[Bug ld/27441] Small inconsistency in between gold and bfd

[i at maskray dot me]

https://sourceware.org/bugzilla/show_bug.cgi?id=27441

--- Comment #21 from Fangrui Song <i at maskray dot me> ---
I made a mistake in #comment 16.

In GNU ld, the rule is probably:

* it is linked at least once in `--no-as-needed` mode (i.e. `--as-needed a.so
--no-as-needed a.so` => needed)
* or it has a definition resolving a non-weak reference by a previous input
file (it works similar to archive member selection)



In gold, the rule is probably:

* it is linked at least once in `--no-as-needed` mode (i.e. `--as-needed a.so
--no-as-needed a.so` => needed)
* or it has a definition resolving a non-weak reference


In LLD, a shared object is needed, if one of the following conditions is true:

* it is linked at least once in `--no-as-needed` mode (i.e. `--as-needed a.so
--no-as-needed a.so` => needed)
* or it has a definition resolving a non-weak reference from a live section
(not discarded by `--gc-sections`)


Yes, the GNU ld behavior matches its documentation. But gold and LLD's
behaviors are probably more useful (less reliance on position dependent
behaviors - the user has more freedom moving it around on the linker command
line without changing symbol resolution) and can make implementation simpler
(no need to handle --as-needed shared object similar to the tricky archive
member selection)

# a.s
call foo
# b.s
.globl foo
foo:
  ret

gold --as-needed b.so a.o && readelf -d a.out
...
 0x0000000000000001 (NEEDED)             Shared library: [b.so]

-- 
You are receiving this mail because:
You are on the CC list for the bug.