I tried a bunch of stuff that all felt awful to use but this is what I've ended up with:

# This file is generated by ./gen-sources.sh.
# Do not edit!
{
  targetCommit = "1bc4296091c51f53a5598050c8956d16e945b0f5";
  files = {
    "sysa/gzip-1.2.4/files/stat_override.c" = "1216xn1536mfjyx9j1b6128ynb3mfdmy68h7y0n656sk9p5rpf75";
    "sysa/gzip-1.2.4/mk/main.mk" = "06c9xl13ym41i2q7rb370kc7affwxbbm6lyqpgbpj6q83bv4dhkq";
  };
}

& a default.nix that mapAttrs that into fetched files to be used like
cp ${live-bootstrap-files."sysa/gzip-1.2.4/mk/main.mk"} Makefile
or
cp ${live-bootstrap-files."sysa/${pname}-${version}/mk/main.mk"} Makefile

I'll clean it up and put it on a branch in the morning but any initial thoughts on the UX?

I considered trimming off the "sysa" part but in theory there could be collisions with stuff "sysc"

In reply to @emilytrau:matrix.org
but am i missing something or could you fetchurl a release tarball and use ungz and untar from mescc-tools-extra?

I tried pulling the whole repo but it was upset by the symlinks.
I'm guessing after gnutar and gnuzip we can do that but we need a couple files for those exact packages 😅

also pulling the whole of live-bootstrap is quite a big source where we'll only use a handful of files hopefully. IDK if the priority will be less bytes pulled in or less total separate derivations.
One nice thing about the separate files is it's very clear exactly which subset of files from live-bootstrap are inherited

I made some other ux changes to get a trimmed subset of files so you can do

lbf = live-bootstrap-files.getSubsetOfFiles "sysa/${pname}-${version}/";
# ...
cp ${lbf."mk/main.mk"} Makefile
cp ${lbf."files/stat_override.c"} some/place/file.c

It grabs live-bootstrap but then deletes most files

https://github.com/schierlm/FullSourceBootstrapFromGit/blob/main/part01.sh#L27-L28

specifically in this folder we see it uses config.h & mescc.scm which we just make ourselves so that's covered

pushed a copy. I noticed some of the files for coreutils come from a different commit so having some capability for multiple target commits would be good

https://github.com/06kellyjac/nixpkgs/commit/a106f98d9fb82a5640413289dd9c7f2dd7f5b6db

I might permanently move the files source of truth to nix & only fetch hashes for items that are empty e.g. "somefile.txt" = ""; and maybe look to parallelise it although I don't think we want to pull many many files that'd make that worth it