Description
Context: As part of our work towards integrating lld into rustc (#36120), we found out
and reported 3 scenarios ("bugs") where lld doesn't behave exactly the same as ld. While one of
the LLD devs was looking over one of the object files that was used one of the bug reports, they
found out something that seemed irregular:
Quoting @rui314's comment from the LLVM bug tracker:
I sent out https://reviews.llvm.org/D24789 to fix the issue.
But I'm not completely sure if the object file created by rustc is valid. At least it seems to be causing a unnecessary trouble by choosing an irregular value for sh_entsize.
hello.o's .debug_gdb_scripts is a mergeable section with sh_entsize = 0. Usually the field is 1 for a section containing mergeable strings.
Jorge, I think you want to change rustc so that it sets 1 instead of 0.
The object mentioned in the comment can be generated with:
$ rustup default nightly-2016-09-15
$ echo 'fn main() { println!("Hello, world!"); }' > hello.rs
$ rustc -g hello.rs
This issue is to discuss if we should change the value of sh_entsize for this particular section.
This is above my paygrade so I'm going to cc @michaelwoerister who knows our debuginfo very well.
P.S. This particular bug has already been fixed in lld by, AIUI, relaxing one of their checks so our
object files are accepted.