[FFroehlich]

Stacked on #3525.

ReturnDataView.__init__ built a dict with the shapes of all array fields. For every view, that read about 35 attributes of the SWIG-wrapped ReturnData and allocated about 40 GC-tracked objects, whether or not any field was read. ExpDataView did the same for its fields.

_field_dimensions is now a class attribute that maps each field to a function of the wrapped object returning the field's shape. It is evaluated when the field is first accessed, and the resulting array is cached as before. Constructing a view is now just a type check plus setting the pointer and an empty cache.

Shapes are now read at the first access of a field rather than at view creation. So an ExpDataView sees changes to the ExpData made before a field is first read; before, this failed with a reshape error. A new test covers this.

Measurements

Setup: Python 3.13, a 2-state model, run_simulations with 9169 ExpDatas and forward sensitivities, 6 calls (55k views). Afterwards llh, sllh and FIM are read from every result.

	view construction	GC time	gen0 collections	reading llh, sllh, FIM
#3525	1.36 s	0.44 s	1049	0.32 s
this PR	0.03 s	0.01 s	77	0.31 s

With 12M long-lived GC-tracked objects in the process (standing in for a large PEtab problem), view construction went from 2.29 s to 0.03 s, and GC time from 1.32 s to 0.82 s. One full collection remains, triggered by the earlier allocations of the large heap.

This is the cost that the profile in #3510 (comment) charged to XArrayFactory.__init__ (about 11 s for 18,338 views). See #3525 for why cProfile charged it there.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.31%. Comparing base (32e3f4e) to head (c28e07b).

Additional details and impacted files

@@                      Coverage Diff                       @@
##           claude/quirky-margulis-928f12    #3526   +/-   ##
==============================================================
  Coverage                          80.30%   80.31%           
==============================================================
  Files                                331      331           
  Lines                              23438    23440    +2     
  Branches                            1631     1631           
==============================================================
+ Hits                               18823    18825    +2     
  Misses                              4609     4609           
  Partials                               6        6           

Flag	Coverage Δ
cpp	74.81% <100.00%> (+<0.01%)	⬆️
cpp_python	34.64% <100.00%> (+0.01%)	⬆️
petab	46.93% <100.00%> (+0.01%)	⬆️
petab_sciml	16.62% <42.85%> (+<0.01%)	⬆️
petab_sciml_benchmarks	15.21% <42.85%> (+<0.01%)	⬆️
python	72.60% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
python/sdist/amici/sim/sundials/_numpy.py	88.12% <100.00%> (+0.15%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.