Marks and mark rings in GNU Emacs

I recently attempted to answer the question of whether experienced Emacs users should consider partially or fully disabling Transient Mark mode, which is (and should be) the default in modern GNU Emacs.

That blog post was meant to be as information-dense as I could make it, but now I’d like to describe the experience I have been having after switching to my custom pseudo-Transient Mark mode, which is labelled “mitigation #2” in my older post.

In summary: I feel like I’ve uncovered a whole editing paradigm lying just beneath the surface of the editor I’ve already been using for years. That is cool and enjoyable in itself, but I think it’s also helped me understand other design decisions about the basics of the Emacs UI better than before – in particular, the ideas behind how Emacs chooses where to display buffers, which were very frustrating to me in the past. I am now regularly using relatively obscure commands like C-x 4 C-o. I see it! It all makes sense now!

I would encourage everyone who has never used Emacs without Transient Mark mode to try turning it off for a while, either fully or partially, just to see what you can learn. It’s fascinating how it can come to seem more convenient and natural to pop the mark just to go back to the end of the current line after fixing up something earlier in the line, even though doing so requires pressing two modified keys instead of just C-e.

Eshell

I was amused to learn some years ago that someone was trying to make Emacs work as an X11 window manager. I was amazed and impressed to learn, more recently, that the project is still going and a fair number of people are using it. Kudos! I suspect that the basic motivation for such projects is that Emacs is a virtual Lisp machine, and it has a certain way of managing visible windows, and people would like to be able to bring both of those to their X11 window management.

However, I am beginning to suspect that the intrinsic properties of Emacs buffers are tightly connected to the ways in which Emacs manages visible windows, and the intrinsic properties of Emacs buffers are at least as fundamental as its status as a virtual Lisp machine. Thus I am not convinced by the idea of trying to use Emacs’ ways of handling visible windows to handle windows which do not contain Emacs buffers. (but it’s certainly nice to learn it’s working out for others)

The more general point is this. Emacs buffers are as fundamental to Emacs as anything else is, so it seems unlikely to be particularly fruitful to move something typically done outside of Emacs into Emacs, unless that activity fits naturally into an Emacs buffer or buffers. Being suited to run on a virtual Lisp machine is not enough.

What could be more suited to an Emacs buffer, however, than a typical Unix command shell session? By this I mean things like running commands which produce text output, and piping this output between commands and into and out of files. Typically the commands one enters are sort of like tiny programs in themselves, even if there are no pipes involved, because you have to spend time determining just what options to pass to achieve what you want. It is great to have all your input and output available as ordinary buffer text, navigable just like all your other Emacs buffers.

Full screen text user interfaces, like top(1), are not the sort of thing I have in mind here. These are suited to terminal emulators, and an Emacs buffer makes a poor terminal emulator – what you end up with is a sort of terminal emulator emulator. Emacs buffers and terminal emulators are just different things.

These sorts of thoughts lead one to Eshell, the Emacs Shell. Quoting from its documentation:

The shell’s role is to make [system] functionality accessible to the user in an unformed state. Very roughly, it associates kernel functionality with textual commands, allowing the user to interact with the operating system via linguistic constructs. Process invocation is perhaps the most significant form this takes, using the kernel’s fork' andexec’ functions.

…

Emacs is … a user application, but it does make the functionality of the kernel accessible through an interpreted language – namely, Lisp. For that reason, there is little preventing Emacs from serving the same role as a modern shell. It too can manipulate the kernel in an unpredetermined way to cause system changes. All it’s missing is the shell-ish linguistic model.

Eshell has been working very well for me for the past month or so, for, at least, Debian packaging work, which is very command shell-oriented (think tools like dch(1)).

The other respects in which Eshell is tightly integrated with the rest of Emacs are icing on the cake. In particular, Eshell can transparently operate on remote hosts, using TRAMP. So when I need to execute commands on Debian’s ftp-master server to process package removal requests, I just cd /ssh:fasolo: in Eshell. Emacs takes care of disconnecting and connecting to the server when needed – there is no need to maintain a fragile SSH connection and a shell process (or anything else) running on the remote end.

Or I can cd /ssh:athena\|sudo:root@athena: to run commands as root on the webserver hosting this blog, and, again, the text of the session survives on my laptop, and may be continued at my leisure, no matter whether athena reboots, or I shut my laptop and open it up again the next morning. And of course you can easily edit files on the remote host.

Update 31/Jan/2021: I now find myself mostly using M-!, M-&, C-x p ! and C-x p & to run shell commands. It would seem that the “shell-ish linguistic model” is less important to the work that I do than I thought it was. Thus, I was getting even less out of running bash in terminal emulators than I thought. I still reach for Eshell for Debian packaging work, I think because tools like dch assume the shell linguistic model.