Also serialize dt->instance. Fixes #5129

[Keno]

dt->instance is one of the jl_value_t*'s that can occur purely in code
and not anywhere else in the AST. All other instances of where this
happends are captured by li->roots, since they need GC roots that way,
but this instance is rooted by the owning type.

@JeffBezanson @vtjnash

[JeffBezanson]

I don't think this is the full fix. What we want to do is iterate over the jl_value_to_llvm table and dump all the needed indexes. This would subsume jl_serialize_gv_syms and all the other ad-hoc calls to jl_serialize_gv. The only thing that makes this difficult is that some entries are bindings, which are not ordinary values. However, bindings deserve a different treatment anyway (#5125).

[Keno]

I can't think of any other instance where this might happen. The reason I didn't instead iterate of the jl_value_to_llvm is that I thought the iteration over backref was there in the first place to avoid serializing values that would not be referenced in the final system image in order to keep the size of the sysimg down if possible.

[JeffBezanson]

I guess that is handled by jl_serialize_globalvals.

[Keno]

Correct.

[Keno]

Updated to remove unnecessary changes.

[vtjnash]

I actually want to go with Jeff's idea in the long term. Anything in jl_value_to_llvm was JIT into the code somewhere, so we should make sure it has a reference.

However, as Jeff pointed out, this will require special handing of jl_binding_t objects, but that's needs to be fixed soon anyways. (It may also require some special handling of function pointers, but that's not really an issue either).

For now however, this is probably fine.

all the other ad-hoc calls to jl_serialize_gv

Those are actually already gone. My last commit before the branch was merged moved all of the serialization of global variable stuff to the end of the file.

[Keno]

Wouldn't the better solution there be to cut out the unreferenced IR rather than including unreferenced julia objects?

[Keno]

Third time's the charm.

[JeffBezanson]

It doesn't look like you actually return the placeholder value anywhere.

[Keno]

Doh! Fixed.