Missing-parameter error reports a count, not the parameter names

[petlenz]

Summary

When required parameters are missing, input_parameter_controller::accept() throws an exception carrying only a count, not the names:

missing 1 required parameter(s)

The names are known at the throw site — they are collected in missing and printed to stdout — but they never reach the exception. A caller that catches the error, or runs anywhere stdout is not visible, is told that something is missing but not what.

Where

include/numsim-core/input_parameter_controller.h, in accept() (step 4):

    // 4. Report all missing required parameters at once
    if (!missing.empty()) {
      std::println("  missing required parameters:");
      for (const auto& key : missing)
        std::println("    - {}", key);
      throw std::invalid_argument(
          "missing " + std::to_string(missing.size()) + " required parameter(s)");
    }

Why it matters

Two paths in this header report the same condition, and they disagree. The single-parameter check already does the right thing:

  void check(ParameterHandler &input) const final override {
    if (!input.contains(this->m_para.name())) {
      throw std::invalid_argument("Parameter " + this->m_para.name() +
                                  " is missing!");
    }
  }

So the quality of the diagnostic depends on which validation path happens to run. The batch path in accept() is the one a document- or deck-driven flow goes through, which is exactly the case where the user did not write the C++ and has no other way to find out which key is wrong.

Concretely, in numsim-materials a JSON material model that omits one cone parameter of a Drucker-Prager material fails with:

missing 1 required parameter(s)

for a document that named eta, beta, K_bulk, G, sigma_0, hardening_source, strain_source and solver_source — eight candidates, one of them wrong, and the error narrows it to none of them. We have a test pinning this behaviour rather than the desirable one (MaterialRegistry.ADruckerPragerDocumentMissingEtaFailsLoudly), with a comment pointing here.

Printing to stdout does not cover the gap:

• a library writing diagnostics to stdout is not usable from a host that owns that stream — an Abaqus UMAT, a Python binding, a GUI;
• the print and the throw can be separated by arbitrary output from other materials being constructed;
• a caller that catches and re-throws with context (as numsim-materials' UMAT boundary does, mapping failures to a solver return code) has nothing to add the names to.

Suggested fix

Interpolate the names that are already collected:

    if (!missing.empty()) {
      std::string names;
      for (const auto& key : missing)
        names += (names.empty() ? "" : ", ") + std::string(key);
      throw std::invalid_argument(
          "missing " + std::to_string(missing.size()) +
          " required parameter(s): " + names);
    }

Two things worth deciding at the same time:

1. Whether the std::println calls in accept() (missing parameters, and the using default: '{}' lines above them) belong in a library at all, or should move behind a caller-supplied sink.
2. Whether the two messages should be unified, so "Parameter eta is missing!" and the batch form read the same way.

Happy to send a PR for the message change if the shape above is what you want.