[pkgimg] Split data from code

[vchuravy]

The goal is to be able to reuse a single .ji for multiple
code caches. Still work-in-progress currently this doesn't
produces usable code caches.

With this:

-rw-r--r-- 1 vchuravy vchuravy 1.4M Aug 26 14:11 ArgTools/aGHFV_akYLn.ji
-rwxr-xr-x 1 vchuravy vchuravy 221K Aug 26 14:10 ArgTools/aGHFV_akYLn.so*

Making it clear that the biggest cost remains the data.

[Keno]

Can we make the split files valid SO's with DT_NEEDED links? Not sure I like the idea of needing a specific side file with a separate lookup mechanism.

[vchuravy]

Can we make the split files valid SO's with DT_NEEDED links? Not sure I like the idea of needing a specific side file with a separate lookup mechanism.

I don't think so.

The idea is:

• .ji: "Plain old cache files, contains everything necessary to load this package"
• .so/.o: "Code cache file, only contains the code and some information necessary to load it."

Instead of what we have currently:

• .ji: "Only contains metadata for vailidty checks"
• .so Contains data and code, requiring us to re run the whol process just to get at O3

The reason for the split was that otherwise we would need to use an elf/coff/mach-o parser just to figure out if the cache file is valid, since dlclose doesn't work properly...

[Keno]

We control the layout of the shared objects - e.g. we can make the ji part a separate load command. We don't need a full parser, these data structures should be traversable in like 30-40 lines of code.