What is it?

It is my fork of Emacs, if you will.

This is Emacs' master branch, with stuff added (and some stuff removed, which I was too lazy to support).

I am using this on a daily basis on macOS with an init file that I can also use with an unchanged master. I make no efforts to find or fix bugs in stuff that I don't use personally.

My init file uses use-package and straight, and I am using packages like vertico, consult, corfu, magit, org, gnus etc., so I'd say it's pretty backwards compatible.

Who's that?

C-h i d m emacs RET s gerd RET

What's not there/not supported

• Documentation
• Tests

Common Lisp packages

What's there

You can use packages much like in Common Lisp. For details, read lisp/emacs-lisp/pkg.el and src/pkg.c, and maybe admin/cl-packages.org, although I haven't kept that up-to-date for some time, I guess.

What's the plan?

None. I don't think CL packages will land in Emacs in my lifetime. The resistance against this, or anything CL for that matter, is simply too high among current Emacs maintainers.

So, why do this? Because I can. Some people like to tinker with their init files, others go a step further :-).

New GC

This is an incremental and generational GC that is based on Ravenbrook MPS.

I ported this to mainline Emacs (branch feature/igc on savannah.gnu.org) and it was developed further there and here. At the moment the future of this branch is unclear.

Works well for me on macOS. I'm using this daily.

Child frames on ttys

This has meanwhile landed in master on savannah.

DTrace support

My Emacs is now beginning to be a DTrace USDT provider. Configure --with-dtrace=no to disable it. This is only tested on macOS. Development is ongoing in a very slow pace.

TTY menus in Lisp

This is an implementation of TTY menus that is entirely written in Emacs Lisp. See lisp/tm.el. That file can also be used with a current master from savannah. Use M-x tm-menu-mode to activate. See the doc string of that function. The menu's behavior is patterned after what macOS does.

It's unclear at the moment if that will land in GNU because I haven't offered it for inclusion, and reason for that is that I'm not interested in the inevitable discussions about the macOS menu behavior and whatnot.

Markers, Text indices, and marker vectors

Emacs internal text encoding is an extended UTF-8. Characters can be between 1 and 5 bytes long in this encoding. Such a variable-length representation requires a conversion between character and byte positions.

This position conversion is currently sped up in Emacs by consulting the doubly-linked list of markers that each buffer has. Each marker contains a character and a byte position. A heuristic is used to pick a suitable marker from whose known position one can scan forward or backward in the buffer text to convert character to byte positions and vice versa.

This has some problems:

• Adding a marker is O(1), but deleting a marker is O(N).

• Iteration over markers to update them when text is inserted/deleted accesses memory all over the place, the marker objects.

• A possibly large number of "cache marker" are produced to make it more likely that the heuristic finds suitable markers.

• The heuristic doesn't really work in some use cases.

What I've done here is:

• Add text indices: A text index is a data structure which supports such position conversions with predictable performance and without relying on markers and heuristics. The implementation can be found in text-index.c. Please read the comment at the start of that file for details.

• Add marker vectors: I had already changed the doubly-linked list of markers to use marker vectors in igc. This is now ported to the non-igc case, so that both can use a common implementation. (Please note that feature/igc still uses the old implementation. It will use the new one should this land in master.)

• Remove positions from markers: Store the character position of a marker in the marker vector and compute the byte position as needed using text indices. This allows position adjustments when text changes t be done by iterating over the marker vector without touching other memory.

Status

I am using this on a daily basis.

There is also a branch scratch/text-index on savannah where I ported this to master. If that lands in master or when is unclear. Stefan Monnier seems interested, and has run some benchmarks that look good. One notorious case was sped up by 2 orders of magnitude in an early version.

Emacs Mac Port

Now merged https://github.com/jdtsmith/emacs-mac, so you can build --with-mac as an alternative to with-ns.

I have ported igc to mac.