Add support for non-string merge sections

[davidlattimore]

When we encounter an input section with the merge flag set, but without the strings flag set, we currently just ignore the merge flag. We should actually deduplicate that input section with other input sections with the same name.

Some links for additional context (thanks to Trevor Gross):

https://inbox.sourceware.org/binutils/CALNs47sfhiiCPi4o=otZ4k3nEt=byB=hv3yEowLO5rKU8CKt+Q@mail.gmail.com/T/#u

rust-lang/libs-team#466

[davidlattimore]

@karolzwolak

To add some additional context, the existing string merging is done in string_merging.rs. I expect that the simplest option is likely to be to reuse that and have some code that skips splitting the section and instead uses the entire section as one string.

The string merging algorithm is kind of complex for performance reasons, but hopefully you won't need to touch most of it.

The function part_id::should_merge_strings will need to change. Mostly likely it'll need to ignore the shf::STRINGS flag. The strings flag would then need to be checked somewhere else, perhaps somewhere like group_merge_string_sections_by_output.

[karolzwolak]

I"ve got hacky MVP working I think. Any tips on setting up the tests for this feature? Oh and please assign me to this. @davidlattimore

[davidlattimore]

Great! For testing, I think either add to string_merging.c or add a new test that does something similar but with non-string sections. In string_merging1.s, where the section is declared .section .rodata.strings, "aSM", @progbits, 1 you can do the same, but without the "S" bit. When declaring the strings - e.g. .ascii "Hello\0" you can remove the \0 at the end.

[karolzwolak]

As discussed with @davidlattimore there's no need to follow up the implementations with additional renaming so this should be closed as completed.

[davidlattimore]

Thanks for implementing this!