[petlenz]

Closes #66, #67, #148, #149 (the two rows of #65 fixed here; #65 stays open for its remaining sub-issues #150–#153).

Three failures that named nothing useful.

A wrong parameter type was a bare bad_any_cast (#65)

get_parameter() let std::bad_any_cast escape: no parameter, no material, no type, and not a class a caller would think to catch. Writing 1 where 1.0 is read -- an int for a double, which both JSON and C++ make easy -- produced bad any_cast and nothing else, at construction, before the material could say anything about itself.

set_parameter() already wrapped this. get, which every material calls several times, did not.

One reader served two different parameters (#66)

The JSON reader registry is keyed on std::type_index alone. block_ref is std::pair<std::string, std::string>, so vector_newton's zero_blocks and weighted_sum's terms are the same C++ type and resolve to the same reader. A malformed weighted_sum term reported

zero_blocks: each entry must be a [row, column] pair

naming a parameter of a different material.

Readers now receive the parameter key -- it was already available at the read() call site and thrown away -- so the shared reader names whatever it was actually given. add_for_key() lets one parameter keep its own wording without disturbing the other user of the type, so zero_blocks still says [row, column].

Any future pair of parameters sharing a C++ type would have hit the same thing.

"type" errors were a dead end (#67)

A missing "type" surfaced nlohmann's own key 'type' not found, from a library the user never named, with no indication of which entry was at fault. A misspelt one echoed the name back with no candidates.

Both now name the entry by its "name", and an unknown type suggests the nearest registered one (j2_plasticty -> j2_plasticity) or lists them all when nothing is close.

Not fixed here

The remaining rows of #65 are its sub-issues: unknown optional keys in the C++ path (#150), non-integral values for integer parameters (#151), check_data_type through the declaration helpers (#152), all missing names in the exception text (#153, handler side numsim-core#20).

[petlenz]

Review findings (design, tmech expressions instead of component loops, modern C++, comments, tests, correctness). Severity: must = wrong result, hang, build break or unreachable feature; should = design, duplication, accuracy; nit = style.

parameter diagnostics — needs changes

• [should] wrapper catches only bad_any_cast; numsim-core's missing-key invalid_argument still names no material; rvalue overload now a copying duplicate → delete; message names neither expected nor stored type; where attached only to the two "type" checks (reader failures and ctx.create errors arrive without the entry); "name": 3 surfaces nlohmann's type_error; "capped" Levenshtein comment is false; two identical read_name_pairs bodies; <print> drift vs main.

Conflicts and landing order (open PRs on main)

• tests/CMakeLists.txt: core: reject malformed models instead of running them (#56, #57, #60, #64, #69) #79 and core: fail loudly when a material reaches for something unwired (#59, #62) #84 both insert after test_property_graph; solvers: report a Runge-Kutta step that did not converge (#58, #61) #83 vs main's foreach; tests: drive vector_newton through a real return map (kinematic hardening) #55 vs main.
• core/material_interface.h: core: fail loudly when a material reaches for something unwired (#59, #62) #84 and io: name the parameter, the material and the entry when a deck is wrong (#66, #67) #85 overlap.
• Order: core: fail loudly when a material reaches for something unwired (#59, #62) #84 → io: name the parameter, the material and the entry when a deck is wrong (#66, #67) #85 (rebased on it) → core: reject malformed models instead of running them (#56, #57, #60, #64, #69) #79 → solvers: report a Runge-Kutta step that did not converge (#58, #61) #83; solvers: drop the relative-residual guard, it fired on rounding noise (#21) #53 merges clean; tests: drive vector_newton through a real return map (kinematic hardening) #55/cmake: make the installed package compile for a consumer (#70, #71) #81/build: clean the warning set and enforce it in CI (#76) #86 rebase onto 7bcba93 and shrink to their non-CMake content.
• STATEV: stress-like tensor history without the engineering-shear factor (#63) #91 ↔ tests: drive vector_newton through a real return map (kinematic hardening) #55: back_stress becomes stress_like.

[petlenz]

Rebased on main; the review findings addressed in the commit:

• get_parameter: one overload (the rvalue one was a copying duplicate); a wrong stored type says what the material reads it as ("is not a real number as this material reads it …"), and a missing key no longer surfaces the handler's "Key K not found" but "material 'x': parameter 'K' is missing".
• create_from_json: the entry (where) is attached to every error — reader failures and the material's own checks included — not only to the two "type" checks; a non-string "name" is reported as such instead of reaching nlohmann first.
• edit_distance is capped as its comment claimed (stops when a whole row exceeds the cap).
• Comments cut to one or two lines throughout; the test essays too. Tests added: missing parameter names material and key, the type read is named, a reader failure and a missing required parameter carry the entry, a numeric "name".

The read_name_pairs duplicate and the <print> drift are not in the rebased diff any more.

The diagnostics, set_parameter and json_model suites pass locally (g++-14); the full run follows in CI.

[petlenz]

Follow-up push: the loops of the PR are std algorithms now — suggest_type uses std::ranges::min_element with the edit distance as projection, std::ranges::sort and std::accumulate; the two block_ref readers share one string_pairs helper built on std::ranges::transform (the last duplicated body). Full local run before this push: 682/682; the four affected suites pass after it.