llvm dev - Jun 2019 - RFC: Two llvm-readobj behaviour queries

Hi all,

I just was discussing a couple of features in llvm-readobj and whether we
should change them slightly:

1) llvm-readobj --all currently matches GNU readelf's --all switch in that
it is equivalent to specifying a bundle of different options (including
--file-headers, --program-headers, --symbols, --version-info etc). However,
--all does not actually specify "all" dump types. For example, it
doesn't
dump ELF section groups or the addrsig section (--elf-section-groups and
--addrsig respectively). I'd propose that it should dump all relevant
dumping information. Ideally, I'd suggest this difference for both
llvm-readobj and llvm-readelf, but I can understand nervousness in the
additional dumps that this would produce in llvm-readelf versus GNU readelf.

2) llvm-readobj --notes prints the SHT_NOTE sections unless the ELF type is
ET_CORE, in which case it prints the contents of the PT_NOTE segments and
NOT the SHT_NOTE sections. It seems to me like there's no good reason to
not try both for all file formats. For example, it's perfectly reasonable
to have PT_NOTE segments in an ET_EXEC or ET_DYN object file. What would
people think about removing this distinction? In this case, I'd propose
printing all PT_NOTE segments and all SHT_NOTE sections not in PT_NOTE
segments. In all likelihood, this would produce no different behaviour to
the old behaviour.

Thoughts?

James
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190624/8dd056ac/attachment.html>

On Mon, 24 Jun 2019 at 16:16, James Henderson via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
>
> Hi all,
>
> I just was discussing a couple of features in llvm-readobj and whether we
should change them slightly:
>
> 1) llvm-readobj --all currently matches GNU readelf's --all switch in
that it is equivalent to specifying a bundle of different options (including
--file-headers, --program-headers, --symbols, --version-info etc). However,
--all does not actually specify "all" dump types. For example, it
doesn't dump ELF section groups or the addrsig section (--elf-section-groups
and --addrsig respectively). I'd propose that it should dump all relevant
dumping information. Ideally, I'd suggest this difference for both
llvm-readobj and llvm-readelf, but I can understand nervousness in the
additional dumps that this would produce in llvm-readelf versus GNU readelf.
>
> 2) llvm-readobj --notes prints the SHT_NOTE sections unless the ELF type is
ET_CORE, in which case it prints the contents of the PT_NOTE segments and NOT
the SHT_NOTE sections. It seems to me like there's no good reason to not try
both for all file formats. For example, it's perfectly reasonable to have
PT_NOTE segments in an ET_EXEC or ET_DYN object file. What would people think
about removing this distinction? In this case, I'd propose printing all
PT_NOTE segments and all SHT_NOTE sections not in PT_NOTE segments. In all
likelihood, this would produce no different behaviour to the old behaviour.
>
> Thoughts?
>
I agree with both proposals. Although I suspect that people parsing
the output of GNU readelf won't be using the --all switch I think it
is reasonable to keep the option sets the same.

Peter
> James
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

On Mon, Jun 24, 2019 at 8:16 AM James Henderson via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hi all,
>
> I just was discussing a couple of features in llvm-readobj and whether we
> should change them slightly:
>
> 1) llvm-readobj --all currently matches GNU readelf's --all switch in
that
> it is equivalent to specifying a bundle of different options (including
> --file-headers, --program-headers, --symbols, --version-info etc). However,
> --all does not actually specify "all" dump types. For example, it
doesn't
> dump ELF section groups or the addrsig section (--elf-section-groups and
> --addrsig respectively). I'd propose that it should dump all relevant
> dumping information. Ideally, I'd suggest this difference for both
> llvm-readobj and llvm-readelf, but I can understand nervousness in the
> additional dumps that this would produce in llvm-readelf versus GNU
readelf.
>
> 2) llvm-readobj --notes prints the SHT_NOTE sections unless the ELF type
> is ET_CORE, in which case it prints the contents of the PT_NOTE segments
> and NOT the SHT_NOTE sections. It seems to me like there's no good
reason
> to not try both for all file formats. For example, it's perfectly
> reasonable to have PT_NOTE segments in an ET_EXEC or ET_DYN object file.
> What would people think about removing this distinction? In this case,
I'd
> propose printing all PT_NOTE segments and all SHT_NOTE sections not in
> PT_NOTE segments. In all likelihood, this would produce no different
> behaviour to the old behaviour.
>
> Thoughts?
>
> James
>
>
I'm fine with both changes.

For the first, I prefer llvm-readelf --all mimics gnu readelf.

- Michael Spencer
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190625/25859f57/attachment.html>