On 1/8/24 04:21, Iain Sandoe wrote:
On 6 Jan 2024, at 22:30, Nathan Sidwell <nat...@acm.org> wrote:
Richard Smith & I discussed whether we should use the module interface's
capability of giving vague linkage entities a strong location. I didn't want to go
messing with that, 'cos it was changing yet more stuff.
But, perhaps we should revisit that? Any keyless polymorphic class in module
purview gets its vtables etc emitted in the module's object file? Likewise
these kinds of entities.
cc'ing Iain, who probably knows more about Clang's state here.
I have been trying to keep up with this thread, but not sure if I can throw a
whole lot of light on things.
There is an on-going attempt (now some 3 or 4 papers in) to try and figure out
how to handle `static inline` entities at least at file scope - but that
appears to be a different case (I can try an locate the latest paper on this if
needed; the topic was discussed in Varna and Kona, but no new paper yet -
perhaps Michael [Spencer] will bring a paper in Tokyo).
clang ran into some issues with vtables and that resulted in some discussion
about whether there should be an amendment to the Itanium ABI to deal with the
module-specific stuff.
https://github.com/itanium-cxx-abi/cxx-abi/issues/170
https://github.com/llvm/llvm-project/pull/75912#discussion_r1444150069
Sorry I cannot be much more specific at present,
That's pretty specific that vtables at least get emitted in the module
whether or not there's a key function. I've asked on that issue why
this only applies to vtables.
Jason
