory ParamHistFunc). On import, it becomes a RooProduct of these two, with the same value in every bin and the same integral. From the second export on, the JSON is stable.

Not supported, with an explicit error:

• Absolute parameters (paramRelative = false): the analytical integral of RooParamHistFunc is then not consistent with its value (it is Σ p_i · w_i · vol_i instead of Σ p_i · vol_i), see [RF] RooParamHistFunc: analytical integral inconsistent with value for paramRelative = false #24003. A mapping to ParamHistFunc would have a different integral than the original. This can be added once it is clear whether this will be fixed in RooParamHistFunc or the class is going to be deprecated.
• An observable whose binning differs from the binning of the histogram (the ParamHistFunc takes its bins from the observable).

RooHistConstraint, the constraint term that goes with RooParamHistFunc in Barlow-Beeston setups, is built on the fly when the likelihood is created and is not persisted in workspaces, so it is not mapped.

Related: #24001 (per-bin parameters of ParamHistFunc in HS3 bin order; one-dimensional functions like the one written here are not affected).

Changes or fixes:

File	Change
JSONFactories_RooFitCore.cxx	New exporter for RooParamHistFunc. It creates the two step functions as temporary objects and queues them for export (registered with autoExportDependants = false, so that the queued objects are exported).
testRooFitHS3.cxx	New test RooParamHistFuncAsHistFactoryProduct: the written structure, values per bin and integral after the round trip, a stable JSON from the second export on, and the errors for absolute parameters and for a differing binning.

Checklist:

• tested changes locally
• updated the docs (if necessary)

Tests run (Linux, Ubuntu 26.04, GCC 15.2, RelWithDebInfo, -Droofit=ON -Dtesting=ON): testRooFitHS3 105 → 106 tests, testJSONInterface 1, testHistFactory 69, all passing. clang-format clean.

AI assistance: Claude Code (claude-opus-5-5) implemented the changes, the test and this description under review of the author. The commit carries Assisted-by: claude-code:claude-opus-5-5.

🤖 Generated with Claude Code

[guitargeek]

Hi @cburgard! If we deprecate RooParamHistFunc and remove in ROOT 6.46 (see #24003 (comment)), would you still want this PR?

[guitargeek]

The deprecation PR was already merged:

• [RF] Rewrite rf709 with HistFactory classes and deprecate RooParamHistFunc and RooHistConstraint #13199

So RooParamHistFunc is frozen, and we don't have to add HS3 export.