[llvm] [llvm-objdump][BPF] --symbolize-operands: infer local labels for BPF (PR #100550)

[James Henderson via llvm-commits]

jh7370 wrote:

> > > @yonghong-song , @4ast , do we want to make --symbolize-operands default for bpf and if so, should this be a part of this commit or a #95103? I understand why @MaskRay does not like it, but usability benefits seem significant.
> > 
> > 
> > The patch can land as is. The default --symbolize-operands does not need to be combined. It would be a pretty significant policy change as we do not allow targets to customize the defaults. That would require more debates.
> 
> No. User experience matters a lot more than some "policy". We're going to keep L in symbol table then.

This feels like an excessively strong reaction to what @MaskRay has been saying, especially as there wasn't even room for discussion.

This is an open source project, with a wide range of stakeholders. These "policies" are what are needed to keep the code in a maintainable state for all of these stakeholders. That being said, there is always an opportunity to discuss changing a policy or even working together to find an alternative solution that will satisfy everyone.

I'm not familiar at all with the BPF target. Ignoring the prior situation of a bug that used the wrong name for temporary symbols (i.e. the "L"/".L" symbols), is BPF significantly different to other targets on the need for a different default for the --symbolize-operands option?

https://github.com/llvm/llvm-project/pull/100550