Since b1901 (rev 44a1a4a #4935), I've been unable to build llama.cpp with the provided Package.swift. I bisected back to this PR as the first breaking commit — b1900 (rev c918fe8) seems to work well.

I'm not terribly comfortable with C++, so to my eye I don't see anything wrong with the patch. I noticed the lack of GGML_API prefix like other types in the file, but I tried changing that locally and didn't notice a difference, though it's very possible I applied said change incorrectly/incompletely. EDIT: it seems committing this change indeed resolved the visibility issue I experienced as a result. Proposed a patch: #5030

I would be more than happy to share any other relevant information or test any proposed fixes to the extent that would be helpful!

Here's the full error log:

<module-includes>:1:9: note: in file included from <module-includes>:1:
#import "/Users/tito/code/llama.cpp/spm-headers/llama.h"
        ^
/Users/tito/code/llama.cpp/spm-headers/llama.h:235:9: error: unknown type name 'ggml_backend_sched_eval_callback'
        ggml_backend_sched_eval_callback cb_eval;