[solvers] IpoptSolver doesn't hard-code the linear solver

[jwnimmer-tri]

Towards #21476. This will simplify the work of changing which linear solver(s) the build system provides.

---

This change is 

[jwnimmer-tri]

+@rpoyner-tri for (low priority) feature review, please.

[rpoyner-tri]

question

Reviewed 6 of 6 files at r1, all commit messages.
Reviewable status: 1 unresolved discussion, needs at least two assigned reviewers

---

solvers/ipopt_solver.cc line 157 at r1 (raw file):

  const std::vector<std::string_view> linear_solvers =
      internal::GetSupportedIpoptLinearSolvers();
  if (!linear_solvers.empty()) {

nit is empty string a legal value for default_linear_solver_ ? If so, cool. If not, shouldn't there be an else ?

[jwnimmer-tri]

Reviewable status: 1 unresolved discussion, needs at least two assigned reviewers

---

solvers/ipopt_solver.cc line 157 at r1 (raw file):

Previously, rpoyner-tri (Rick Poyner (rico)) wrote…

nit is empty string a legal value for default_linear_solver_ ? If so, cool. If not, shouldn't there be an else ?

Yes, empty is a legal value. When empty, these lines earlier in this file ...

  if (!default_linear_solver.empty()) {
    set_string_option("linear_solver", default_linear_solver);
  }

... will not try to set any Drake-specific default, instead allowing IPOPT's built-in upstream default to remain unchanged.

This could happen if the user is building Drake as a Bazel external and has replaced our build of @ipopt with their own customized build.

[jwnimmer-tri]

+@SeanCurtis-TRI for platform review per schedule, please.

[SeanCurtis-TRI]

Reviewed 6 of 6 files at r1, all commit messages.
Reviewable status: complete! all discussions resolved, LGTM from assignees rpoyner-tri(platform),SeanCurtis-TRI(platform)