Skip to content

The little hot-reloadin' static site maker from shell.


Notifications You must be signed in to change notification settings


Folders and files

Last commit message
Last commit date

Latest commit


Repository files navigation


The little hot-reloadin' static site generator from shell. Assumes Bash 4.4+.

WARNING: Here be yaks!

shite's job is to help me make my website: Thus, shite's scope, (mis)feature set, polish will always be production-grade, where production is "works on my machine(s)" :)

much write. such Bash. very hotreload. wow.

Table of Contents


Well, shite aims to make websites.

  • It is a wee publishing system made of pipelined workflows, optionally driven by streams of file events (for the hotreloadin' bits).

  • It will not surprise a Perl/PHP gentleperson hacker from the last century.

  • It exists because one whistles silly tunes and shaves yaks.

This is baaasically what it does (ref: the shite_templating_publish_sources function).

cat "${watch_dir}/sources/${url_slug}" |
    __shite_templating_compile_source_to_html ${file_type} |
    __shite_templating_wrap_content_html ${content_type} |
    __shite_templating_wrap_page_html |
    ${html_formatter_fn} |
    tee "${watch_dir}/public/${slug}.html"
  • It publishes content from org-mode files.
  • And html, and markdown.
  • It hot-builds.
  • It hot-reloads (no Javascript).
  • It does neither if you disdain creature comforts.
  • It does not demand any server process for local publishing.
  • It is quite small.
    # The complete "business logic" is 300-ish lines as of this comment,
    # counted as all lines except comments and blank lines.
    grep -E -v "\s?\#|^$" \
        ./bin/{events,metadata,templating,utils,hotreload}.sh |
       wc -l
  • It is Bash-ful.
  • I like it.

Before you get too exshited, may I warn you that the MIT license means I don't have to give a shite if this little shite maker fails to make your shite work. Contributing is replete with more warnings.

And last but not least, I hereby decree that all texsht herein be read in Sean Connery voish.

Dreams and desires

In my shite dreams, I desire...

  • Above all, to keep it (the "business logic") small. Small enough to cache, debug, and refactor in my head.

  • To install and use without superuser permission.

  • To extremely avoid toolchains and build dependencies. No gems / npms / venvs / what-have-yous. Thus, Bash is the language, because Bash is everywhere. And standard packages like pandoc or tidy, when one needs specific advanced functionality.

  • Dependency-free templating with plain-ol' HTML set in good ol' heredocs.

  • Simple metadata system, content namespacing, static asset organisation etc.

  • Web server optional (or any kind of server process for that matter). We aim for static sites, after all, which work just fine with file:// navigation.

  • To construct it from small, composable, purely functional, Unix-tool-like parts, because I like that sort of stuff a lot.

  • To give myself a seamless REPL-like edit-save-build-preview workflow.

    • Hot-build page processing (compile + build on save.)
    • Javascript-free browser hot-reloading. It works. It's terrible. It's awesome!
    • TODO: Potentially also extend the same mechanism to hot-deploy, on git push to a private repo on my own VPS somewhere. Maybe.


I accidentally restarted blogging after a long haitus. Before I could get words into the cloud, I muddled about with "modern" Static Site Generators. Because WordPress is so last century (or so I told myself). Then I got annoyed by the SSG Jamstack bespoke templating building etc. magic. Now I am on the dark path of making this. It is being blogged about at: shite: static sites from shell: part 1/2

Usage Demo

I use shite mainly in "hotreload" mode, mainly to write posts (in orgmode) and live preview them (in Firefox). Less mainly, to hot-preview modifications to styles and/or page templates. Least mainly, after labouring on a post interminably, I use it in "don't hotreload" mode to do a full site rebuild.

shite demo examples below.

Hot-reloaded shite editing

Basically this means that if I create, update, delete any file under sources, it must automatically translate to HTML, be published locally to public, and cause an appropriate page navigation or reload action in the web browser, where my site is open.

hotreload begins

invoke shite in hotreload mode

Call the "main" script in a clean new terminal session or tmux pane.


It helpfully opens the index file in Firefox, according to the defaults I've set in shite_global_data array in ./

hotreload content edits

hotreload content edits

In your Emacs or Vim, open some content file under sources. Edit, save, and watch the content appear in the browser. (Yes specifying Emacs/Vim is goofy, because I trigger hot actions based on inotify events. Apparently different editors do file updates differently. I use Emacs or Vim, so I watch for the events they cause, so it works on my machine. :)).

Frequently the browser remembers the scroll position, which is neat. Sometimes the hotreload is, well, shite. So I just hit space and save the content file to trigger hotreload again.

hotreload style edits

hotreload style edits

Go to some static asset, like a CSS stylesheet. Alter a thing, like background color value. Save and watch the color change in the browser.

hotreload template edits

hotreload template edits

Tweak some template fragment in, blog post template. Then switch to some blog post content file and modify it to trigger page build with the modified template (e.g. hit space and save).

hot rebuild indices and feeds

hot rebuild indices and feeds

This is a hack. The root page under sources is special. If I modify it, then it means I want to rebuild posts lists for the index page, for tags, and also rebuild related meta-files like the RSS feed, sitemap, robots.txt etc.

Full site builds

full site build

In a clean new terminal session call with "no", and optionally with the base_url of the deployment environment:

Rebuild full site for "local" file:/// navigation. Truly "serverless" :)

./ "no"

Rebuild full site for publication under my domain.

./ "no" ""

Environment Variables and Debug flags

These flags alter the behaviour of the system.

  • Setting SHITE_BUILD to "hot" will run the event system in "monitor" mode, which in turn drives hotreload behaviour. Setting it to "no" will suppress browser hotreload.
  • Setting SHITE_DEBUG_TEMPLATES to "debug" will cause templates to be sourced first, before publishing any templated source content.

Design and Internals

shite is quite Unixy inside. Or so I'd like to think.

Code is functional programming-style Bash. Everything is a function. Most functions are pure functions---little Unix tools in themselves. Most logic is pipeline-oriented. This works surprisingly well, because Shell ain't a bad place to FP.

I also wanted a live interactive REPL-like experience when writing with shite, because I like working in live/interactive runtimes like Clojure and Emacs.

So, shite has become this fully reactive event-driven system capable of hot build-and-reload-on-save.

File and URL naming scheme

There are three main directory namespaces:

  • sources housing the "source" content, such as blog posts written in orgmode, as well as CSS, Javascript, and other static assets.
  • public target for the compiled / built artefacts
  • bin for the shite-building code

The URL naming scheme follows sub-directory structure under sources, and is replicated as-is under the pubilic directory structure. Since this is a bog standard URL namespacing scheme, it also, applies directly to published content. Like so:



Code organisation

All "public" functions are namespaced as shite_the_func_name. All "private" functions are namespaced as __shite_the_func_name.

Functions exist to:

  • define common page fragments (meta, header, footer etc.)
  • compose full pages from components, metadata, and body content
  • assemble the site... build and publish sources into public targets
  • detect and process event streams to drive various site building features site builds, and browser hot reloading
  • react to processed events and drive hot compile of pages, hot build of site, and browser hot reload / navigation
  • provide convenience utilities for manual builds, local development

Calling the code

In a clean new terminal session:

  • CD to the root of this project
  • Source the dev utility code into the environment. This will bring in all the business logic, templates, as well as dev utility functions.
    source ./bin/
  • Hit shitTABTAB or __shiTABTAB at the command line for autocompletions.
  • Enter type -a func_name to print the function's definition and read its API.
  • Set shite_global_data and shite_page_data as needed.
  • Call functions at the command line. Call them individually, and/or composed with other functions to test / exercise parts of the system.

Templating system

Templates exist for page fragments (like header, footer, navigation), and for full page definitions (like the default page template). These are written as plain HTML wrapped in heredocs. ./bin/ provides these.

Templates are filled-in with variable data from different sources:

  • Bash associative arrays: shite_global_data contains site-wide metadata, and shite_page_data contains page-specific metadata. Some outside process must pre-set these arrays prior to processing any page.
  • stdin: to inject content into the templates that are wrappers for content.
  • function calls: to expand fragments like HTML metadata, links etc.

For example, a full page may be constructed as follows:

cat ./sample/ |
    pandoc -f markdown -t html |
    cat <<EOF
    <!DOCTYPE html>
        <body ${shite_page_data[page_id]}>
              $(cat -)

Metadata and front matter system

shite's metadata system is defined as key-value pairs. Keys name the metadata items, and would be associated with whatever value of that type. Examples below.

As noted earlier, run-time metadata is carried in the environment by the associative arrays shite_global_data and shite_page_data. These maybe be populated by direct construction, as well as updated from external sources.

Each page may specify its own metadata in "front matter" at the top of the page. This will be used in addition page metadata derived from other sources.

shite expects us to write front matter using syntax that is compatible with the given content type, as follows.

For orgmode content

Use comment lines # SHITE_META to demarcate the org-style metadata that shite should also parse as page-specific metadata.

#+title: This is a Title
#+slug: this/is/a/slug
#+date: Friday 26 August 2022 03:38:01 PM IST
#+tags: foo bar baz quxx
#+more_org_metadata: but not processed as shite metadata
#+still_more_org_metadata: and still not processed as shite metadata

* this is a top level heading

this is some orgmode content

#+TOC: headlines 1 local

** this is a sub heading
   - this is a point
   - this is another point
   - a third point

For markdown content

Write Jekyll-style YAML front matter, boxed between --- separators.

TITLE: This is a Title
slug: this/is/a/slug
DATE: Friday 26 August 2022 03:38:01 PM IST
TAGS: foo BAR baz QUXX

# this is a heading

this is some markdown content

## this is a subheading
  - this is a point
  - this is another point
  - a third point

For html content

We can simply use standard <meta> tags, that obey this convention: <meta name="KEY" content="value">.

<meta name="TITLE" content="This is a Title">
<meta name="slug" content="this/is/a/slug">
<meta name="DATE" content="Friday 26 August 2022 03:38:01 PM IST">
<meta name="TAGS" content="foo BAR baz QUXX">

<h1>This is a heading</h1>
<p>This is some text</p>
<h2>This is a subheading</h2>
    <li>This is a point</li>
    <li>This is another point.</li>
    <li>This is a third point.</li>

Bashful Hot Reloading Sans Javascript

Here be Yaks!

Being entirely spoiled by Clojure/Lisp/Spreadsheet style insta-gratifying live interactive workflows, I want hot reload and hot navigate in shite-making too.

But there does not seem to exist a standalone live web development server / tool that does not also want me to download half the known Internet as dependencies. As I said before, a thing I extremely do not want to do.

DuckSearch delivered Emacs impatient-mode, which is quite hot, but I don't want to hardwire this my Emacs. Luckily, it also delivered this exciting brainwave featuring 'inotify-tools' and 'xdotool':

Hot copy!

Because what could be hotter than my computer slammin' that F5 key for me? As if it knew what I really wanted deep down in my heart.

The event system

The event subsystem is orthogonal to everything else, and composes with the rest of the system.

The design is bog standard streaming architecture, viz. watch for file system events, then filter, deduplicate, analyse, and route them (tee) to different event processors. Currently there are just two such processors; one to compile and publish the page or asset associated with the event, another to hot reload the browser (or hot navigate) depending on the same event.

Baaasically this:

# detect file events
__shite_detect_changes ${watch_dir} 'create,modify,close_write,moved_to,delete' |
    __shite_events_gen_csv ${watch_dir} |
    # hot-compile-and-publish content, HTML, static, etc.
    tee >(shite_templating_publish_sources > /dev/null) |
    # browser hot-reload
    tee >(__shite_hot_cmd_public_events ${window_id} ${base_url} |

Events are simply a stream of CSV records structured like this:


We use different parts of the event record to cause different kinds of actions.

Liveness criterion

The afore-linked inotify-refresh script tries to periodically refresh a set of browser windows. We, however, want to be very eager. Any edit action on our content files and/or static assets must insta-trigger hot reload/navigate actions in the browser tab that's displaying our shite.

Hot reload scenarios

We want to define distinct reload scenarios: Mutually exclusive, collectively exhaustive buckets into which we can map file events we want to monitor.

If we do this, then we can model updates as a sort of write-ahead-log, punching events through an analysis pipeline, associate them with the exact-match scenario, and then finally cause the action. For example:

Refresh current tab when

  • static asset create, modify, move, delete

Go home when

  • current page deleted

Navigate to content when

  • current page content modified
  • any content page moved or created or modified

Hot reload behaviour

Since we are making the computer emulate our own keyboard actions, it can mess with our personly actions. If we stick to writing our shite in our text editor, and let the computer do the hotreloady thing, we should remain non-annoyed.

Unrealised Ambitions

There are many Yaks in the world.

From any source dir to any publish dir from anywhere on my box

For truly pervasive multi-site publishing mojo:

  • shite should be available on my PATH
  • I should be able to configure any source / public pair per site
  • Everything else should "just work" as it does

This is a small yak. I'll probably yakshave it soon.

Hot deployment

Obviously one can use the CI jobs of popular git hosts to trigger shite builds. But why use clunky current-century tech, when we have already advanced to the state of the art of the late 1900s... fully streaming and fully reactive?

Sarcasam aside, I don't see why the same event system cannot be used to add hot-deploy support, on a remote machine I run.

On the remote box:

  • a web server serves the public pages of the site
  • a clone of the site sources is enshrined
  • the selfsame hotreload process is live against sources (minus the browser-watching).
  • a git checkout auto-triggers on receiving a git push
  • which should cause hot-build against the modified sources (with some special case to trigger full build if a template changes)

On my local box:

  • edit, preview locally with local hotreloadin'
  • git commit, push sources to remote
  • hit F5 on the appropriate public URL
  • the hot-build should have completed in the time it takes to get to F5

Hot deployment with local hot reload

Do something over SSH to bring browser refresh back to local box, in case of hot deploys to remote server.

  • maybe Shell "Session Portability"? video, slides.
  • maybe tap the browser hotreload commands and stream only those back to my box, with a "server-mode" for hot-publish at the remote box and a "client-mode" for hot-reload on my local box?
  • maybe I'll find out it all "just work" with Emacs/TRAMP?

More creature comforts

Maybe some "Dev-ing/Drafting" time setup/Teardown scenario? Maybe a 'dev_server' function that we use to kick start a new shite writing session?

  • xdg-open a new tab in the default browser (say, firefox), and goto the home page of the shite based on config.
  • xdotool 'set_window --name' to a UUID for the life of the session.
  • Close the tab when we kill the dev session.


If you got all the way down here, and still want to contribute...


Why in the name of all that is holy and good, would you want to? Is it not blindingly obvious that this is the work of a goofball? Haven't you heard that Bash is Not Even A Real Programming Language? And isn't it face-slappingly obvious that your PRs will languish eternally, and your comments will fall into a nameless void?

Yes, sending patches is a terrible idea.

But please email me your hopes and dreams about your shite maker! I read email at my firstname dot lastname at gmail.

Together we can whistle silly tunes, and co-yak-shave our respective yaks, in our own special ways.

May The Source be with us.


This work is dual-licensed under the MIT license and the CC By-SA 4.0 license.

  • The Bash source code for making shite is licensed under the MIT license.
  • My website's content which I've included in this project, for demo purposes, commit is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License (CC By-SA 4.0).

SPDX-License-Identifier: mit OR cc-by-sa-4.0