Relicensing (most of) binutils/libctf to LGPLv3+?

[Nick Alcock]

On 19 Sep 2024, Arsen Arsenović outgrape:

> Hi Nick,
>
> Nick Alcock <nick.alcock@oracle.com> writes:
>
>> When we contributed libctf to binutils, we made it GPL because we didn't
>> know of a use case for making it LGPL. Since then, however, a large
>> number of uses have turned out to need it, from uses in the kernel's
>> build system (GPLv2-only) all the way to use in libside to potentially
>> allow entirely arbitrary programs containing libside probes to translate
>> their own CTF info into the libside type system. These programs might be
>> licensed under anything at all.
>>
>> It has become clear that at least the read side of libctf is really a
>> sort of compiler runtime library component: it's reading things GCC can
>> emit into arbitrary binaries, and surely those arbitrary binaries should
>> be allowed to read the type system describing their own code!
>>
>> There is a split here that we should consider. libctf.so links against
>> libbfd, so it must remain GPLv3+, as does libctf/ctf-open-bfd.c: these
>> pieces implement ctf_open() and ctf_fdopen(), which read CTF out of an
>> executable object using libbfd.
>>
>> But there is also libctf-nobfd.so, which does not include this feature
>> (specifically because some consumers don't need or want BFD). This
>> component uses nothing but libiberty (and on some platforms libz and
>> libintl). This, and its source constituents -- everything under libctf/
>> other than the build scripts and libctf-open-bfd.c, and
>> include/ctf{-api,}.h -- can be LGPLv3+'d without issue, and I think
>> should be.
>
> Note that source files utilizing GPLv3+ APIs do not have to be GPLv3(+),
> and can be LGPLv3+.  I have a preference for keeping all files GPLv3+,
> but if that is not workable for whatever reason, keeping a uniform
> license across the entire (sub)project might be desirable.

Well... I doubt it's going to make people here too *happy* if they see
bfd being used by (say) binary-only proprietary llds simply because it's
being pulled in by libctf.so and we elected to make that LGPLv3+.

I think it makes more sense to try to model what we want rather than
just what is simplest: i.e., don't try to pull bfd into something LGPL
and instead just keep the non-BFD-using subset (which we already have!)
LGPL and keep the bfd-using portion GPL. It's all partitioned off into
files already...

-- 
NULL && (void)