Release/2.5

[olivhoenen]

Notes for release 2.5.0

New features

• Add --no-provenance CLI option to remove information from the plots (Generalize no-provenance option to all plotting script #31 )
• Add terminal plotting in idsprint to allow simple profile view when there is no X (Feature/show shape and terminal plot in idsprint #28 )

Improvements

• More flexible handling of matplotlib backends (Feature/allow custom matplotlib backend #33 Let matplotlib choose its default backend #35 )
• Add separatrix and boundary in plotequilibrium (Feature/add separatrix and magnetic axes in plotequilibrium #20 )
• plotequilibrium option --rho is renamed as --phi, with better description
• Improve failsafe in plotequicomp when some data is missing (Improve failsafe for plotequicomp when data is missing #18 )

Bug fixes

• Fix idslist when using UDA URIs (Fix/incorrect handling of error/warning when idslist is applied to a uda uri #38 )
• Fix missing labels with idsprint -p on several quantities (idsprint -p needs labels for all signals #29 )
• Fix plot of edge data (use polycollection from matplotlib and use cells instead of nodes #22 )
• Fix core_sources calculations and plots (fix: correct electron power calculation and improve plot visibility #21 )
• Fix 2D grid orientation consistency (fix: correct psi 2D grid orientation for rectangular equilibrium grids #17 )
• Fix compatibility with NumPy >= 2.0 (AttributeError: np.NaN was removed in the NumPy 2.0 release. Use np.nan instead. #10 )

[olivhoenen]

@SimonPinches in the EB script associated with this release, we need in the module to either unset MPLBACKEND or setenv MPLBACKEND=TkAgg (to counterbalance the MPLBACKEND=Agg which is in matplotlib module and keep more user friendly default behaviour).

[SimonPinches]

I found three user-facing crashes in code paths this release ships (empty scalar leaf in idsprint, GGD data without a "cells" subset in plotedgeprofiles, and dbperf on entries with no readable IDS), plus a few behavioural changes worth either fixing or calling out in the release notes (forced cache_mode=none on UDA and the silent removal of --show-labels). Details in the line comments.

[SimonPinches]

Bug (crash + dropped output). value is only assigned in the else branch, so for an empty primitive (has_value == False) the Columns([txt, value]) line raises UnboundLocalError. And when the primitive does have a value, tree.add(group) only happens inside if compact: — in the default (non-compact) mode the value is built and then silently discarded, so idsprint shows only the type header.

Suggested change

	elif isinstance(structure, imas.ids_primitive.IDSPrimitive):
	tree = Tree(f"[magenta]{type(structure)}")
	if not structure.has_value:
	tree.add(f"[grey62]{structure.metadata.name}[/]")
	else:
	value = Pretty(structure.value)
	txt = f"[bright_yellow]{structure.metadata.name}[/]:"
	group = Columns([txt, value])
	if compact:
	group = Columns([f"[bright_yellow]{structure.metadata.name}[/]"])
	tree.add(group)
	return tree
	elif isinstance(structure, imas.ids_primitive.IDSPrimitive):
	tree = Tree(f"[magenta]{type(structure)}")
	if not structure.has_value:
	tree.add(f"[grey62]{structure.metadata.name}[/]")
	return tree
	if compact:
	group = Columns([f"[bright_yellow]{structure.metadata.name}[/]"])
	else:
	txt = f"[bright_yellow]{structure.metadata.name}[/]:"
	group = Columns([txt, Pretty(structure.value)])
	tree.add(group)
	return tree

[SimonPinches]

Bug (crash on missing "cells" grid subset). get_cell_polygons_and_values() is documented to return (None, None) when the "cells" grid subset is unavailable, but this immediately iterates polygons → TypeError: 'NoneType' object is not iterable, so the graceful "no data" branch below never runs. Same pattern in view_ion_density (line 177) and view_neutral_density (line 226).

Suggested change

	polygons, cell_values = self.edge_profiles_compute.get_cell_polygons_and_values(time_slice, ne_values)
	all_r = np.concatenate([v[:, 0] for v in polygons])
	polygons, cell_values = self.edge_profiles_compute.get_cell_polygons_and_values(time_slice, ne_values)
	if polygons is None or len(polygons) == 0:
	logger.warning("edge_profiles: no cell polygons available, cannot plot electron density")
	return None
	all_r = np.concatenate([v[:, 0] for v in polygons])

(Ideally the empty-data text shown in the existing else branch would be reused, but the guard above at least prevents the crash. Applies to all three density views.)

[SimonPinches]

Bug (NameError on sparse entries). If the entry contains no readable IDS (empty availables_ids, or every get() raises), src_dd_version is never assigned and the DBEntry(..., dd_version=src_dd_version) calls below raise NameError. Initialising it to None falls back to the default DD, which restores the previous behaviour. The unused except Exception as e will also trip flake8 (F841).

Suggested change

	_dummy_hdfde = imas.DBEntry(
	imas.ids_defs.HDF5_BACKEND, args.database, pulse, run, args.user, data_version=args.version
	)
	_dummy_hdfde.open()
	availables_ids = get_available_ids_and_occurrences(_dummy_hdfde)
	for idsname, occurrence in availables_ids:
	try:
	_dummy = _dummy_hdfde.get(idsname, lazy=True, autoconvert=False)
	src_dd_version = _dummy.ids_properties.version_put.data_dictionary.value
	break
	except Exception as e:
	pass
	_dummy_hdfde.close()
	src_dd_version = None
	_dummy_hdfde = imas.DBEntry(
	imas.ids_defs.HDF5_BACKEND, args.database, pulse, run, args.user, data_version=args.version
	)
	_dummy_hdfde.open()
	availables_ids = get_available_ids_and_occurrences(_dummy_hdfde)
	for idsname, occurrence in availables_ids:
	try:
	_dummy = _dummy_hdfde.get(idsname, lazy=True, autoconvert=False)
	src_dd_version = _dummy.ids_properties.version_put.data_dictionary.value
	break
	except Exception:
	continue
	_dummy_hdfde.close()

[SimonPinches]

Ineffective check + regression risk. Two points here:

1. profiles2d.grid.dim1 is not None is never False for IMAS array nodes — an equilibrium with grid_type == 1 but empty grid.dim1/dim2 sails through and produces an empty contour. A length check does what was intended:

Suggested change

	if profiles2d.grid_type.index == 1 and profiles2d.grid.dim1 is not None and profiles2d.grid.dim2 is not None:
	if profiles2d.grid_type.index == 1 and len(profiles2d.grid.dim1) > 0 and len(profiles2d.grid.dim2) > 0:

2. Previously this function also supported entries that filled 2-D profiles_2d.r/z directly (it fell back to them, or built the mesh from dim1/dim2 only when r was empty). Those entries now get "Only rectangular cylindrical grid (grid_type=1) is supported" and None. If that's intentional, worth a line in the release notes. Also: the returned keys are still named r2d/z2d but now hold 1-D arrays — fine for the in-repo contour(r, z, psi.T) call sites (all updated, I checked), but misleading for any external consumer; consider renaming to r/z or noting it in the docstring.

[SimonPinches]

Nit: integer fill value passes validation. _valid_scalar uses abs(EMPTY_FLOAT) ≈ 9e40 as threshold, so EMPTY_INT (−999999999) passes and bnd_type can be set to the fill value; the separatrix fallback below then silently takes the limiter/lcfs path. Suggest an explicit check:

Suggested change

	bnd_type = int(bnd.type)
	if _valid_scalar(bnd_type):
	bnd_type = int(bnd.type)
	if bnd_type != imas.ids_defs.EMPTY_INT and _valid_scalar(bnd_type):

[SimonPinches]

Nit: duplicate import (F811) — duplicate of the (deliberately early) import on line 15.

Suggested change

from idstools.view.common import PROVENANCE_TITLE_STYLE, PlotCanvas

[SimonPinches]

Silent CLI removal. --rho→--phi and --plots→--profiles are covered in the docs, but --show-labels was removed without a mention — existing scripts using it will now exit with an argparse error. Worth a line in the 2.5.0 release notes (or keeping it as a deprecated no-op for one release).

[SimonPinches]

Handler accumulation + function size. Each call to view_magnetic_poloidal_flux registers a new pick_event handler on the figure — and plotequilibrium calls it twice (base plot + create_overlays), so handlers accumulate. Today the stale handler's leg_map misses the new legend artists so it's benign, but it's a leak pattern that will bite as soon as artists are reused. Suggest connecting once per figure (e.g. keep the cid on the figure and disconnect before reconnecting).

More broadly, this function has grown to ~380 lines doing contours, six kinds of overlay, annotations and an interactive legend — extracting the overlay/legend sections into helpers would make it much easier to maintain.

[SimonPinches]

Pre-existing bug worth a follow-up issue. Not introduced by this PR, but visible in the surrounding context of the np.nan changes: there are several bare len() calls (if len() < 1: — lines 839, 878, 908, 915, 1092) in the edge-profiles branches. These raise TypeError whenever those paths execute, so the fallbacks they guard can never work. Suggest a follow-up issue rather than folding into this release PR.

[SimonPinches]

Nit: misleading log message. The message says "falling back to node interpolation", but ne_list is only populated when no per-subset density entry was found — when we reach this branch (entry found but wrong length) it is always empty, so the function returns (None, None, None) despite the message. Either build the node fallback in this case too, or reword the message.