[Lldb-commits] [PATCH] D123226: [lldb] use one shared ThreadPool and task groups

[Alexandre Ganea via Phabricator via lldb-commits]

aganea added inline comments.


================
Comment at: lldb/source/Core/Debugger.cpp:1969
+llvm::ThreadPool &Debugger::GetThreadPool() {
+  static llvm::ThreadPool threadpool(llvm::optimal_concurrency());
+  return threadpool;
----------------
clayborg wrote:
> aganea wrote:
> > Ideally this should be explicitly created on the stack & passed along on the callstack or in a context, like MLIR does: https://github.com/llvm/llvm-project/blob/main/mlir/include/mlir/IR/MLIRContext.h#L48
> We need one instance of the thread pool so we can't create on the stack. 
> 
> Static is not great because of the C++ global destructor chain could try and use if "main" exits and we still have threads running. I would do something like the "g_debugger_list_ptr" where you create a static variable in Debugger.cpp:105 which is a pointer, and we initialize it inside of Debugger::Initialize() like we do for "g_debugger_list_ptr". Then the thread pool will not cause shutdown crashes when "main" exits before other threads.
> 
I meant having the `ThreadPool` in a context created by main() or the lib caller, before getting here in library/plugin code, and passed along. LLD does that too: https://github.com/llvm/llvm-project/blob/main/lld/ELF/Driver.cpp#L94 - the statics in `Debugger.cpp` seems like a workaround for that. In any case, it's better than having the `static` here.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123226/new/

https://reviews.llvm.org/D123226