xcode 16 vs. RandomSource ==> linker errors
Closed this issue · 5 comments
What happened?
$ bazel build //systems/primitives:random_source_test
...
Undefined symbols for architecture arm64:
"drake::systems::RandomSource<drake::symbolic::Expression>::~RandomSource()", referenced from:
vtable for drake::systems::RandomSource<drake::symbolic::Expression> in random_source_test.o
"drake::systems::RandomSource<drake::symbolic::Expression>::~RandomSource()", referenced from:
vtable for drake::systems::RandomSource<drake::symbolic::Expression> in random_source_test.o
"drake::systems::RandomSource<drake::symbolic::Expression>::SetRandomState(drake::systems::Context<drake::symbolic::Expression> const&, drake::systems::State<drake::symbolic::Expression>*, drake::RandomGenera\
tor*) const", referenced from:
vtable for drake::systems::RandomSource<drake::symbolic::Expression> in random_source_test.o
"drake::systems::RandomSource<drake::symbolic::Expression>::SetDefaultState(drake::systems::Context<drake::symbolic::Expression> const&, drake::systems::State<drake::symbolic::Expression>*) const", referenced\
from:
vtable for drake::systems::RandomSource<drake::symbolic::Expression> in random_source_test.o
It definitely has something to do with is_symbolic_convertible in the unit test. I've tried to peel things apart further, but I feel like I'm getting contradictory results as I delve down into the implementations of is_dynamic_castable and NiceTypeName.
For a full CI build with the same failure, see https://drake-jenkins.csail.mit.edu/job/mac-arm-sequoia-unprovisioned-clang-bazel-experimental-release/9/consoleFull
Version
master circa 1.34
What operating system are you using?
macOS 14 (Sonoma)
What installation option are you using?
compiled from source code using Bazel
Relevant log output
No response
My guess is that instead of EXPECT_FALSE(is_symbolic_convertible ...) we should do EXPECT_TRUE(is_not_symbolic_convertible .. ) -- by adding a new helper function ("is_not_...") for that purpose.
What I haven't yet investigated are the other existing instances of EXPECT_FALSE(is_symbolic_convertible), why they don't fail, and why they might be different.
That's a fair question, but dynamic_cast<MyNonSymbolicSystem<Expression>> is unquestionably invalid code, so I was just thinking we stop doing that, see what happens, and maybe just move on if it works.
So, the difference that matters is ... RandomSource is declared final. The other potential victims are not.
It appears that (for apple clang 16 alone afaict) final means "fill out my instantiation's vtable, no matter what". Ubuntu (all compilers tried) proceed just fine.
Fixed with merge of #22188.