Geiser

1.1 Modus operandi

As already mentioned, Geiser relies on a running Scheme process to obtain the information it makes accessible to the programmer. There’s little effort, on the Elisp side, to understand, say, the module system used by the Scheme implementation at hand; instead, a generic interface between the two worlds is defined, and each supported Scheme includes a library implementing that API, together with some wee shims in Elisp allowing the reuse of the Emacs-side framework, which constitutes the bulk of the code.

While being as generic as possible, the Scheme-Elisp interface makes some assumptions about the capabilities and interaction mode of the corresponding REPL. In particular, Geiser expects the latter to support namespaces in the form of a module system, and to provide a well-defined way to establish the REPL’s current namespace (or module), as well as the current file’s module (or namespace). Thus, all evaluations performed by Geiser either in the REPL or in a source code buffer happen in the context of the current namespace. Every time you switch to a different file, you’re switching namespaces automatically; at the REPL, you must request the switch explicitly (usually just using means provided by the Scheme implementation itself).

If your favourite Scheme supports the above modus operandi, it has all that’s needed for a bare-bones Geiser mode. But Geiser can, and will, use any metadata available: procedure arities and argument lists to display interactive help, documentation strings, location information to jump to definitions, export lists to provide completion, and so on and so forth. Although this is not an all-or-none proposition (Geiser can operate with just part of that functionality available), i initially concentrated in supporting those Schemes with the richest (to my knowledge) introspection capabilities, namely, Guile and Racket. Later on, Dan Leslie added support for Chicken, and several other schemes followed suit.

1.2 Showing off

When working with a fully conniving Scheme, Geiser can offer the following functionality:

• Form evaluation in the context of the current file’s module.
• Macro expansion.
• File/module loading and/or compilation.
• Namespace-aware identifier completion (including local bindings, names visible in the current module, and module names).
• Autodoc: the echo area shows information about the signature of the procedure/macro around point automatically.
• Jump to definition of identifier at point.
• Access to documentation (including docstrings when the implementation provides it).
• Listings of identifiers exported by a given module.
• Listings of callers/callees of procedures.
• Rudimentary support for debugging (when the REPL provides a debugger) and error navigation.
• Support for multiple, simultaneous REPLs.
• Support for image display in those Schemes that treat them as first class values.

In the following pages, i’ll try to explain what these features actually are (i’m just swanking here), and how to use them for your profit. But, before that, let’s see how to install Geiser.

2.1 Must needs

If Geiser came with any guarantees, you’d break all of them by not using GNU Emacs 24.4 (or better: i regularly use it with a recent Emacs snapshot) and at least one of the supported Schemes, namely:

• Chez 9.4 or better, via geiser-chez
• Chibi 0.7.3 or better, via geiser-chibi
• Chicken 4.8.0 or better, via geiser-chicken
• Gambit 4.9.3 or better, via geiser-gambit
• Gauche 0.9.6 or better, via geiser-gauche
• Guile 2.2 or better, via geiser-guile
• GNU Kawa 3.1 or better, via geiser-kawa
• MIT/GNU Scheme 9.0 or better, via geiser-mit
• Racket 7.0 or better, via geiser-racket
• Stklos 1.50 or better, via geiser-stklos

Since Geiser supports multiple REPLs, having all of them will just add to the fun.

You’ll also need Geiser itself. The quickest installation is via your favourite implementation’s ELPA packages (as of this writing, they’re all available in MELPA and (most of them) also in NonGNU ELPA, which comes included with the batteries of Emacs 28 or better).

2.2 The quick and easy way

Did i mention that the easiest way of installing Geiser is using its ELPA package? If you’re using Emacs 24, ELPA is already there; for earlier versions, the page i just linked to twice will tell you where to find the goodies.

ELPA packages live in repositories accessible via HTTP. You can find Geiser’s package in either NonGNU ELPA or, if you like living on the bleeding edge, MELPA (directly from the git repo). To tell Emacs that an ELPA repo exists, you add it to package-archives¹:

(require 'package)

(add-to-list 'package-archives
  '("nongnu" . "https://elpa.nongnu.org/nongnu/"))

(package-initialize)

And then installing your favourite Geiser is as easy as (if, say, you’re a MIT aficionado):

M-x package-install RET geiser-mit RET

Rinse and repeat for each of the scheme implementations that you would like to use. Some of them (e.g. Gambit or Chicken) have a bit of additional setup, specific to them, so make sure you also check their respective package documentation.

With that, you are pretty much all set up. See The REPL to start using Geiser.

And, by the way, if you prefer to keep reading this manual within Emacs, C-h i m Geiser RET will bring you to the info version of it that you just installed!

2.3 Installing from source

All Geiser packages are ready to be used out of the box without much more ado. For the sake of concreteness, let’s assume you put its source in the directory ~/lisp/geiser. All you need to do is to add the following line to your Emacs initialisation file (be it ~/.emacs, ~/.emacs.d/init.el or any of its moral equivalents):

(add-to-list 'load-path "~/lisp/geiser/elisp")

and, if your, say, geiser-gambit checkout lives in ~/lisp/geiser-mit add to that:

(add-to-list 'load-path "~/lisp/geiser-gambit")

The autoloads defined in those packages should be enough to start scheming.

2.4 Friends

Although Geiser does not need them, it plays well with (and is enhanced by) the following Emacs packages:

• Paredit. Regardless of whether you use Geiser or not, you shouldn’t be coding in any Lisp dialect without the aid of Taylor Campbell’s structured editing mode.
• Company. Nikolaj Schumacher’s and Dmitry Gutov’s company-mode provides a generic front-end for completion engines (such as Geiser’s), with pretty and automatic completion lists.
• macrostep-geiser provides support for in-buffer macro expansion, using the macrostep package.
• ac-geiser If you prefer auto-complete-mode to company-mode, Xiao Hanyu’s ac-geiser, which provides a Geiser plugin for the popular Emacs Auto Completion Mode, is the package for you. Like Geiser, ac-geiser is available in MELPA, and also as an el-get package.

You just need to install and setup them as usual, for every package’s definition of usual. Geiser will notice their presence and react accordingly.

3.1 Starting the REPL

To start a Scheme REPL (meaning, a Scheme process offering you a Read-Eval-Print Loop), Geiser provides the generic interactive command geiser. If you invoke it (via, as is customary in Emacs, M-x geiser), you’ll be saluted by a prompt asking which one of the supported implementations you want to launch—yes, you can stop the asking, see below. Tabbing for completion will offer you, as of this writing, guile, racket, chicken, mit, chibi and chez. Just choose your poison, and a new REPL buffer will pop up (by default, the REPL will appear in a new window: if that annoys you, just set geiser-repl-use-other-window to nil and the current window will be used).

If all went according to plan, you’ll be facing an implementation-dependent banner, followed by an interactive prompt. Going according to plan includes having the executable of the Scheme you chose in your path. If that’s not the case, you can tell Emacs where it is, as described in a moment. Returning to our REPL, the first thing to notice is that the funny prompt is telling you your current module: its name is the part just after the @ sign (in Guile, that means guile-user, while Racket’s and Chicken’s top namespaces don’t have a name; cf. discussion in Switching context). Other than that, this is pretty much equivalent to having a command-line interpreter in a terminal, with a bunch of add-ons that we’ll be reviewing below. You can start typing sexps right there: Geiser will only dispatch them for evaluation when they’re complete, and will indent new lines properly until then. It will also keep track of your input, maintaining a history file that will be reloaded whenever you restart the REPL.

If you’re not happy with the faces Geiser is using for the REPL’s prompt and evaluated input, you can customise geiser-font-lock-repl-prompt and geiser-font-lock-repl-input to better-looking faces.

Connecting to an external Scheme

There’s an alternative way of starting a Geiser REPL: you can connect to an external Scheme process, provided it’s running a REPL server at some known port. How to make that happen depends on the Scheme implementation.

If you use Guile, you just need to start your Guile process (possibly outside Emacs) passing to it the flag --listen. This flag accepts an optional port as argument (as in --listen=1969), if you don’t want to use the default.

In Racket, you have to use the REPL server that comes with Geiser. To that end, put Geiser’s Racket scheme directory in Racket’s collection search path and invoke start-geiser (a procedure in the module geiser/server) somewhere in your program, passing it the desired port and, if desired, network interface name. This procedure will start the REPL server in a separate thread. For an example of how to do that, see the script bin/geiser-racket.sh in the source distribution, or, if you’ve compiled Geiser, bin/geiser-racket-noinst in the build directory, or, if you’ve installed Geiser, geiser-racket in <installation-prefix>/bin. These scripts start a new interactive Racket that is also running a REPL server (they also load the errortrace library to provide better diagnostics, but that’s not strictly needed).

With your external Scheme process running and serving, come back to Emacs and execute M-x geiser-connect, M-x connect-to-guile or M-x connect-to-racket. You’ll be asked for a host and a port, and, voila, you’ll have a Geiser REPL that is served by the remote Scheme process in a dedicated thread, meaning that your external program can go on doing whatever it was doing while you tinker with it from Emacs. Note, however, that all Scheme threads share the heap, so that you’ll be able to interact with those other threads in the running Scheme from Emacs in a variety of ways. For starters, all your (re)definitions will be visible everywhere. That’s dangerous, but will come in handy when you need to debug your running web server.

The connection between Emacs and the Scheme process goes over TCP, so it can be as remote as you need, perhaps with the intervention of an SSH tunnel.

3.2 First aids

A quick way of seeing what else Geiser’s REPL can do for you, is to display the corresponding entry up there in your menu bar. No, i don’t normally use menus either; but they can come in handy until you’ve memorized Geiser’s commands, as a learning device. And yes, i usually run Emacs inside a terminal, but one can always use La Carte to access the menus in a convenient enough fashion.

Or just press C-h m and be done with that.

Among the commands at your disposal, we find the familiar input navigation keys, with a couple twists. By default, M-p and M-n are bound to matching items in your input history. That is, they’ll find the previous or next sexp that starts with the current input prefix (defined as the text between the end of the prompt and your current position, a.k.a. point, in the buffer). For going up and down the list unconditionally, just use C-c M-p and C-c M-n. In addition, navigation is sexp-based rather than line-based.

There are also a few commands to twiddle with the Scheme process. C-c C-q will gently ask it to quit, while C-u C-c C-q will mercilessly kill the process (but not before stowing your history in the file system). Unless you’re using a remote REPL, that is, in which case both commands will just sever the connection and leave the remote process alone. If worse comes to worst and the process is dead, C-c C-z will restart it. However, the same shortcut, issued when the REPL is alive, will bring you back to the buffer you came from, as explained in this section.

The remaining commands are meatier, and deserve sections of their own.

3.3 Switching context

In tune with Geiser’s modus operandi, evaluations in the REPL take place in the namespace of the current module. As noted above, the REPL’s prompt tells you the name of the current module. To switch to a different one, you can use the command geiser-repl-switch-to-module, bound to C-c C-m. You’ll notice that Geiser simply uses a couple of meta-commands provided by the Scheme REPL (the stock ,m in Guile and Chicken and the (geiser-defined) ,enter in Racket), and that it doesn’t even try to hide that fact. That means that you can freely use said native ways directly at the REPL, and Geiser will be happy to oblige. In Racket, ,enter works like Racket’s standard enter! form, but you can also provide a path string as its argument (e.g., ,enter "/tmp/foo.rkt" is equivalent to ,enter (file "/tmp/foo.rkt")). Like enter!, ,enter accepts also module names (as in, say, ,enter geiser/main). As mentioned, in Guile and Chicken, ,m is used as is.

Once you enter a new module, only those bindings visible in its namespace will be available to your evaluations. All Schemes supported by Geiser provide a way to import new modules in the current namespace. Again, there’s a Geiser command, geiser-repl-import-module, to invoke such functionality, bound this time to C-c C-i. And, again, you’ll see Geiser just introducing the native incantation for you, and you’re free to use such incantations by hand whenever you want.

One convenience provided by these two Geiser commands is that completion is available when introducing the new module name, using the TAB key. Pressing it at the command’s prompt will offer you a prefix-aware list of available module names.

Which brings me to the next group of REPL commands.

3.4 Completion and error handling

We’ve already seen Geiser completion of module names in action at the minibuffer. You won’t be surprised to know that it’s also available at the REPL buffer itself. There, you can use either C-. or M-` to complete module names, and TAB or M-TAB to complete identifiers. Geiser will know what identifiers are bound in the current module and show you a list of those starting with the prefix at point. Needless to say, this is not a static list, and it will grow as you define or import new bindings in the namespace at hand. If no completion is found, TAB will try to complete the prefix after point as a module name.

REPL buffers use Emacs’ compilation mode to highlight errors reported by the Scheme interpreter, and you can use the next-error command (M-g n) to jump to their location. By default, every time you enter a new expression for evaluation old error messages are forgotten, so that M-g n will always jump to errors related to the last evaluation request, if any. If you prefer a not-so-forgetful REPL, set the customization variable geiser-repl-forget-old-errors-p to nil. Note, however, that even when that variable is left as t, you can always jump to an old error by moving to its line at the REPL and pressing RET. When your cursor is away from the last prompt, TAB will move to the next error in the buffer, and you can use BACKTAB everywhere to go to the previous one.

Caveat about completion & the REPL

It is possible for Geiser to hang your Emacs process when trying to complete symbols. This can happen in the REPL itself or even in a Scheme buffer that is attached to the REPL process. If this happens, you’ve probably entered a module that changes the REPL prompt from what Geiser was expecting to see.

Unfortunately, there’s no general solution for this issue right now (as it is a daunting task to try to make a regexp that can encompass all possible REPL prompts). The best solution for now is to fix this issue on a case-by-case basis by adjusting your prompt regexp variable so that it matches the default prompt as well as your Scheme module’s special prompt.

For example, XREPL is a Racket module that implements a better Racket REPL. You might be interested in toying around with some of its functions, but when you try to enter XREPL via, say, C-c C-m xrepl, you’ll notice that the REPL prompt has changed to something like this:

<pkgs>/xrepl-lib/xrepl/main>

If you start typing symbols, and then you try to auto-complete those symbols, your Emacs process may hang. This is because Geiser expects the REPL prompt to match this regexp (for Racket):

"\\(mzscheme\\|racket\\)@[^ ]*> "

Therefore, we can fix this issue by changing our default prompt regexp like so:

(setq geiser-racket--prompt-regexp "<pkgs>.*> \\|\\(mzscheme\\|racket\\)@[^ ]*> ")

Note that you may have to run M-x geiser-reload after setting this variable so that your changes will take effect.

Again, you’ll have to change the regexp to fit every prompt that causes this issue, but the only alternative (that we can think of right now) is to create a regexp that will match every possible prompt. Obviously, that is going to be more than a little tricky. However, if you have a better solution than that, please share it with the Geiser developers; we’ll be more than happy to hear it.

3.5 Autodoc and friends

Oftentimes, there’s more you’ll want to know about an identifier besides its name: What module does it belong to? Is it a procedure and, if so, what arguments does it take? Geiser tries to help you answering those questions too.

Actually, if you’ve been playing with the REPL as you read, you might have notice some frantic activity taking place in the echo area every now and then. That was Geiser trying to be helpful (while, hopefully, not being clippy), or, more concretely, what i call, for want of a better name, its autodoc mode. Whenever it’s active (did you notice that A in the mode-line?), Geiser’s gerbils will be scanning what you type and showing (unless you silence them with C-c C-d C-a) information about the identifier nearest to point.

If that identifier corresponds to a variable visible in the current namespace, you’ll see the module it belongs to and its value. For procedures and macros, autodoc will display, instead of their value, the argument names (or an underscore if Geiser cannot determine the name used in the definition). Optional arguments are surrounded by parentheses. When the optional argument has a default value, it’s represented by a list made up of its name and that value. When the argument is a keyword argument, its name has “#:” as a prefix.

If that’s not enough documentation for you, C-c C-d d will open a separate documentation buffer with help on the symbol at point. This buffer will contain implementation-specific information about the identifier (e.g., its docstring for Guile, or its contract, if any, for Racket), and a handy button to open the corresponding manual entry for the symbol, which will open an HTML page (for Racket and Chicken) or the texinfo manual (for Guile). If you’d rather go directly to the manual, try C-c C-d i, which invokes geiser-doc-look-up-manual as the handy button does.

Geiser can also produce for you a list, classified by kind, of the identifiers exported by a given module: all you need to do is press C-c C-d m, and type or complete the desired module’s name.

The list of exported bindings is shown, again, in a buffer belonging to Geiser’s documentation browser, where you have at your disposal a bunch of navigation commands listed in our cheat-sheet.

We’ll have a bit more to say about the documentation browser in a later section.

If that’s still not enough, Geiser can jump, via M-., to the symbol’s definition. A buffer with the corresponding file will pop up, with its point resting upon the identifier’s defining form. When you’re done inspecting, M-, will bring you back to where you were. As we will see, these commands are also available in Scheme buffers. M-. also works for modules: if your point is on an unambiguous module name, the file where it’s defined will be opened for you.

3.6 Seeing is believing

In schemes that support images as values (currently, that means Racket), the REPL will display them inline if you’re using them in a graphics-aware Emacs.

For the terminal, images will appear as buttons: press return on them to invoke an external viewer (configurable via geiser-image-viewer) that will show you the image at hand. You can also ask for the same behaviour on all emacsen by customising geiser-repl-inline-images-p to nil.

Geiser keeps a cache of the last displayed images in the directory geiser-image-cache-dir, which defaults to the system’s temp directory, with up to geiser-image-cache-keep-last files. You can invoke the external image viewer on any of them with M-x geiser-view-last-image, which takes a prefix argument to indicate which image number you want, 0 corresponding to the newest one.

3.7 Customization and tips

The looks and ways of the REPL can be fine-tuned via a bunch of customization variables. You can see and modify them all in the corresponding customization group (by using the menu entry or the good old M-x customize-group geiser-repl), or by setting them in your Emacs initialisation files (as a rule, all knobs in Geiser are tunable this way: you don’t need to use customization buffers if you don’t like them).

I’m documenting below a proper subset of those settings, together with some related tips.

Choosing a Scheme implementation

Instead of using the generic geiser command, you can directly start your Scheme of choice using any of the following commands:

• run-racket
• run-guile
• run-chicken
• run-mit
• run-chibi
• run-chez

In addition, the variable geiser-active-implementations contains a list of those Schemes Geiser should be aware of. Thus, if you happen to be, say, a racketeer not to be beguiled by other schemes, you can tell Geiser to forget about the richness of the Scheme ecosystem with something like:

(setq geiser-active-implementations '(racket))

in your initialisation files.

When starting a new REPL, Geiser assumes, by default, that the corresponding Scheme binary is in your path. If that’s not the case, the variables to tweak are (depending on which Scheme you choose):

• geiser-guile-binary
• geiser-racket-binary
• geiser-chicken-binary
• geiser-mit-binary
• geiser-chibi-binary
• geiser-chez-binary

They should be set to a string with the full path to the requisite binary.

Before starting the REPL, Geiser will check whether the version of your Scheme interpreter is good enough. This means that it will spend a couple tenths of a second launching and quickly discarding a Scheme process, but also that the error message you’ll get if you’re on the wrong Scheme version will be much more informative. If you one to avoid version checks, just check geiser-repl-skip-version-check-p to t in your configuration.

Init files and load paths

The startup behaviour of the REPL can be also fine tuned with a couple more initialisation parameters.

Many Scheme implementations provide a configuration variable to specify a Geiser-specific init file (e.g., geiser-guile-init-file for Guile), and, sometimes a global list of paths to add to the interpreter’s load path (that’d be geiser-guile-load-path for Guile).

There is also a generic mechanism to specify how to add directories to the initial load path when geiser-repl-current-project-function is set: you can then customize geiser-repl-add-project-paths to a list of subdirectories of the project’s root to add to the load path. When this option is set, the working directory of the REPL’s buffer (i.e., the value of the elisp variable default-directory) will be set to the directory returned by geiser-repl-current-project-function).

These variables controlling your scheme’s initialisation process are good candidates for an entry in a project’s .dir-locals.el file, so that they are automatically set to a sensible value whenever you start a REPL in the project’s directory.

Startup waiting time

When starting a scheme implementation in old or very busy computers, Geiser might have to wait a bit more than it expects (which is ten seconds, or ten thousand milliseconds, by default). If you find that Geiser is giving up too quickly and complaining that no prompt was found, try to increase the value of geiser-repl-startup-time to, say, twenty seconds:

(setq geiser-repl-startup-time 20000)

If you prefer, you can use the customize interface to, well, customise the above variable’s value.

History

By default, Geiser won’t record duplicates in your input history. If you prefer it did, just set geiser-repl-history-no-dups-p to nil. History entries are persistent across REPL sessions: they’re saved in implementation-specific files whose location is controlled by the variable geiser-repl-history-filename. For example, my Geiser configuration includes the following line:

(setq geiser-repl-history-filename "~/.emacs.d/geiser-history")

which makes the files geiser-history.guile and geiser-history.racket to live inside my home’s .emacs.d directory.

Autodoc

If you happen to love peace and quiet and prefer to keep your REPL’s echo area free from autodoc’s noise, geiser-repl-autodoc-p is the customization variable for you: set it to nil and autodoc will be disabled by default in new REPLs. You can always bring the fairies back, on a per-REPL basis, using C-c C-d C-a.

Remote connections

When using any of the connection commands (e.g. geiser-connect, connect-to-guile, connect-to-racket, etc.) you’ll be prompted for a host and a port, defaulting to “localhost” and 37146. You can change those defaults customizing geiser-repl-default-host and geiser-repl-default-port, respectively.

Killing REPLs

If you don’t want Emacs to ask for confirmation when you’re about to kill a live REPL buffer (as will happen, for instance, if you’re exiting Emacs before closing all your REPLs), you can set the flag geiser-repl-query-on-kill-p to nil. On a related note, the customizable variable geiser-repl-query-on-exit-p controls whether Geiser should ask for confirmation when you exit the REPL explicitly (via, say, C-c C-q, as opposed to killing the buffer), and is set to nil by default.

4.1 Activating Geiser

With Geiser installed following any of the procedures described in The quick and easy way or Installing from source, Emacs will automatically activate geiser-mode when opening a Scheme buffer. Geiser also instructs Emacs to consider files with the extension rkt part of the family, so that, in principle, there’s nothing you need to do to ensure that Geiser’s extensions will be available, out of the box, when you start editing Scheme code.

Indications that everything is working according to plan include the ’Geiser’ minor mode indicator in your mode-line and the appearance of a new entry for Geiser in the menu bar. If, moreover, the mode-line indicator is the name of a Scheme implementation, you’re indeed in a perfect world; otherwise, don’t despair and keep on reading: i’ll tell you how to fix that in a moment.

The menu provides a good synopsis of everything Geiser brings to the party, including those keyboard shortcuts we Emacsers love. If you’re seeing the name of your favourite Scheme implementation in the mode-line, have a running REPL and are comfortable with Emacs, you can stop reading now and, instead, discover Geiser’s joys by yourself. I’ve tried to make Geiser as self-documenting as any self-respecting Emacs package should be. If you follow this route, make sure to take a look at Geiser’s customization buffers (M-x customize-group RET geiser): there’s lot of fine-tuning available there. You might also want to take a glance at our cheat sheet.

Since geiser-mode is a minor mode, you can toggle it with M-x geiser-mode, and control its activation in hooks with the functions turn-on-geiser-mode and turn-off-geiser-mode. If, for some reason i cannot fathom, you prefer geiser-mode not to be active by default, customizing geiser-mode-auto-p to nil will do the trick.

And if you happen to use a funky extension for your Scheme files that is not recognised as such by Emacs, just tell her about it with:

(add-to-list 'auto-mode-alist '("\\.funky-extension\\'" . scheme-mode))

Now, geiser-mode is just a useless wretch unless there’s a running Scheme process backing it up. Meaning that virtually all the commands it provides require a REPL up and running, preferably corresponding to the correct Scheme implementation. In the following section, we’ll see how to make sure that that’s actually the case.

4.2 The source and the REPL

As i’ve already mentioned a couple of times, geiser-mode needs a running REPL to be operative. Thus, a common usage pattern will be for you to first call geiser (or one of its variants), and then open some Scheme files; but there’s nothing wrong in first opening a couple Scheme buffers and then starting the REPL (you can even find it more convenient, since pressing C-c C-z in a Scheme buffer will start the REPL for you). Since Geiser supports more than one Scheme implementation, though, there’s the problem of knowing which of them is to be associated with each Scheme source file. Serviceable as it is, geiser-mode will try to guess the correct implementation for you, according to the algorithm described below. If you find that Geiser is already guessing right the Scheme implementation, feel free to skip to the next subsection.

How Geiser associates a REPL to your Scheme buffer

To determine what Scheme implementation corresponds to a given source file, Geiser uses the following algorithm:

1. If the file-local variable geiser-scheme-implementation is defined, its value is used. A common way of setting buffer-local variables is to put them in a comment near the beginning of the file, surrounded by -*- marks, as in:

   ;; -*- geiser-scheme-implementation: guile -*-
   

2. If you’ve customized geiser-active-implementations so that it’s a single-element list (as explained in here), that element is used as the chosen implementation.
3. The contents of the file is scanned for hints on its associated implementation. For instance, files that contain a #lang directive will be considered Racket source code, while those with a define-module form in them will be assigned to a Guile REPL.
4. The current buffer’s file name is checked against the rules given in geiser-implementations-alist, and the first match is applied. You can provide your own rules by customizing this variable, as explained below.
5. If we haven’t been lucky this far and you have customized geiser-default-implementation to the name of a supported implementation, we’ll follow your lead.
6. See? That’s the problem of being a smart aleck: one’s always outsmarted by people around. At this point, geiser-mode will humbly give up and ask you to explicitly choose the Scheme implementation.

As you can see in the list above, there are several ways to influence Geiser’s guessing by means of customizable variables. The most direct (and most impoverishing) is probably limiting the active implementations to a single one, while customizing geiser-implementations-alist is the most flexible (and, unsurprisingly, also the most complex). Here’s the default value for the latter variable:

(((regexp "\\.scm$") guile)
 ((regexp "\\.ss$") racket)
 ((regexp "\\.rkt$") racket))

which describes the simple heuristic that files with .scm as extension are by default associated to a Guile REPL while those ending in .ss or .rkt correspond to Racket’s implementation (with the caveat that these rules are applied only if the previous heuristics have failed to detect the correct implementation, and that they’ll match only if the corresponding implementation is active). You can add rules to geiser-implementations-alist (or replace all of them) by customizing it. Besides regular expressions, you can also use a directory name; for instance, the following snippet:

(eval-after-load "geiser-impl"
  '(add-to-list 'geiser-implementations-alist
                '((dir "/home/jao/prj/frob") guile)))

will add a new rule that says that any file inside my /home/jao/prj/frob directory (or, recursively, any of its children) is to be assigned to Guile. Since rules are first matched, first served, this new rule will take precedence over the default ones.

A final tip: if you want Geiser to start automatically a REPL for you if it notices that there’s no one active when it enters geiser-mode, you can customize geiser-mode-start-repl-p to t.

Managing multiple scheme projects

By default, Geiser will re-use a single REPL for all buffers sharing the same scheme implementation. This works well enough in many cases, but may become problematic (or at least annoying) when working on multiple projects with separate dependencies and include paths.

Geiser provides optional support for using separate REPLs for each project, which can be enabled by customizing geiser-repl-current-project-function and selecting your Emacs project-management library of choice (eg. project.el or projectile). With this configured, if you want new REPLs to automatically associate themselves with the current project, so that all Geiser commands will ignore REPLs that are not associated with the project, customize the toggle geiser-repl-per-project-p to t and you’re all set up.

This can be very convenient when used with a .dir-locals.el in the project root to set include paths, ensuring that Geiser REPLs will always know where to find your project’s modules or dependencies. Geiser automatically handles the common case of the project root belonging to the load path: unless you tell it otherwise (using the customisable flag geiser-repl-add-project-path-p, which defaults to t), it will add the result of calling geiser-repl-current-project-function to the REPLs load path on startup.

Switching between source files and the REPL

Once you have a working geiser-mode, you can switch from Scheme source buffers to the REPL or C-c C-z. Those shortcuts map to the interactive command geiser-repl-switch.

If you use a numeric prefix, as in C-u C-c C-z, besides being teleported to the REPL, the latter will switch to the namespace of the Scheme source file, as if you had used C-c C-m in the REPL, with the source file’s module as argument; cf. discussion in Switching context. This command is also bound to C-c C-a.

Once you’re in the REPL, the same C-c C-z shortcut will bring you back to the buffer you jumped from, provided you don’t kill the Scheme process in between. This is why the command is called geiser-repl-switch instead of switch-to-repl, and what makes it really handy, if you ask me.

If for some reason you’re not happy with the Scheme implementation that Geiser has assigned to your file, you can change it with C-c C-s, and you probably should take a look at the previous subsection to make sure that Geiser doesn’t get confused again.

A note about context

As explained before (see Modus operandi), all Geiser activities take place in the context of the current namespace, which, for Scheme buffers, corresponds to the module that the Scheme implementation associates to the source file at hand (for instance, in Racket, there’s a one-to-one correspondence between paths and modules, while Guile relies on explicit define-module forms in the source file).

Now that we have geiser-mode happily alive in our Scheme buffers and communicating with the right REPL instance, let us see what it can do for us, besides jumping to and fro.

4.3 Documentation helpers

Autodoc redux

The first thing you will notice by moving around Scheme source is that, every now and then, the echo area lights up with the same autodoc messages we know and love from our REPL forays. This happens every time the Scheme process is able to recognise an identifier in the buffer, and provide information either on its value (for variables) or on its arity and the name of its formal arguments (for procedures and macros). That information will only be available if the module the identifier belongs to has been loaded in the running Scheme image. So it can be the case that, at first, no autodoc is shown for identifiers defined in the file you’re editing. But as soon as you evaluate them (either individually or collectively using any of the devices described in To eval or not to eval) their signatures will start appearing in the echo area.

Autodoc activation is controlled by a minor mode, geiser-autodoc, which you can toggle with M-x geiser-autodoc-mode, or its associated keyboard shortcut, C-c C-d a. That /A indicator in the mode-line is telling you that autodoc is active. If you prefer that it be inactive by default (e.g., because you’re connecting to a really remote scheme and want to minimize network exchanges), just set geiser-mode-autodoc-p to nil in your customization files. Even when autodoc mode is off, you can use geiser-autodoc-show, bound by default to C-c C-d s, to show the autodoc string for the symbol at point.

The way autodoc displays information deserves some explanation. It will first show the name of the module where the identifier at hand is defined, followed by a colon and the identifier itself. If the latter corresponds to a procedure or macro, it will be followed by a list of argument names, starting with the ones that are required. Then there comes a list of optional arguments, if any, enclosed in parentheses. When an optional argument has a default value (or a form defining its default value), autodoc will display it after the argument name. When the optional arguments are keywords, their names are prefixed with “#:” (i.e., their names are keywords). An ellipsis (…) serves as a marker of an indeterminate number of parameters, as is the case with rest arguments or when autodoc cannot fathom the exact number of arguments (this is often the case with macros defined using syntax-case). Another way in which autodoc displays its ignorance is by using an underscore to display parameters whose name is beyond its powers.

It can also be the case that a function or macro has more than one signature (e.g., functions defined using case-lambda, or some syntax-rules macros, for which Geiser has often the black magic necessary to retrieve their actual arities). In those cases, autodoc shows all known signatures (using the above rules for each one) separated by a vertical bar (|).

As you have already noticed, the whole autodoc message is enclosed in parentheses. After all, we’re talking about Scheme here.

Finally, life is much easier when your cursor is on a symbol corresponding to a plain variable: you’ll see in the echo area its name, preceded by the module where it’s defined, and followed by its value, with an intervening arrow for greater effect. This time, there are no enclosing parentheses (i hope you see the logic in my madness).

You can change the way Geiser displays the module/identifier combo by customizing geiser-autodoc-identifier-format. For example, if you wanted a tilde surrounded by spaces instead of a colon as a separator, you would write something like:

(setq geiser-autodoc-identifier-format "%s ~ %s")

in your Emacs initialisation files. There’s also a face (geiser-font-lock-autodoc-identifier) that you can customize (for instance, with M-x customize-face) to change the appearance of the text. And another one (geiser-font-lock-autodoc-current-arg) that controls how the current argument position is highlighted.

Other documentation commands

Sometimes, autodoc won’t provide enough information for you to understand what a function does. In those cases, you can ask Geiser to ask the running Scheme for further information on a given identifier or module.

For symbols, the incantation is M-x geiser-doc-symbol-at-point, or C-c C-d C-d for short. If the associated Scheme supports docstrings (as, for instance, Guile does), you’ll be teleported to a new Emacs buffer displaying Geiser’s documentation browser, filled with information about the identifier, including its docstring (if any; unfortunately, that an implementation supports docstrings doesn’t mean that they’re used everywhere).

Pressing q in the documentation buffer will bring you back, enlightened, to where you were. There’s also a handful of other navigation commands available in that buffer, which you can discover by means of its menu or via the good old C-h m command. And feel free to use the navigation buttons and hyperlinks that justify my calling this buffer a documentation browser.

For Racket, which does not support docstrings out of the box, this command will provide less information, but the documentation browser will display the corresponding contract when it’s available, as well as some other tidbits for re-exported identifiers.

You can also ask Geiser to display information about a module, in the form of a list of its exported identifiers, using C-c C-d C-m, exactly as you would do in the REPL.

In both cases, the documentation browser will show a couple of buttons giving you access to further documentation. First, you’ll see a button named source: pressing it you’ll jump to the symbol’s definition. The second button, dubbed manual, will open the Scheme implementation’s manual page for the symbol at hand. For Racket, that will open your web browser displaying the corresponding reference’s page (using the HTML browser in Racket’s configuration, which you can edit in DrRacket’s preferences dialog, or by setting plt:framework-pref:external-browser directly in ~/.racket/racket-prefs.rktd), while in Guile a lookup will be performed in the texinfo manual.

For Guile, the manual lookup uses the info indexes in the standard Guile info nodes, which are usually named “guile” or “guile-2.0”. If yours are named differently, just add your name to the customizable variable geiser-guile-manual-lookup-nodes.

A list of all navigation commands in the documentation browser is available in our cheat-sheet.

You can also skip the documentation browser and jump directly to the manual page for the symbol at point with the command geiser-doc-look-up-manual, bound to C-c C-d i.

4.4 To eval or not to eval

One of Geiser’s main goals is to facilitate incremental development. You might have noticed that i’ve made a big fuss of Geiser’s ability to recognize context, by dint of being aware of the namespace where its operations happen.

That awareness is especially important when evaluating code in your scheme buffers, using the commands described below. They allow you to send code to the running Scheme with a granularity ranging from whole files to single s-expressions. That code will be evaluated in the module associated with the file you’re editing, allowing you to redefine values and procedures to your heart’s (and other modules’) content.

Macros are, of course, another kettle of fish: one needs to re-evaluate uses of a macro after redefining it. That’s not a limitation imposed by Geiser, but a consequence of how macros work in Scheme (and other Lisps). There’s also the risk that you lose track of what’s actually defined and what’s not during a given session. But, in my opinion, those are limitations we lispers are aware of, and they don’t force us to throw the baby with the bathwater and ditch incremental evaluation. Some people disagree; if you happen to find their arguments convincing, you don’t have to throw away Geiser together with the baby: M-x geiser-restart-repl will let you restart the REPL as many times as you see fit. Moreover, you can invoke geiser-compile-current-buffer and geiser-load-current-buffer with a prefix argument (that’d be something like C-u C-c C-k for compilation, for instance), to tell Geiser to restart the REPL associated with a buffer before compiling or loading its current contents.

For all of you auld bearded lispers still with me, here are some of the commands performing incremental evaluation in Geiser.

geiser-eval-last-sexp, bound to C-x C-e, will eval the s-expression just before point. If you use a prefix, as in C-u C-x C-e, besides evaluating it the expression is inserted in the the buffer.

geiser-eval-definition, bound to C-M-x, finds the topmost definition containing point and sends it for evaluation. The variant geiser-eval-definition-and-go (C-c M-e) works in the same way, but it also teleports you to REPL after the evaluation.

geiser-eval-region, bound to C-c C-r, evals the current region. Again, there’s an and-go version available, geiser-eval-region-and-go, bound to C-c M-r. And, if you want to extend the evaluated region to the whole buffer, there is geiser-eval-buffer, bound to C-c C-b and its companion geiser-eval-buffer-and-go, bound to C-c M-b.

For all the commands above, the result of the evaluation is displayed in the minibuffer, unless it causes a (Scheme-side) error (see To err: perchance to debug), or, for schemes supporting them (such as Racket), the evaluation yields an image, in which case you’ll see it in popping up in the Geiser debug buffer (if your Emacs runs under the auspices of a graphical toolkit), or via an external viewer if you set geiser-image-viewer to the path of an appropriate visualization program (see also Seeing is believing for more on image support).

At the risk of repeating myself, i’ll remind you that all these evaluations will take place in the namespace of the module corresponding to the Scheme file from which you’re sending your code, which, in general, will be different from the REPL’s current module. And, if all goes according to plan, (re)defined variables and procedures should be immediately visible inside and, if exported, outside their module.

Besides evaluating expressions, definitions and regions, you can also macro-expand them. The corresponding key bindings start with the prefix C-c C-m and end, respectively, with C-e, C-x and C-r. The result of the macro expansion always appears in a pop up buffer.

All the evaluations and expansions performed by the commands above are asynchronous², so that you can move around while the answer is being computed. The command geiser-eval-interrupt, bound to C-c C-i will interrupt any on-going evaluation and, when the scheme implementation supports a debugger, bring you to a buffer where you can perform buffer actions in the interrupted evaluation’s context.

Oh, didn’t i mention we have support for debuggers? Let’s talk about that next.

4.5 To err: perchance to debug

When an error occurs during evaluation, it will be reported according to the capabilities of the underlying Scheme REPL.

In most schemes, you’ll be presented with a backtrace, in a new buffer where file paths locating the origin of the error are click-able (you can navigate them using the TAB key, and use RET or the mouse to jump to the offending spot; or invoke Emacs’ stock commands next-error and previous-error, bound to M-g n and M-g p by default).

By default, Geiser will tele-transport your pointer to the debug buffer: if you prefer to stay in the source buffer, set geiser-debug-jump-to-debug to nil.

For schemes with good debug support (Guile is one), the debug buffers offer a debugging menu, accessible via the , (that’s a comma) key. If you press it, a transient menu will appear, offering you a variety of actions, including showing local variable values or a more detailed backtrace or frame display. This is the same interface you’ll encounter the in case of interrupted evaluations, either by your explicit C-c C-i command or because a breakpoint has been previously set.

In addition, Geiser will sometimes report warnings for otherwise successful evaluations. In those cases, it won’t enter the debugger, just report the warnings in a debug buffer.

4.6 Jumping around

This one feature is as sweet as it is easy to explain: M-. (geiser-edit-symbol-at-point) will open the file where the identifier around point is defined and land your point on its definition. To return to where you were, press M-, (geiser-pop-symbol-stack). This command works also for module names: Geiser first tries to locate a definition for the identifier at point and, if that fails, a module with that name; if the latter succeeds, the file where the module is defined will pop up.

Sometimes, the underlying Scheme will tell Geiser only the file where the symbol is defined, but Geiser will use some heuristics (read, regular expressions) to locate the exact line and bring you there. Thus, if you find Geiser systematically missing your definitions, send a message to the mailing list, and we’ll try to make the algorithm smarter.

You can control how the destination buffer pops up by setting geiser-edit-symbol-method to either nil (to open the file in the current window), 'window (other window in the same frame) or 'frame (in a new frame).

4.7 Geiser writes for you

No self-respecting programming mode would be complete without completion. In geiser-mode, identifier completion is bound to M-TAB, and will offer all visible identifiers starting with the prefix before point. Visible here means all symbols imported or defined in the current namespace plus locally bound ones. E.g., if you’re at the end of the following partial expression:

(let ((default 42))
  (frob def

and press M-TAB, one of the possible completions will be default.

After obtaining the list of completions from the running Scheme, Geiser uses the standard Emacs completion machinery to display them. That means, among other things, that partial completion is available: just try to complete d-s or w-o-t-s to see why this is a good thing. Partial completion won’t work if you have disabled it globally in your Emacs configuration: if you don’t know what i’m talking about, never mind: Geiser’s partial completion will work for you out of the box.

If you find the M modifier annoying, you always have the option to activate geiser-smart-tab-mode, which will make the TAB key double duty as the regular Emacs indentation command (when the cursor is not near a symbol) and Geiser’s completion function. If you want this smarty pants mode always on in Scheme buffers, customize geiser-mode-smart-tab-p to t.

Geiser also knows how to complete module names: if no completion for the prefix at point is found among the currently visible bindings, it will try to find a module name that matches it. You can also request explicitly completion only over module names using M-` (that’s a backtick).

Besides completion, there’s also this little command, geiser-squarify, which will toggle the delimiters of the innermost list around point between round and square brackets. It is bound to C-c C-e [. With a numeric prefix (as in, say, M-2 C-c C-e [), it will perform that many toggles, forward for positive values and backward for negative ones.

Caveat about completion

It is possible for Geiser to hang your Emacs process when trying to complete symbols. This can happen in the REPL itself or even in a Scheme buffer that is attached to the REPL process. For more details on how to fix this problem, Caveat about completion & the REPL

5.1 Scheme buffers

5.2 REPL

5.3 Documentation browser