RFC: llvm-readelf Mach-O & COFF options

Hi all,

llvm-readelf is an alias for llvm-readobj which aims for GNU compatibility and is likely the tool that most people migrating to the LLVM binutils will adopt instead of llvm-readobj. Because it is just an alias, it has inherited the functionality provided by llvm-readobj, including for non-ELF targets, with Mach-O and COFF-specific switches available in its interface. People migrating from GNU readelf won’t need these switches.

I have a change up for review to expand the documentation for llvm-readelf to list all its supported switches (see https://reviews.llvm.org/D63826). This patch currently includes the Mach-O and COFF specific switches. It was suggested that these could be surprising and unnecessary, and could be removed from the patch. If we go ahead and remove them from the documentation, I think we should remove them from the help text, or possibly even the available options entirely. I’d be happy to do the work for this, if the community agrees that llvm-readelf should only support ELF-related options.

Thoughts?

James

My perspective is that llvm-readobj is a testing tool whose goal should be to make testing LLVM easy, not 100% GNU readelf compatibility, but I think that ship has sailed.

I don’t really care if you decide to move the functionality, I just want to have some binary in LLVM for exploratory dumpers, where I don’t have to care too much about consistency with another tool that makes different arbitrary formatting decisions. So, if you want to move it, I guess we need some new binary where we can put experimental code. “llvm-objutil” mirroring llvm-pdbutil?

llvm-readobj is fine as is, and as far as I know, there’s no issue with it needing to be GNU compatible. llvm-readelf on the other hand should be, based on various discussions over the last couple of years. I’m not proposing removing the Mach-O and COFF options from llvm-readobj at all. I think it makes sense to keep them where they are. What I’m suggesting is removing them from the options in llvm-readelf ONLY (detected by the name of the executable, similar to how we already decide what aliases to support). Ideally, I’d expect these options to still be available in llvm-readelf, and they’d do exactly the same as what llvm-readobj does, just that they wouldn’t be printed by llvm-readelf --help.

To reiterate, this proposal does not affect the behaviour of the llvm-readobj executable, unless it has been renamed to llvm-readelf. Does that make sense?

James

Sounds good to me, sorry, when you said remove the functionality from readelf, I thought you meant to remove it from readobj. Disabling the MachO and COFF options in readelf mode so they don’t show up in --help sounds good to me.