Repository navigation
Conversation
Test Results 23 files 23 suites 3d 22h 31m 33s ⏱️ For more details on these failures, see this check. Results for commit 625d361. |
This was referenced Nov 4, 2026
cburgard
changed the title
cburgard
force-pushed
the
hs3-conformance-fixes
branch
4 times, most recently
from
November 5, 2026 12:55
91bf247 to
74ff7df
Compare
guitargeek
pushed a commit
that referenced
this pull request
Nov 5, 2026
Fix several deviations of RooFit's HS3 layer from the HS3 standard (hep-statistics-serialization-standard master at cb3e9eb). None of the changes depend on pending HS3 PRs; the ones that do follow in #23840. Stricter checks only apply to documents that declare a recent enough hs3_version, so files written by older ROOT versions remain readable. Metadata and names: * hs3_version must be a string. Its MAJOR.MINOR is parsed, and a warning is printed if it can't be parsed or is newer than what RooFit writes. Importers can query it with RooJSONFactoryWSTool::hs3VersionInput(). * Duplicate names of top-level objects are an error from HS3 0.3 on, and only a warning for older documents. * On export, likelihoods are named "<analysis>_likelihood", analyses are named after their ModelConfig, and transformed helper functions that several pdfs share are written only once. Import: * The "default_domain" is optional. Without it, variables get the range of the first domain that lists them. * Parameters that only appear in a parameter point are created. They are constant if "const" is true or if no domain lists them. A string value sets the label of a category. Index categories of combined distributions are now created before the parameter points. * Observables of unbinned data without a domain range take their range from the "min"/"max" of the data axes. Infinite bounds (null) are supported. * "init" of an analysis is restored as the ModelConfig snapshot under the same name. A missing parameter point is reported. * An explicit "extended" flag of mixture_dist is checked against the number of coefficients. * Malformed input is rejected instead of being read out of bounds: covariance and correlation matrices must be square, "errors" must have as many entries as "contents", and "edges" must contain at least two strictly ascending values. * Errors of zero in binned data stay zero. Before, they could be read as squared weight = content. Export: * Axes of unbinned data only carry "name", "min" and "max". * The ModelConfig snapshot is written as "init" of the analysis, which no longer has the side effect of loading the snapshot values. * Ternaries in formulas are written as ifelse(c, a, b), and ifelse() is translated back to a ternary on import. Type keys (the old keys are still imported, with a deprecation warning): * ARGUS_dist -> argus_dist * binsampling -> binsampling_dist. Before, RooFit could not read back its own export. * rate_extended_dist uses the fields "dist" and "rate" instead of "pdf" and "norm". HistFactory: * "LogNormal" is accepted as a constraint type, alongside "Lognormal". Log-normal constraints of gamma parameters are rejected instead of silently becoming Gaussian. * The parameter of a normfactor is named after "parameter". "name" is only the label of the modifier. * normsys/histosys modifiers without constraint are unconstrained from HS3 0.2 on, with one warning per channel. Older documents, and objects imported with importJSONElement(), keep the implicit unit Gaussian. On export, unconstrained modifiers are allowed. A normsys with a function as parameter is refused. * The staterror constraint type is read from the modifier ("constraint", or "constraint_type" as written by ROOT 6.38 and 6.40.00), with the channel-level setting as fallback. It is written in both places, so Gaussian stat errors are no longer lost in a round trip. * A modifier without interpolation and without "default_interpolation" is an error from HS3 0.3 on. Older documents keep code 4, with a deprecation warning. * Fix the export of normsys modifiers recognized from formulas like 1+0.1*alpha: hi/lo were written as 0.1/-0.1 instead of the absolute factors 1.1/0.9, and the constraint was dropped. Other: * Warnings of RooJSONFactoryWSTool are logged on the ObjectHandling topic. Before, they went to the IO topic, which no default stream shows. * Add ModelConfig::GetSnapshotName(). * Update the RooFitHS3 developer documentation. * Make the constraints and interpolation of the rf515_hfJSON tutorial explicit. The fit result is unchanged. Files written with this version use the new type keys and ifelse(), which ROOT 6.40 and older cannot read. Assisted-by: claude-code:claude-opus-5-5
1 of 2 tasks
HS3 PR "json-fixes-0.3" defines `hs3_version` as a `"MAJOR.MINOR"` string and the version of the revised standard as `"0.3"`. RooFit now writes this version. Documents with a newer version are imported with a warning. This commit depends on the HS3 PR and should only be merged together with or after it. Assisted-by: claude-code:claude-opus-5-5
HS3 PR "json-fixes-0.3" renames the analysis keys to `parameters_of_interest` and `domains` (a list of domain names), which is what RooFit already writes, including one entry each for the nuisance parameter, global observable and parameter-of-interest domains. The keys of HS3 v0.2.9, `parameter_of_interest` and `domain` (a single name), are now also read, with a deprecation warning. The plural keys take precedence. Assisted-by: claude-code:claude-opus-5-5
…f writing null RooFit wrote infinite bounds of product-domain axes as JSON `null`, which HS3 doesn't allow. HS3 PR "json-fixes-0.3" makes `min` and `max` of domain axes optional, with an absent bound meaning an unbounded range. Infinite bounds are now omitted on export. Since that PR also requires both bounds for a binning, the default binning is only written for bounded axes. Axes without any bound are still written (with only their `name`), which the HS3 PR keeps valid. RooFit needs them to tell an unbounded RooRealVar from a RooConstVar. On import, an absent bound means -inf/+inf, and `null` is still accepted for files written by older ROOT versions. The RooGaussianConstVarSigmaExport and ProductDomainEntriesExportExplicitBounds tests are updated to expect omitted instead of null bounds. Assisted-by: claude-code:claude-opus-5-5
…ll_dist and crystalball_doublesided_dist HS3 PR root-project#135 defines `crystalball_dist` as the single-sided Crystal Ball (`sigma`, `alpha`, `n`, as RooCBShape) and the new `crystalball_doublesided_dist` as the double-sided one (`alpha_L`, `n_L`, `alpha_R`, `n_R`, and `sigma` or `sigma_L`/`sigma_R`, without a symmetric shorthand). * RooCBShape is exported as `crystalball_dist`, as before. * A single-sided RooCrystalBall is exported as `crystalball_dist`, and read back as a RooCBShape (same function, same sign convention for alpha). Before, it was written as an incomplete `crystalball_doublesided_dist` without `alpha_R`/`n_R`, which could not be read back. * A double-sided RooCrystalBall is exported as `crystalball_doublesided_dist`, with `sigma` if both sides share it, and the tails always as `alpha_L`/`n_L`/`alpha_R`/`n_R`. * `crystalball_doublesided_dist` is imported as RooCrystalBall, also with a single `sigma`, and invalid field combinations are rejected. * `crystalball_dist` with any `_L`/`_R` field is the double-sided form of HS3 versions before 0.3. It is still read as RooCrystalBall (including `alpha`/`n` for both tails with `sigma_L`/`sigma_R`), with a one-time deprecation warning. The `crystalball_doublesided_dist` factory expression and the RooCrystalBall export keys are removed, since the new importer and exporter handle all cases. Assisted-by: claude-code:claude-opus-5-5
…on't contribute to the staterror HS3 PRs root-project#135 and root-project#124 make the "errors" of a HistFactory sample optional: a sample without "errors" contributes no statistical uncertainty to a staterror modifier. RooFit used the content as squared error instead, i.e. assumed Poisson uncertainties. That is kept for documents before HS3 0.3. In HS3 >= 0.3 documents, this only matters if the staterror is constrained through the legacy "constraint" key, since otherwise the gammas are either unconstrained or use the named constraint distributions. Assisted-by: claude-code:claude-opus-5-5
…eference constraints by name HS3 PR root-project#124 replaces the `parameter` and `constraint` (a constraint type like `Gauss`) of HistFactory modifiers with the lists `parameters` and `constraints`. Each entry of `constraints` names a distribution in the top-level `distributions`, and `null` leaves a parameter unconstrained. `shapesys` loses its `data.vals`, since the uncertainties are encoded in the constraint distributions. HS3 PR root-project#129 states that the constraint terms belong to the prediction, which is how RooFit already builds the model. Import: * `parameters` and `constraints` are read for all modifiers, and take precedence over the legacy `parameter`, `constraint`, `constraint_name` and `constraint_type`, which are still read for older files. * For `staterror`, the named constraint distributions are used instead of creating Poisson terms. Without any constraint information, the gammas are unconstrained in documents with `hs3_version` >= 0.3; older documents still get the Poisson constraints that RooFit used to create implicitly. * The `nom_*` global observable is now only created when RooFit builds a constraint itself (legacy `Gauss`). Export: * All modifiers write `parameters` and, if any parameter is constrained, `constraints` with the names of the constraint distributions, which are exported alongside. `parameter`, `constraint` and `shapesys` `data.vals` are no longer written. * The RooPoisson (or RooGaussian) constraint terms of the staterror gammas are exported as named `poisson_dist` (or `gaussian_dist`) and referenced from the staterror modifiers. root-project#124 recommends a `barlow_beeston_lite_poisson_constraint_dist`, which RooFit doesn't support yet; any distribution type is allowed. * The channel-level `staterror.constraint_type` and the `constraint` type on staterror modifiers (written since the previous HistFactory commits) are no longer written, but still read. Tests that checked the exported `constraint` key are updated to the new form, and a new test covers import and export by name. This commit depends on HS3 PRs root-project#124 and root-project#129 and should only be merged together with or after them. Assisted-by: claude-code:claude-opus-5-5
…max on irregular binned axes RooFit wrote only `edges` for binned axes with an irregular binning. HS3 PR root-project#124 requires `name`, `min`, `max` and exactly one of `nbins` or `edges` on the axes of a `histfactory_dist`, while HS3 PR root-project#135 makes `min`/`max` optional next to `edges`, but allows them if they equal the first and last edge. Writing them satisfies both, so binned axes written via `exportAxis()` (binned data, histograms, ParamHistFunc and HistFactory axes) now always carry `min` and `max`. The `axes` of generic formulas are unchanged, because their reader requires either `edges` or `min`/`max`/`nbins`. Assisted-by: claude-code:claude-opus-5-5
… named constraints in rf515_hfJSON Write the HistFactory modifiers of the rf515_hfJSON tutorial in the form of HS3 PR root-project#124: every modifier names its parameters in `parameters`, and its constraint distributions by name in `constraints`. The constraint terms that RooFit used to create implicitly are now explicit: * unit Gaussians (mean 0, sigma 1) for the normsys parameters, as recommended by root-project#124; * Poisson terms for the two staterror gammas, with the tau values (400 and 100) that follow from the sample errors. root-project#124 recommends a `barlow_beeston_lite_poisson_constraint_dist`, which RooFit doesn't support yet, so `poisson_dist` is used. The gammas are declared in a domain and in the default parameter point, so that the functions referring to them can be imported. The document now declares `hs3_version` 0.3. The likelihood is the same as before up to an additive constant of 1.3e-5 (checked at several parameter points), and the fit result agrees within the minimizer tolerance. Assisted-by: claude-code:claude-opus-5-5
… shapefactor as "shapefactor" Following the proposal for HS3 PR root-project#124, HS3 has a single per-bin modifier type, `shapefactor`, whose parameters are constrained by the distributions named in `constraints`, if given, and free otherwise. `shapesys` is a deprecated alias. * Export: HistFactory's ShapeSys and ShapeFactor are both written as `shapefactor`, with the constraint distributions found for the gammas in `constraints` (null for free gammas, omitted if all are free). The type is no longer derived from the name of the ParamHistFunc, so free gammas no longer make the export fail with "cannot find constraint", and constraints of any type are allowed (the unused relative errors, which only supported Gaussian and Poisson constraints, are no longer computed). * Import: `shapesys` is read as `shapefactor`, with a deprecation warning (once per channel) for HS3 >= 0.3. The ParamHistFunc is named `<channel>_<name>_ShapeSys` if constrained and `<channel>_<name>_shapeFactor` if free, as in HistFactory. * Without `constraints`, RooFit only creates constraints from `data.vals` for documents before HS3 0.3 or with a legacy constraint key; otherwise the parameters are free and `data.vals` is ignored with a warning. The HistFactory test of the HS3 import of ShapeFactors now round-trips a free ShapeFactor, which needed constant gammas before. Assisted-by: claude-code:claude-opus-5-5
cburgard
force-pushed
the
hs3-conformance-fixes
branch
from
November 6, 2026 06:16
74ff7df to
6115e08
Compare
cburgard
changed the title
dpiparo
pushed a commit
that referenced
this pull request
Nov 6, 2026
Fix several deviations of RooFit's HS3 layer from the HS3 standard (hep-statistics-serialization-standard master at cb3e9eb). None of the changes depend on pending HS3 PRs; the ones that do follow in #23840. Stricter checks only apply to documents that declare a recent enough hs3_version, so files written by older ROOT versions remain readable. Metadata and names: * hs3_version must be a string. Its MAJOR.MINOR is parsed, and a warning is printed if it can't be parsed or is newer than what RooFit writes. Importers can query it with RooJSONFactoryWSTool::hs3VersionInput(). * Duplicate names of top-level objects are an error from HS3 0.3 on, and only a warning for older documents. * On export, likelihoods are named "<analysis>_likelihood", analyses are named after their ModelConfig, and transformed helper functions that several pdfs share are written only once. Import: * The "default_domain" is optional. Without it, variables get the range of the first domain that lists them. * Parameters that only appear in a parameter point are created. They are constant if "const" is true or if no domain lists them. A string value sets the label of a category. Index categories of combined distributions are now created before the parameter points. * Observables of unbinned data without a domain range take their range from the "min"/"max" of the data axes. Infinite bounds (null) are supported. * "init" of an analysis is restored as the ModelConfig snapshot under the same name. A missing parameter point is reported. * An explicit "extended" flag of mixture_dist is checked against the number of coefficients. * Malformed input is rejected instead of being read out of bounds: covariance and correlation matrices must be square, "errors" must have as many entries as "contents", and "edges" must contain at least two strictly ascending values. * Errors of zero in binned data stay zero. Before, they could be read as squared weight = content. Export: * Axes of unbinned data only carry "name", "min" and "max". * The ModelConfig snapshot is written as "init" of the analysis, which no longer has the side effect of loading the snapshot values. * Ternaries in formulas are written as ifelse(c, a, b), and ifelse() is translated back to a ternary on import. Type keys (the old keys are still imported, with a deprecation warning): * ARGUS_dist -> argus_dist * binsampling -> binsampling_dist. Before, RooFit could not read back its own export. * rate_extended_dist uses the fields "dist" and "rate" instead of "pdf" and "norm". HistFactory: * "LogNormal" is accepted as a constraint type, alongside "Lognormal". Log-normal constraints of gamma parameters are rejected instead of silently becoming Gaussian. * The parameter of a normfactor is named after "parameter". "name" is only the label of the modifier. * normsys/histosys modifiers without constraint are unconstrained from HS3 0.2 on, with one warning per channel. Older documents, and objects imported with importJSONElement(), keep the implicit unit Gaussian. On export, unconstrained modifiers are allowed. A normsys with a function as parameter is refused. * The staterror constraint type is read from the modifier ("constraint", or "constraint_type" as written by ROOT 6.38 and 6.40.00), with the channel-level setting as fallback. It is written in both places, so Gaussian stat errors are no longer lost in a round trip. * A modifier without interpolation and without "default_interpolation" is an error from HS3 0.3 on. Older documents keep code 4, with a deprecation warning. * Fix the export of normsys modifiers recognized from formulas like 1+0.1*alpha: hi/lo were written as 0.1/-0.1 instead of the absolute factors 1.1/0.9, and the constraint was dropped. Other: * Warnings of RooJSONFactoryWSTool are logged on the ObjectHandling topic. Before, they went to the IO topic, which no default stream shows. * Add ModelConfig::GetSnapshotName(). * Update the RooFitHS3 developer documentation. * Make the constraints and interpolation of the rf515_hfJSON tutorial explicit. The fit result is unchanged. Files written with this version use the new type keys and ifelse(), which ROOT 6.40 and older cannot read. Assisted-by: claude-code:claude-opus-5-5 (cherry picked from commit d61fecf)
This was referenced Nov 6, 2026
This branch has not been deployed
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.
Important
#23996 is merged; this PR is based on
master, and all 9 commits are for review. It stays in draft until HS3 PRs #135 and #124 are merged.Rebased onto
masterafter the merge of #23996, including the follow-up commits added there during review (e.g.isHS3VersionBefore()instead ofisLegacyHS3Version(), analyses named after their ModelConfig).This Pull request:
Follows up on #23996 (small HS3 conformance fixes against HS3
masteratcb3e9eb) with the changes that depend on pending HS3 v0.3 PRs: the normative JSON fixes of HS3#135, and HistFactory constraints in the form of HS3 #124 and #129
(constraint distributions referenced by name). Larger structural items are left to the PRs
and issues stacked on this one:
array_dist(HS3 #109), also for the HistFactory constraint terms: #24013;histfactoryfunction of #129: #23985, deferred until HS3 #143 is settled;Background: the HS3–RooFit compatibility audit (October 2026).
Related ROOT PR: #23839 (pinned HS3TestSuite with an
expected-failure list).
The PR has two groups of commits (Group I, the HS3-independent part, is #23996):
[after HS3 json-fixes]) follows HS3 PR#135 "Small normative JSON fixes for v0.3" (branch
json-fixes-0.3).[after HS3 #124/#129]or[after HS3 #124]) follows HS3 PRs #124"Histfactory Specfication Sync" and #129 "Move HistFactory to an appendix and add the
histfactory function" (merged).
If #124 is delayed, Group II can be merged on its own.
Each commit builds on its own and passes
testRooFitHS3,testJSONInterfaceandtestHistFactory.Pre-flight findings
Behaviour with #23996 applied:
exportChannel()inJSONFactories_HistFactory.cxx): RooFitwrites
"constraint": "<name of the constraint distribution>"(e.g."alpha_syst1Constraint") and exports thatdistribution separately; the
staterrormodifier carries"constraint": "Poisson". On import,constraintisfirst resolved as a distribution name and then as a legacy type (
Gauss, ...). Group II keeps this; Group IIIreplaces it by the
parameters/constraintslists of HS3 #124.Domains.cxx,ProductDomain::writeJSON()): RooFit writes infinite bounds as JSONnull. The axis itself is always written; there is no "skip the axis entirely" shortcut. Group II item 3therefore changes
nullto an omitted key.Changes or fixes:
Group II: after HS3 PR #135
Checked against HS3 PR #135 at head
3ec2f79(2026-11-05). Compared with the head0e5d5a3originally checked, #135 moved its changes to binned data axes to HS3 PR #126 and issue#137, split the Crystal Ball into a single-sided and a double-sided type (item 6), and left
shapefactor/shapesysto HS3 PR #124. Re-check this table if the PR changes before merging.hs3_version: "0.3".parameters_of_interestanddomains(with NP/GO/POI domains as entries). Now also import the legacyparameter_of_interestanddomain(a string), with a deprecation warning. Test:AnalysisLegacySingularKeys.min/maxof domain axes instead of writingnull, and write a default binning only for bounded axes (the HS3 PR requires both bounds fornbins). Name-only axes are kept (valid per the HS3 PR); RooFit needs them to tell aRooRealVarfrom aRooConstVar.nullis still read.min/maxnext toedges(Group III item 2), which is valid in any case.errorserrorsoptional: a sample withouterrorscontributes no statistical uncertainty to thestaterror. RooFit used the content as squared error instead (i.e. assumed Poisson uncertainties); that is kept for documents before HS3 0.3. In 0.3 documents this only matters if the staterror is constrained via the legacyconstraintkey. Test:HistFactorySampleWithoutErrors.crystalball_distis the single-sided Crystal Ball (sigma,alpha,n),crystalball_doublesided_distthe double-sided one (alpha_L/n_L/alpha_R/n_R, andsigmaorsigma_L/sigma_R, no symmetric shorthand). RooCBShape and single-sided RooCrystalBall →crystalball_dist(read back as RooCBShape; the single-sided RooCrystalBall used to be written as an incompletecrystalball_doublesided_distthat could not be read back). Double-sided RooCrystalBall →crystalball_doublesided_dist, withsigmaif both sides share it and the tails always written for both sides. Invalid field combinations are rejected.crystalball_distwith_L/_Rfields (the double-sided form before HS3 0.3) is still read as RooCrystalBall, with a deprecation warning. Test:CrystalBallTypes.Group III: after HS3 PRs #124 and #129
Checked against #124 at head
bb3b367(2026-11-06; originallybc5f670, since then #124 merged HS3master, added thecustommodifier, made the modifiernameoptional and deprecatedshapesys) and #129 at heade76fc4a.JSONFactories_HistFactory.cxxparametersandconstraints(names of distributions,null= unconstrained) for all modifiers; they take precedence over the legacyparameter,constraint,constraint_nameandconstraint_type, which are still read.staterroruses the named constraints instead of creating Poisson terms; without any constraint information its gammas are unconstrained forhs3_version≥ 0.3 (older documents get the implicit Poisson terms as before). Export writesparameters/constraintsand exports the constraint distributions; the staterror RooPoisson terms become namedpoisson_dist.parameter,constraint,shapesysdata.valsand the channel-levelstaterrorare no longer written.HistFactoryConstraintsByName, updatedHistFactoryStatErrorConstraint,HistFactoryConstraintKeyMigration,HistFactoryLegacyConstraintTypeInConstraintKey,HistFactoryModifierWithoutConstraintIsUnconstrainedRooJSONFactoryWSTool.cxxexportAxis()always carrymin/max, also withedges. #124 requires them onhistfactory_distaxes; #135 allows them if they equal the first and last edge.IrregularBinnedAxesHaveRangetutorials/roofit/roofit/rf515_hfJSON.jsonparameters/constraints; the formerly implicit constraints are explicitgaussian_dist(mean 0, sigma 1) andpoisson_dist(staterror, tau 400/100) entries;hs3_version0.3.shapefactorJSONFactories_HistFactory.cxxshapesysa deprecated alias ofshapefactor, whose parameters are constrained by the distributions inconstraints, if given, and free otherwise. HistFactory ShapeSys and ShapeFactor are both written asshapefactor(constraintswithnullfor free gammas, omitted if all are free); the type is no longer derived from the name of theParamHistFunc, so free gammas no longer make the export fail with "cannot find constraint", and constraints of any type are allowed.shapesysis read asshapefactor, with a deprecation warning (once per channel) for HS3 ≥ 0.3. TheParamHistFuncis named<channel>_<name>_ShapeSysif constrained and<channel>_<name>_shapeFactorif free, as in HistFactory. Withoutconstraints, constraints fromdata.valsare only created for documents before HS3 0.3 or with a legacy constraint key; otherwisedata.valsis ignored with a warning.HistFactoryShapeFactor, updated HistFactory test of the ShapeFactor importNotes on Group III:
histfactoryfunction (#129): not implemented. It is a new primitive; it's left for a follow-up,#23985. #129's statement that the constraint terms belong to the prediction is how RooFit already builds
HistFactory models.
barlow_beeston_lite_poisson_constraint_dist, which RooFit can'tread or write. Since #124 allows any distribution type, the existing RooPoisson terms are exported as
poisson_dist.barlow_beeston_lite_poisson_constraint_distis going to be deprecated in favour ofarray_dist(HS3 #109), which is then to be recommended for the
staterrorinstead; RooFit'sarray_distsupport, includingthe grouping of HistFactory constraints, is #24013. No importer for the BB-lite distribution is planned.
name: optional in #124 (required forcustommodifiers, where it names the function). RooFitwrites and reads it as before.
min/maxonhistfactory_distaxes. Do not allow short reads. #135 no longer touches binned axes, so the two don't conflict any more.is kept there, so that [RF][HS3] Bring the HS3 import and export closer to the HS3 standard #23996 is correct against the current standard and can be merged on its own; Group III
replaces its export.
Behaviour changes users may notice
In addition to those of #23996:
hs3_versionis"0.3", and infinite domain bounds are omitted instead ofnull.hs3_version"0.3", the duplicate-name check introduced in [RF][HS3] Bring the HS3 import and export closer to the HS3 standard #23996 would make RooFit unable to read its own output: its export still produces clashes between different kinds of objects (a named range, a snapshot or a data set named like a pdf), which HS3 does not allow. From 0.3 on, a name used twice within one section, or by a distribution and a function, is still an error; other clashes only warn, as before 0.3 (testNameClashesInOlderDocumentsextended accordingly). Avoiding these clashes on export is left for a follow-up: [RF][HS3] Avoid name clashes between different kinds of objects in the HS3 export #24020.crystalball_doublesided_dist(as onmaster, but nowalso with a single
sigma), a single-sided one ascrystalball_dist.crystalball_distwith_L/_Rfieldstriggers a deprecation warning.
hs3_version≥ 0.3, a HistFactory sample withouterrorsno longer contributesPoisson uncertainties to the
staterror.parameters/constraintsand named constraintdistributions (including
poisson_distterms forstaterror);parameter,constraint,shapesys(now
shapefactor), itsdata.valsand the channel-levelstaterrorare gone. Free ShapeFactor gammas can beexported. In documents with
hs3_version≥ 0.3, astaterrorwithout
constraintsis unconstrained. Binned axes withedgesalso carrymin/max.rf515_hfJSON: with explicit, named constraints its likelihood equals the one ofmasterup to an additiveconstant of 1.3e-5, checked at several parameter points; the fit agrees within the minimizer tolerance
(
mu1.10955 vs. 1.10947).Changed tests and reference data
No reference JSON files exist for these tests; expectations are inline. The following existing expectations or
fixtures were changed because of intended behaviour changes:
RooGaussianConstVarSigmaExport,ProductDomainEntriesExportExplicitBounds(Group II.3): expect omittedinstead of
nullinfinite bounds. The second test now also checks thatnullis still read.tutorials/roofit/roofit/rf515_hfJSON.json(Group III): rewritten with named constraint distributions.HistFactoryConstraintKeyMigration,HistFactoryLegacyConstraintTypeInConstraintKey,HistFactoryModifierWithoutConstraintIsUnconstrained,HistFactoryStatErrorConstraint(Group III): expect theexported
constraintslist instead ofconstraint.HS3ImportShapeFactorModifierintestHistFactory(Group III.4): round-trips a free ShapeFactor, written asshapefactorwithout constraints; it needed constant gammas before.
HistFactory round trip
roofit/histfactory/test/ref_6.16_example_UsingC_combined_meas_model.root(workspacecombined), exported with#23996 and with this branch. Every changed key:
metadata.hs3_version:"0.2"→"0.3"parameter→parameters(list) andconstraint→constraints(list ofnames; for
staterrorthe names of the gamma constraints instead of"Poisson"); the channel-levelstaterror.constraint_typeis gone; new distributionsgamma_stat_channel1_bin_{0,1}_constraint(
poisson_dist), functionsgamma_stat_channel1_bin_{0,1}_poisMean(product) and their constant parametersgamma_stat_channel1_bin_{0,1}_tauandnom_gamma_stat_channel1_bin_{0,1}.Checklist:
Tests run (Linux, Ubuntu 26.04, GCC 15.2, RelWithDebInfo,
-Droofit=ON -Dtesting=ON):ctest -R 'RooFitHS3|JSONInterface': 105 + 1 tests pass onmaster; 111 + 1 with this PR (6 new). The two testbinaries and
testHistFactory(69 tests) were also built and run for every single commit.Not run: the PyROOT tests (
bindings/pyroot/pythonizations/test/roofit.py, which uses HistFactory JSON viaimportJSONElement), therf515_hfJSON.pytutorial itself (its model was fitted from C++ instead), and theoptional HS3 test suite (
test_roofit_hs3testsuite). My build haspyroot=OFF.AI assistance: Claude Code (claude-opus-5-5) implemented the changes, the tests and this description from a
written plan, under review of the author. The commits carry
Assisted-by: claude-code:claude-opus-5-5.🤖 Generated with Claude Code