[lldb] Add parent address to Task synthetic provider #10936
Conversation
|
@swift-ci test |
| parent_addr = process_sp->ReadPointerFromMemory( | ||
| m_task_ptr + ChildFragmentOffset, status); | ||
| if (status.Fail() || parent_addr == LLDB_INVALID_ADDRESS) | ||
| parent_addr = 0; |
There was a problem hiding this comment.
is this really what we want to do here? When the task has no parent, does the runtime write a 0 to its slot, or do we expect an error?
If the runtime is explicitly writing a 0 and we don't expect the memory read to fail, I think we can just return nullptr here.
There was a problem hiding this comment.
For the case of a parent-less task, should we try to customize the formatter output to say something "root task" or "no parent", instead of 0s?
I think 0s could give users the impression that the debugger failed to get the value
There was a problem hiding this comment.
When the task has no parent, does the runtime write a 0 to its slot
A swift Task can have tail allocated "fragments", including a ChildFragment. Child tasks have a ChildFragment, while root tasks do not.
One way to model that is to make the existence of a parent conditional. In other words, for a root task, no parent member will be shown at all. The reason I didn't do that is:
- Users might find it confusing (why is there a
parentsome times, but not other times) - It makes the synthetic child provider more complex (which is fine if we have good reason)
For the case of a parent-less task, should we try to customize the formatter output to say something "root task" or "no parent", instead of 0s?
The parent member needs a data type, which currently is UnsafeRawPointer. To show a task-specific string like "root task", we'd need to use some other data type, one we invent ourselves. This adds complexity, and will be exposed to users who might wonder "what's a LLDBParentTaskPointer?".
We could put a string like "root task" into Task's summary formatter. In which case the parent = 0x0000... would be accompanied with "root task" in the summary string, which should explain the parent address.
On this subject, one of the questions I have for reviewers is the name of this member. I used parent, but felt like make it should be parent_addr. The main reason I didn't do that is the provider has an address field, and I didn't like the inconsistency of using address and addr, and I didn't like the verbose choice of parent_address. Maybe we can rename address to addr, and then use parent_addr. Thoughts? I think including addr in the name might make it more clear that a bunch of zeros means root task/no parent.
There was a problem hiding this comment.
A swift Task can have tail allocated "fragments", including a
ChildFragment. Child tasks have aChildFragment, while root tasks do not.
I misspoke. A root task will of course have a ChildFragment if it has children.
There was a problem hiding this comment.
Adrian's suggestion is to have the parent's type be UnsafeRawPointer?, and will show nil when there's no parent. Works for me, what do you think Felipe?
|
@swift-ci test |
|
@swift-ci test |
|
@swift-ci test |
Display the address of a Task's parent (or 0x0000000 for root tasks).
Exposing a Task's parent will allow:
task info <addr>)task backtrace <addr>)task select <addr>)rdar://147851193