In Erlang 27 and earlier, the atoms in the AtU8 chunk are encoded as one byte of length, followed by the UTF-8 encoding of the atom text.

In Erlang 28, this is changed, so the length is either one or two bytes. The encoding of the length is a bit involved, and it seems this is due to reusing of a scheme originally introduced when the JIT was added, but it makes it possible to specify the length of the atom even when it consists of code-points that encode to several bytes in UTF-8. In Erlang 28, the id of the chunk is still AtU8 but the number of atoms is negative instead of positive.

I made some more changes:

The github action adds Erlang 27 and drops Erlang 23

I dropped the code to handle latin-1 encoded atoms when the chunk id is Atom. The AtU8 chunk for UTF-8 encoded atoms was introduced in Erlang 20. I don't know when Erlang dropped support for the old chunk, but in Erlang 27, support was dropped to read beam files from 24 and older, if I remember right.

Instead of constructing the chunks and the top-level form with updated chunk lengths and paddings, I changed to use beam_lib:build_module which has been around since Erlang 18. So the code now only needs to concern itself with the atom chunk. (As a side note, I found beam_lib functionality only for getting a list of atoms from the atom chunk, beam_lib:chunks(Beam,[atoms]), but no function to make a chunk from a list of atoms, so the code still needs to know the encoding of the atom chunk.)

Since the code now uses beam_lib to traverse chunks and put it back together to a module, there is no longer any big reason to document the beam file format. My intention was to replace it with a link to the documentation of it, but I could not find it anywhere. Eventually, I linked to the wayback machine.