Extend the 2.0 format to cover the full WellEventTimeline API demonstrated
by well_event_schedule.py.
Add built-in TUBING, VALVE and STATE event types mapping to
add_tubing_event, add_valve_event (including the AICD parameters) and
add_state_event, validated with the same required/unknown-attribute
handling as PERFORATION.
Pass any other event type inside a WELL block through generically as an
Eclipse well keyword via add_well_keyword_event with the well name
injected, so WRFTPLT, WCONPROD and the rest work without enumeration.
Event types that closely resemble a misspelled built-in are handled by
the on_unknown_event policy with a did-you-mean hint instead of being
passed through.
Add a SCHEDULE block: a bare SCHEDULE line opens a block of well-less
keyword events (RPTRST, GRUPTREE, TUNING, ...) applied through
add_keyword_event; completion events in a SCHEDULE block are rejected.
Parsed schedule events live in OrionDocument.schedule_events.
Accept ISO datetimes in event dates (@2024-05-15T14:45:30.500); the
time-of-day survives into the generated DATES keyword's TIME field with
millisecond precision. Day-offset arithmetic applies to datetimes too.
Ship well_event_schedule_orion.py, an ORIONEVENTS counterpart to
well_event_schedule.py, and extend the test suite from 55 to 70 tests
including an integration test that applies every event kind and verifies
the generated schedule text.
Replace the 1.0 grammar with a version 2.0 designed to be easier to parse
and easier to write. Variables are now typed: DATE, DURATION (whole days)
and WELL alias declarations replace the untyped SET, and using a variable
of the wrong type produces an error citing both the use and declaration
site. Well blocks open with an explicit WELL keyword (alias or
double-quoted literal name), quoting is double quotes throughout, and
date arithmetic accepts signed chains of integers and DURATION variables
such as @START + RAMP - 2. ORIONEVENTS 1.x headers are rejected with a
migration message.
FILTER and PERFID are no longer parse errors; any KEY=VALUE attribute
parses and the applier warns and ignores the not-yet-supported keys
(FILTER, PERFID, DSHIFT) while still applying the rest of the event.
The parser now recovers per line and reports every error in one pass via
OrionParseError.errors, unknown variables, event types and keywords get
difflib-based did-you-mean hints, and a stdlib-only validator CLI runs as
python3 -m rips.orion_events <file>. The eager orion_events import in
rips/__init__.py is replaced by a PEP 562 lazy import so the -m entry
point does not trigger runpy's double-import warning.
Update the shipped sample file and example script to the 2.0 syntax and
extend the test suite from 27 to 55 tests covering typed declarations,
offset arithmetic, alias resolution, multi-error diagnostics and the CLI.
Add a pure-Python parser (rips.orion_events) that reads an ORIONEVENTS
text file into a structured representation, plus an applier that drives
the existing WellEventTimeline API to create perforation, WCONHIST and
WELTARG events.
The parser is split into a stdlib-only Layer A (parse_orion_events ->
OrionDocument) that is unit-testable without a running ResInsight, and a
Layer B applier (apply_orion_document) that maps the document onto
add_perf_event/add_well_keyword_event and resolves well names via the
project. RADIUS is mapped to diameter, dated SET variables support '+ N'
day arithmetic, and FILTER/PERFID are reserved but currently raise a
located error while DSHIFT is ignored with a warning.
Ship a documented grammar in the module docstring, a clean sample input
(example_input_files/well_events.orion), a runnable example
(import_orion_events.py), and tests covering parsing, mapping and an
end-to-end integration test against a loaded case.
* #14411 Enforce name uniqueness for surfaces, polygons and folders
Names must be unique among siblings sharing the same parent folder. Items in different folders may keep identical names, and comparison is case sensitive.
Siblings are the objects held by the same caf::PdmChildArrayField. Items and folders live in two distinct child arrays in caf::PdmNestedCollection, so a folder and an item may share a name under the same parent.
The Python API gets an OnNameConflict flag on AddFolder, CreatePolygon, ImportSurface and NewRegularSurface, supporting FAIL (default), AUTO_RENAME and OVERWRITE. NewSurface is left out, as grid case surfaces derive their tree label from the case and K index and carry no name of their own.
Compute the depth related geometry properties (DEPTH, DX, DY, DZ, TOPS, BOTTOM) in RimCornerPointCase::createFromCoordinatesArray and replaceGridFromCoordinatesArray, matching the behavior when importing a grid from file. The computation is guarded by the same autoComputeDepthRelatedProperties preference used by the file importers.
Address review feedback on #14199: deleting and recreating the legend left
views bound to the old legend pointing at a deleted object. Add
RimColorLegendCollection::updateColorLegend and a matching UpdateColorLegend
scriptable method that mutate the existing custom legend (name and items) so
referring views keep their binding and are notified once. A new legend is only
created when none is registered for the result or the registered legend is a
standard legend.
set_discrete_property_category_names now performs a single atomic call instead
of delete + create + per-item appends. The empty-dict removal path still uses
DeleteColorLegend.
RimColorLegendCollection::deleteColorLegend erased the default-legend map
entry before looking the legend up, so findDefaultLegendForResult always
returned null and the old legend was never removed from m_customColorLegends.
Each set_discrete_property_category_names call for the same property left an
orphaned legend behind and appended a new one.
Look up the legend before erasing the map key, and guard against deleting a
standard legend that may have been registered as a default.
The return value of copyPdmObjectFromRipsToCaf was discarded when copying method parameters in CallPdmObjectMethod. Check the std::expected result and return a gRPC INVALID_ARGUMENT status (with telemetry and logging) when parameter validation fails, instead of silently proceeding to execute the method.
Make RiaGrpcCommandService::assignPdmObjectValues return std::expected<void, QString> so the result of copyPdmObjectFromRipsToCaf is no longer discarded. Validation failures are now propagated up and returned to the client as a gRPC INVALID_ARGUMENT status instead of being silently ignored. This also resolves the C4834 [[nodiscard]] build warning.
Carries the "Fault Distance" rename through to the C++ class names, file
names, PDM/scriptable keywords and the Python API (add_fault_distance,
class FaultDistance). The RicNewFaultDistanceResultFeature command class
keeps its name.
The Completion Type result was registered only as a dynamic result, which
requires simulation time steps. Loading only grid geometry (e.g. a ROFF grid)
therefore made the result unavailable, since there are no time steps to compute
it for.
Register and compute a STATIC_NATIVE variant of Completion Type in addition to
the existing DYNAMIC_NATIVE one, computed as a single time-step-independent
frame. The shared per-frame computation is factored out of
computeCompletionTypeForTimeStep so both variants reuse it. The virtual
perforation transmissibilities already handle the no-restart case by clamping to
a single frame, so the static result can be computed at calc time step 0.
The legend/category predicates and the completion-type result clearing are
broadened to recognise the static variant so it renders with the same discrete
named-category legend as the dynamic one.
The Event Timeline stores event timestamps with full time-of-day precision, but the exported SCHEDULE deck only emitted DAY/MONTH/YEAR in the DATES keyword, dropping any sub-day precision the user had set.
Emit the optional TIME field (HH:MM:SS[.SSS]) in datesKeyword() whenever the timestamp is non-midnight, formatting with millisecond precision when present. Date-only events keep their previous DAY/MONTH/YEAR-only output. The first-date comment in the schedule generator mirrors the same time so it stays consistent with the keyword.
A grid created via the create_corner_point_grid API populated the internal
active cell info but never created an ACTNUM result property, so the grid did
not expose ACTNUM the way a grid loaded from an Eclipse file does. Since the
corner point case has no reader interface, the data cannot be loaded lazily and
is now materialized as a STATIC_NATIVE result sized per active cell, mirroring
native Eclipse ACTNUM.
Plain pip has no time-based cutoff, so route the workflow's package
installs through uv and set UV_EXCLUDE_NEWER to refuse releases
published within the last week as supply-chain hardening. Mirror the
policy in the rips pyproject for local uv installs.
Add an align_columns option to schedule generation. When enabled, each
keyword is serialized with a '--'-prefixed column-header comment and data
rows right-aligned into fixed-width columns, instead of the compact default
form. The default output is unchanged.
A new RimKeywordFactory::deckKeywordToAlignedString introspects the OPM deck
model and renders each item in its own column (consecutive defaults are not
collapsed into 'N*'). The flag is threaded through RicScheduleDataGenerator
and exposed on the GenerateSchedule scriptable method as align_columns in the
Python generate_schedule_text wrapper.
Reparent RimFaultDistanceResultCollection so it is owned directly by
RimEclipseView instead of RimFaultInViewCollection, and show it as a
top-level node next to the Faults node in the project tree.
RimFaultDistanceResult::calculateValueOptions now reaches the fault list
through the view (the fault collection is a sibling rather than an
ancestor after the move). RicNewFaultDistanceResultFeature resolves the
view from the selection and takes both the distance collection and the
all-faults fallback from it.
The scriptable API moves with the collection: add_fault_distance_result
is now exposed on RimEclipseView (view.add_fault_distance_result(...))
via the new RimcEclipseView, replacing RimcFaultInViewCollection.
Expose the new RimFaultDistanceResult facility to Python by marking the
relevant PDM classes and fields scriptable. RimEclipseView's
m_faultCollection, RimFaultInViewCollection, RimFaultInView,
RimFaultDistanceResult, and RimFaultDistanceResultCollection are now
scriptable, so faults can be enumerated from Python via
view.fault_collection().faults().
A new Rimc method add_fault_distance_result on RimFaultInViewCollection
takes a name and a list of fault objects, creates the result
(auto-numbered FAULTDIST<n> when the name is empty), and returns the
new RimFaultDistanceResult. An empty fault list falls back to every
fault in the view, matching the static-native FAULTDIST behaviour.
Add a Python test exercising both a subset call and a parity check
that asserts the all-faults result matches the existing static-native
FAULTDIST cell-for-cell.
Some commercial simulators reject a DATES keyword whose value equals the
simulation start date. The earliest event date is, by construction, that start
date, so generate_schedule_text now emits it as a comment line (e.g.
"-- Date: 1 JAN 2024") instead of a DATES keyword by default.
The new first_date_as_comment parameter (default True) threads from the Python
wrapper through the GenerateSchedule scriptable method into
RicScheduleDataGenerator. Only the first/earliest date is affected; later dates
are always emitted as DATES keywords. Pass first_date_as_comment=False to
restore the previous behavior.
The well list driving schedule generation originates from a
std::set<RimWellPath*>, so it was ordered by pointer address and the
per-well keyword records (WELSPECS, COMPDAT, COMPLUMP, MSW keywords, well
control) came out in a non-deterministic order that only looked sorted by
coincidence.
Sort the well list once by the deck export name in generateSchedule, so
every per-well keyword is emitted in a stable, well-name-sorted order,
matching the convention already used by the file-based completion export.
Schedule-level keywords (GRUPTREE, RPTRST, TUNING) are left untouched.
buildKeyword placed every user-supplied item into a single DeckRecord,
so multi-record keywords such as TUNING (3 records, each terminated by
its own '/') were emitted with only one record/slash and rejected by the
simulator.
Distribute items across the keyword's records using the OPM ParserKeyword
schema: each item is matched to the record that defines it. Every record
is emitted (empty records are written as a bare '/') so the schema's
record structure is preserved, and items unknown to the schema are now
reported via a warning. Single-record keywords keep their previous
behaviour, including appending non-schema items after the canonical
block.
Add a completion_number parameter to the add_perf_event Python API
(AddPerfEvent scriptable method) and a matching field on
RimWellEventPerf. When a timeline perforation event is materialised into
a RimPerforationInterval, a non-zero completion number is copied across,
so it flows through the existing completion-data path onto
RigCompletionData.
The schedule generator now emits COMPLUMP alongside COMPDAT: COMPDAT
generation is folded into generateCompletionsForWell, which computes the
completion data once and builds both keywords from it via the existing
RimKeywordFactory::complumpKeyword (added in #13273). COMPLUMP only
produces records for perforations that have a completion number and is
merged across wells like COMPDAT.
WELSEGS and COMPSEGS carry a per-well header record followed by that
well's segment records, so segments from different wells cannot share a
single header. Commit 6ef024c merged records of the same keyword across
wells, which is invalid for these two keywords.
Accumulate WELSEGS and COMPSEGS as separate per-well blocks in a
dedicated container and serialise each block individually, while the
remaining keywords (WELSPECS, COMPDAT, WSEGVALV, WSEGAICD, well-control
and schedule keywords) continue to merge. Emission preserves the
existing priority ordering.
Replace the include_welsegs / include_compsegs boolean flags on the
schedule-generation API with a list of wells (export_msw_for_wells) that
gates the multi-segment-well keywords. WSEGVALV and WSEGAICD are now gated
the same way as WELSEGS and COMPSEGS: all four are emitted only for wells
present in the list. An empty list (the default) suppresses MSW export for
all wells, while COMPDAT and well-control keywords are unaffected.
C++: RicScheduleDataGenerator takes a std::set<const RimWellPath*> mswWells
and generateMswForWell early-returns for wells not in the set.
RimcWellEventTimeline_generateSchedule replaces the two bool fields with a
scriptable PdmPtrArrayField<RimWellPath*>, resolved via
ptrReferencedObjectsByType().
Python: generate_schedule_text / generate_schedule expose
export_msw_for_wells (List[WellPath]) instead of the boolean flags. Updates
the well_event_schedule example and adds regression tests covering the
empty-list default, per-well selection, and WSEGVALV/WSEGAICD gating.
Replace the integer eclipse_case_id parameter on
RimcWellEventTimeline_generateSchedule with a
caf::PdmPtrField<RimEclipseCase*>. The execute() method no longer needs
to walk the project tree and look up the case by ID — the gRPC layer
resolves the object pointer for us.
generate_schedule_text passes the Eclipse case object through to the
underlying generate_schedule call, matching how add_perf_event /
add_well_keyword_event already take a WellPath object rather than an
ID.
Add include_welsegs and include_compsegs flags (default True, preserving
current behaviour) to the schedule-generation API. Plumbed through
RimcWellEventTimeline_generateSchedule -> RicScheduleDataGenerator ->
generateMswForWell, gating only the WELSEGS / COMPSEGS merge calls.
WSEGVALV and WSEGAICD are unaffected.
Python: generate_schedule_text gains include_welsegs / include_compsegs
keyword arguments that forward to generate_schedule. Callers wanting
control-only / well-header-only schedule updates can now suppress the
verbose MSW table blocks.
The schedule generator previously concatenated full keyword blocks
(`KEYWORD ... /\n/\n`) — one per well — producing verbose output where
each WELSPECS / COMPDAT / WCONHIST / WELSEGS / COMPSEGS / etc. block
appeared once per contributing well. Eclipse keywords are designed to
hold multiple records under a single header; the per-well repetition is
syntactically valid but bloats the file and obscures intent.
Switch `RicScheduleDataGenerator` to accumulate
`std::map<QString, Opm::DeckKeyword>` instead of raw concatenated text.
Each per-well helper now returns an `Opm::DeckKeyword`, and a new
`mergeKeyword` primitive appends its records into the accumulator entry,
creating the entry from the helper's keyword on first encounter. At
emission time each accumulated keyword is serialised once.
To support this, `RifEventKeywordFormatter` is split: each `format*`
QString function now wraps a corresponding `build*` function that
returns `std::optional<Opm::DeckKeyword>`. `RimWellEventKeyword` and
`RimKeywordEvent` expose a parallel `generateDeckKeyword` accessor.
The text-scraping `extractKeywordName` helper becomes dead code and is
removed.
Schedule-level keywords whose OPM ParserRecord is a single ALL-sized item
(RPTRST, RPTSCHED, ...) are not positional — they take a free-form list
of mnemonic tokens like `BASIC=2 DEN ROCKC NORST=1`. Previously each
`keyword_data` entry became its own DeckItem, and OPM emitted them
either as truncated positional values or as a quoted, space-collapsed
single string.
Detect that schema shape in `RifEventKeywordFormatter::formatKeyword`
and pack the user's entries into one RawString DeckItem of mnemonic
tokens: `KEY` for FLAG items, `KEY=VALUE` otherwise.
To distinguish flag mnemonics (`DEN`) from `KEY=1`-style entries, add a
new `FLAG` value to `RimWellEventKeywordItem::ItemType` plumbed end to
end (Python -> GRPC -> Rimc -> storage -> formatter). The Python API
now maps `bool True` to FLAG and silently drops `bool False`.
For positional keywords (WCONHIST etc.) FLAG items have no value to
emit and are rendered as default markers ("1*").
A `rawStringItem` helper is added to `RifOpmDeckTools` so the
mnemonic-list path can produce DeckItems whose OPM output is not
quoted.
`add_well_keyword_event` previously emitted items in `keyword_data` dict
insertion order. Eclipse keywords such as WCONHIST are positional, so a
non-canonical order silently corrupted the simulator input.
`RifEventKeywordFormatter::formatKeyword` now walks the OPM
`ParserRecord` schema and emits items in canonical order, inserting
`1*` defaults for intermediate gaps and dropping trailing unspecified
items. Items that are not part of the keyword schema (e.g. mnemonic-list
keywords like RPTRST/RPTSCHED) are appended after the canonical block in
caller-supplied order to preserve existing behaviour.
Fixes#14065.
Consolidate the hand-written PropertyType / NNCPropertyType enums in
enums.py into the auto-generated StrEnum classes that already exist in
generated_classes.py. enums.py and all its imports are removed.
The auto-generated class for caf::AppEnum<RiaDefines::ResultCatType>
was named "ResultType" because the existing scriptable m_resultType
fields used keyword "ResultType". Switch those fields to
CAF_PDM_InitScriptableFieldWithScriptKeywordNoDefault, keeping the XML
keyword "ResultType" (so project files still load) but exposing the
script keyword as "PropertyType". The generator now emits a single
PropertyType StrEnum that:
- matches the proto PropertyType for setters (FLOAT/INTEGER aside, the
members callers actually use are identical),
- gains ALLAN_DIAGRAMS that the C++ side already knew about,
- loses the proto-only REMOVED/UNDEFINED sentinels (unused in any
Python caller).
rips.NNCPropertyType was exported but unreferenced anywhere — dropped.
Updated Python callers:
- rips.ResultType.X -> rips.PropertyType.X across examples and tests.
- result_type= kwarg / .result_type attribute -> property_type /
.property_type (the script keyword change renamed those everywhere
the affected PdmObject fields are accessed from Python).
- view.apply_cell_result's hand-written wrapper now takes property_type
instead of result_type and forwards via cell_result.property_type.
All 271 Python tests pass, ruff and clang-format clean.
Show how property_data_type, discrete_property_category_names, and
discrete_property_category_colors round-trip the values just written by
set_active_cell_property + set_discrete_property_category_names.
Replace the hand-written PropertyDataType wrapper with first-class enum
return support in the code generator. Scriptable methods that derive
from the new caf::PdmEnumObjectMethod<T> template return a typed enum
value across the gRPC boundary; the Python generator emits a StrEnum
class with the texts pulled from the underlying caf::AppEnum<T> and
wraps the call site so callers receive `EnumClass(value)` directly.
caf::PdmDataContainerEnum is the small PdmObject the framework uses to
carry the enum text on the wire (single scriptable QString field).
RimcEclipseCase_propertyDataType becomes declarative: implement
executeEnum() returning the typed RiaDefines::ResultDataType and pick a
returnEnumScriptName() — the base handles serialization and reports
the four AppEnum members (UNKNOWN, FLOAT, DOUBLE, INTEGER) to the
generator.
Drop the hand-written PropertyDataType in enums.py — the generated one
in resinsight_classes is wire-equivalent and now the single source of
truth.
The auto-generated stub returns a DataContainerString, forcing call
sites to use .values[0]. Add a hand-written wrapper on Reservoir
(which both EclipseCase and the other reservoir types inherit from)
that unwraps the single-value container and returns the typed
PropertyDataType so a caller can simply write
case.property_data_type(...) == rips.PropertyDataType.INTEGER
Update the round-trip tests to use the cleaner form.
Add the inverse APIs to the existing discrete-property setters:
- case.property_data_type(property_type, property_name) returns "FLOAT" or
"INTEGER", matching the data_type flag passed to set_active_cell_property
/ set_grid_property.
- case.discrete_property_category_names(property_name) and
case.discrete_property_category_colors(property_name) return the
{int: str} mappings previously bound via
set_discrete_property_category_names, or empty dicts when nothing is bound.
Implemented as PdmObjectMethods on RimEclipseCase and
RimColorLegendCollection so the bindings surface automatically through
generated_classes.py.
Fill the async-variant gaps in the Grid class. Both sync forms are
already exercised by test_10k; the new tests iterate the streaming
variants, accumulate the chunks, and assert length and a spot-check
cell against the sync result.
Fixes#4469.
Fill the async-variant gaps in Properties / NNCProperties coverage that
already have sync tests: selected_cells_async, selected_cell_property_async,
nnc_connections_static_values_async, nnc_connections_generated_values_async.
Each test accumulates the streamed chunks and asserts equality with the
corresponding synchronous call.
Fixes#4465.
Cover four more Project APIs. test_plot_lookup mirrors the existing
views/view test using the WellLogPlot collection in 10KWithWellLog.rsp.
test_export_well_paths exercises the WELLPATHS export-folder path.
test_scale_fracture_template_and_set_containment opens
small-completion-export-fractures.rsp, asserts the scale factors round
trip via PDM descendants, then verifies set_fracture_containment by
saving the project and inspecting the .rsp XML (RimFractureContainment
fields are not scriptable). test_summary_cases uses
SIMPLE_SUMMARY2.SMSPEC and asserts summary_cases() reflects
import_summary_case.
The Python wrapper Project.scale_fracture_template passes dFactor and
conductivity, but the proto message only declared id, halfLength, and
height. Constructing the protobuf message raised ValueError, making the
API uninvokable. The C++ Ricf handler already declares matching
scriptable fields, so the additions wire up automatically via the
reflection-based proto-to-PDM assignment in RiaGrpcCommandService.
Verify import_formation_names via the FORMATION_NAMES branch of
available_properties on the target case. Verify import_well_log_files
via the well_log_folder argument, writing a minimal synthetic LAS file
into a tempdir and checking the returned well-path names contain the
LAS WELL header value.
Fixes#4468.
Cover the easily-reachable gaps in Project API test coverage: save() round
trip, views()/view() lookup, well_path_by_name() hit and miss, the
grid_case_groups()/grid_case_group(id) lookup pair, and the async
key_values iterator.
Cover the remaining gaps in the Case Properties + NNC test batch:
grid_property_async, nnc_connections_async, and available_properties for
DYNAMIC_NATIVE and GENERATED branches.
Closes#4467.
Remove the committed rips/_version.py and add it to .gitignore. The file
is generated by CMake from rips/_version.py.cmake at configure time, so
checking it in caused the source tree to carry stale/duplicated state.
Addresses review feedback on PR #13847.
Replace legacy setup.py packaging with modern pyproject.toml (PEP 517/518/621).
Version is dynamically resolved via setuptools attr: directive from rips/_version.py,
which CMake generates.
Consolidate requirements.txt, dev-requirements.txt, and build-requirements.txt
into pyproject.toml dependencies and [dev] optional-dependencies.
Consolidate rips/mypy.ini into [tool.mypy] section of pyproject.toml.
Update GitHub Actions workflow and CMakeLists.txt accordingly.
Adds case.filtered_cells(filter, time_step, grid_index) which applies any
RimCellFilter to an Eclipse case and returns a per-cell 0/1 mask. The mask
length and ordering match case.grid_property(..., grid_index), so callers
can combine the two vectors element-wise.
Implemented as a scriptable Rimc method (PdmVoidObjectMethod) on
RimEclipseCase that writes the mask via the existing key-value store, with
a thin Python wrapper mirroring the visible_cells() pattern. No proto or
gRPC service changes.
Lets a script attach a cell filter to a RimPerforationInterval as a
typed reference so subsequent work can gate completion-export
connections by the filter's cell mask. Pure metadata in this commit:
no consumer of RimPerforationInterval is modified, the field exists
purely as an API surface.
Add a non-scriptable caf::PdmPtrField<RimCellFilter*> m_cellFilter to
RimPerforationInterval with public cellFilter()/setCellFilter()
accessors and a "Cell Filter" group in the property panel between the
existing geometry/flow group and the Date Settings group. The field is
deliberately non-scriptable: a scriptable PdmPtrField round-trips in
Python as an opaque "ClassName:0xADDR" string rather than a resolved
object, defeating the point. Two CAF object methods on
RimcPerforationInterval expose the relationship:
- AddFilter (perf.add_filter(filter)) sets the pointer; replaces any
existing filter (single filter per perforation).
- cell_filter (perf.cell_filter()) resolves to the referenced
RimCellFilter object on read, mirroring the valve.template() pattern.
Composition stays in RimCombinedFilter: a script that wants AND/OR
across multiple filters attaches a combined filter as the single
filter on the perforation.
The Python example in PythonExamples builds a case-level AND combined
filter (PERMX 100..20000 AND K=5..10), creates a modeled well path
with two perforation intervals, and shares the same filter between
them. test_perforation_filter.py covers single-set, replace-on-second-
add, and shared-pointer round-trips.
Kristian Bendiksen
Magne Sjaastad
Jon Jenssen
Håkon Hægland