krytarowski added inline comments.
================
Comment at: include/lldb/Core/Debugger.h:379
lldb::ListenerSP m_forward_listener_sp;
- std::once_flag m_clear_once;
+ llvm::once_flag m_clear_once;
----------------
labath wrote:
> The code in llvm says you should only ever use the LLVM_DEFINE_ONCE_FLAG to
> declare flags. I am guessing it's because it forces the `static` keyword into
> the declaration, which in turn guarantees your object will be
> zero-initialized by the linker. If you declare the flag as a local variable
> like this, it will be initialized to a random value, and you will have a fun
> time debugging issues in the future (it will only affect netbsd, as
> std::call_once platforms will still be correctly initialized).
I was thinking about it, there is `m_clear_once()` in
`lldb/source/Core/Debugger.cpp` used in the initializer. But apparently there
is no zeroing it with this.
Similar case is in DWARFDataSegment.
```
struct DWARFDataSegment {
llvm::once_flag m_flag;
lldb_private::DWARFDataExtractor m_data;
};
```
And another one with `static llvm::ManagedStatic<llvm::once_flag> g_once_flag;`
(I don't fully understand the impact here).
Is there an option to turn `llvm::once_flag` to behave like `std::once_flag`?
For example changing once_flag to a class like:
```
class once_flag
{
once_flag() : status(Uninitialized) {};
llvm::sys::cas_flag status;
};
```
and replace occurrences of `flag` with `flag.status`.
Perhaps it would change the code to mechanical replacement of `std::once_flag`
and `std::call_ones` to `llvm::` versions.
Repository:
rL LLVM
https://reviews.llvm.org/D29288
_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
