thor/AGENTS.md

Thor — Odin Static Site Generator

Thor is a static site generator written in Odin, replacing Hugo for the sbrow.github.io blog. It lives at ./thor/ as a git subtree with its own flake.nix.

Architecture

thor.json          ← site config (title, base_url, author, params, modules)
content/           ← markdown and HTML content files
layouts/           ← Mustache templates + partials (user overrides)
assets/            ← CSS (Tufte-based), JS, fonts, images
thor/defaults/     ← bundled default templates (embedded via #directory)
public/            ← build output (generated)

Package structure

thor/
├── treesitter/         # FFI types + grammar management (standalone package)
├── markdown/           # Content transformation pipeline (imports ../treesitter)
├── mustache/           # Template engine with lambdas + pipe filters
├── content.odin        # Page struct, scan_content, load_page
├── render.odin         # Template rendering, data structs, RSS, sitemap
├── site.odin           # Config (Flags, Config_File, Site), init_site
├── minify.odin         # HTML/CSS minification (imports treesitter)
├── feed.odin           # RSS + sitemap generation
├── vfs.odin            # Union file system (defaults → modules → site)
├── assets.odin         # VFS-based asset copying
├── opengraph.odin      # Open_Graph struct + og_init/og_for_page
├── frontmatter.odin    # JSON frontmatter parser
├── defaults.odin       # DEFAULTS_PATH constant (#directory)
├── main.odin           # Entry point
└── defaults/layouts/   # Bundled default templates

Source files (main package)

File	Responsibility
main.odin	Entry point. Sets context.logger, calls init_site, build_vfs, site_load_content, render_site. Optional Spall profiling via SPALL config flag.
site.odin	Flags (CLI), Config_File (thor.json), Site (runtime state + arena + VFS + pages + modules). Feature enum. 5-step init_site. Imports md "markdown" for Extension enum.
content.odin	Page struct, scan_content (section-aware walk that handles leaf bundles), load_page, infer_layout. Calls md.process() for the markdown pipeline.
render.odin	Template rendering: render_site, render_page_html, render_home_html, render_section. Data structs (Base_Data, Page_Data, Home_Data, Section_Data). VFS-based template loading with fallback chain (get_template).
minify.odin	HTML/CSS minification via tree-sitter. Imports ts "treesitter".
feed.odin	RSS feed + sitemap XML. Uses page.url for canonical URLs.
vfs.odin	Union file system: VFS, build_vfs, mount_dir, mount_subdir, mount_recursive, vfs_get. Layers defaults → modules → site.
assets.odin	copy_assets_dir — iterates VFS entries with assets/ prefix, minifies CSS, copies verbatim or via os.copy_file.
opengraph.odin	Open_Graph struct (fields ordered per OGP spec). og_init(site) for site defaults, og_for_page(site, page, base) for page-specific OG data.
frontmatter.odin	JSON frontmatter parser ({ } delimited). Supports layout field for template override.
defaults.odin	DEFAULTS_PATH constant, resolved at compile time via #directory so bundled templates ship in the binary.

Subpackages

Package	Files	Responsibility
treesitter/	treesitter.odin	FFI types (Parser, Node, Query, etc.), @(link_prefix="ts_") foreign bindings, grammar management (ensure_parser, load_grammar, grammar_cache), statically-linked HTML/CSS grammars
markdown/	markdown.odin	Extension enum, DEFAULT_EXTENSIONS, process(body, ext, file_path) — full pipeline, parse_extension_list, apply_extension_config
	footnotes.odin	strip_definitions (pre-cmark), inject_notes (post-cmark)
	alerts.odin	inject_alerts — GitHub alert blocks (> [!NOTE]) → styled blockquotes with semantic class names (alert-note etc.)
	emoji.odin	expand_emoji — :shortcode: → unicode emoji
	sectionate.odin	wrap_sections — splits HTML at <h2> into <section> wrappers
	highlight.odin	Syntax highlighting via tree-sitter. Imports ../treesitter.
mustache/	See Mustache engine below	Template engine

Icon SVGs live as HTML partials in layouts/partials/icons/ (home, github, rss, chevron_up, star).

Data flow

thor.json → find_config → init_site (5-step)
  → build_vfs (defaults/layouts → modules → site/layouts, site/assets)
  → site_load_content (scan_content + url computation)
  → render_site
    → load_partials + get_template (VFS + fallback chain)
    → render_page_html / render_home_html / render_section
    → optional minify_html
    → public/

Page struct

Page :: struct {
    section:     string,      // "" for root, "posts", etc.
    slug:        string,
    layout:      string,      // inferred or frontmatter override
    permalink:   string,      // relative URL path
    url:         string,      // full canonical URL (base_url + permalink)
    title:       string,
    description: string,
    date:        string,
    menu:        string,
    body_html:   string,
    draft:       bool,
    is_starred:  bool,
    _is_index:   bool `private`,
}

No Page_Type enum — page type is inferred from section + _is_index. Layout is inferred via infer_layout(section, is_index):

• Home (root index): "home"
• Section index: "<section>_index" (e.g. "posts_index")
• Section page: singularized section (e.g. "post")
• Root page: "page"

Template fallback chain (in get_template): for content pages, post → page → base; for section indexes, posts_index → section_index → page → base. Fallbacks logged at debug level. Frontmatter layout field overrides the inferred value.

Config system

Config is split into three structs with a clear 5-step initialization flow:

• Flags — CLI args only. Parsed by core:flags. Includes path overrides (--content, --assets, --output, --layouts), build-mode toggles (-drafts, -watch, -minify), and -ext/-no-ext for markdown extension overrides.
• Config_File — parsed from thor.json via json.unmarshal_string. Holds title, paths, markdown_extensions (JSON), params (JSON), modules (JSON array of relative paths).
• Site — runtime state: arena, pages, modules, VFS, features: bit_set[Feature], markdown_extensions: bit_set[md.Extension].

Feature enum — Drafts, Minify, Watch. Checked with .Minify in site.features.

markdown.Extension enum (in the markdown package, not main) — Emoji, Sidenotes, Alerts, Highlight, Sections. Default is md.DEFAULT_EXTENSIONS (currently .Emoji, .Sidenotes, .Alerts). Configurable via:

• thor.json: "markdown_extensions": { "emoji": true, "highlight": false, ... }
• CLI: -ext:highlight,sections (enable) / -no-ext:emoji (disable). Comma-separated, case-insensitive.

find_config — walks up from CWD looking for thor.json. Falls back to ./thor.json.

Config precedence: CLI flags > thor.json values > hardcoded defaults.

{
  "title": "...",
  "base_url": "...",
  "author": "...",
  "modules": ["../path/to/module"],
  "markdown_extensions": {
    "emoji": true,
    "sidenotes": true,
    "alerts": true,
    "highlight": true,
    "sections": true
  },
  "params": {
    "social": [
      { "name": "github", "url": "...", "icon": "icons/github" }
    ]
  }
}

VFS (Union File System)

Layered directory resolution for templates and assets: site layouts/ → module layouts/ → defaults/layouts/.

VFS :: struct { files: map[string]VFS_Entry }
VFS_Entry :: struct { fs_path: string, data: []byte }

build_vfs mounts in reverse precedence (defaults first, site last overwrites). DEFAULTS_PATH resolved at compile time via #directory, so bundled templates ship inside the binary. Modules configured via "modules": ["../path"] in thor.json — each module contributes layouts/ and assets/ subdirectories. vfs_get lazily reads file contents on first access.

Content is not yet in the VFS — scan_content still uses direct filesystem reads. (See TODOS.md.)

Open Graph

Open_Graph struct in opengraph.odin with fields ordered per ogp.me spec. Site defaults set via og_init(site) (site_name, description, default image, locale). Page-specific fields via og_for_page(site, page, base) (copies base, overrides url/title/type/is_article/section/published_time). Templates access via {{og.url}}, {{og.title}}, {{#og.is_article}}, etc.

Markdown pipeline

Lives in the markdown package. Entry point: md.process(body, ext, file_path). All .html content files skip the pipeline entirely — body is used as-is.

raw markdown
  → md.strip_definitions     (if .Sidenotes — pre-cmark)
  → cmark markdown_to_html   (Unsafe mode for HTML passthrough)
  → md.expand_emoji          (if .Emoji — post-cmark)
  → md.inject_notes          (if .Sidenotes — post-cmark)
  → md.inject_alerts         (if .Alerts — post-cmark)
  → md.highlight_code        (if .Highlight — post-cmark)
  → md.wrap_sections         (if .Sections — post-cmark)

Each step is gated by bit_set[md.Extension].

Template system

Templates use Mustache with template inheritance ({{<base}} / {{$block}}):

<!-- base.html -->
<body>{{> nav}}{{$content}}{{/content}}{{> footer}}</body>

<!-- page.html (content layout) -->
{{<base}}
{{$content}}
<main><article><h1>{{page_title}}</h1>{{&body}}</article></main>
{{/content}}
{{/base}}

Data is passed as typed structs (not map[string]any). Mustache resolves struct fields via Odin reflection, including using-embedded fields. Date presence is checked via string truthiness ({{#date_iso}}) — no separate has_date bool needed.

Base_Data :: struct {
    now:    datetime.DateTime,
    author: string,
    params: json.Value,
    body:   string,
    title:  string,
    og:     Open_Graph,
}
Page_Data :: struct {
    using base:   Base_Data,  // fields promoted via reflection fallback
    page_title:   string,
    date_iso:     string,
    date_display: string,
}
Home_Data :: struct {
    using base: Base_Data,
    pages:      [dynamic]Page_Context,
}
Section_Data :: struct {
    using base: Base_Data,
    page_title: string,
    posts:      [dynamic]Page_Context,  // flat list; year grouping done in template via pipe
}

render_site pre-parses all partials and the base layout once (via mustache.parse), then per-layout templates are cached in get_template. Year-based grouping on section index pages is done in the template via {{#posts | group_by year}} (see Pipes extension below) — there is no Year_Section Go-side struct.

Pipes extension

Section tags may transform the resolved value before iteration:

{{#posts | group_by year}}
  {{key}}: {{#items}}{{title}}, {{/items}}
{{/posts}}

Currently only group_by <field> is implemented. Filter results live in context.temp_allocator (render-scoped). See mustache/EXTENSIONS.md for syntax details, caps (MAX_PIPES, MAX_PIPE_ARGS), and the Group struct shape.

Comments

page.html includes {{> comments}}. The comments.html partial self-guards with {{#og.is_article}} so it only renders on article pages — no separate is_post flag.

Syntax highlighting

Build-time highlighting via Tree-sitter C FFI. No client-side JavaScript.

• HTML and CSS grammars statically linked via Nix (mkGrammarStaticLib in thor/flake.nix). Always available, no dlopen.
• Other grammars (bash, odin, nu, etc.) loaded via dlopen from Helix's compiled .so files.
• Highlight queries (.scm) loaded from Helix's runtime directory.
• Paths hardcoded in treesitter/treesitter.odin (GRAPHS_PATH, QUERIES_PATH) — Nix store paths, Helix-version-dependent. (See TODOS.md.)
• Grammar loading split: ensure_parser (parser only, used by minify) vs load_grammar (parser + query, used by highlight).
• Capture names mapped to CSS classes: keyword → .hl-keyword, etc.
• Atom-one-dark color theme in main.css.

Minification

Optional, enabled with -minify flag (.Minify in Feature bit_set).

• HTML — tree-sitter parses output, strips comments, removes inter-tag whitespace, preserves <pre>/<code>/<textarea>/<script>/<style> content. Applied after template rendering.
• CSS — tree-sitter parses .css files in assets/, strips comments, collapses whitespace, trims around {};:,. Applied during copy_assets_dir.
• Non-CSS files in assets/ copied verbatim.

Memory management

• Site owns a mem.Dynamic_Arena
• init_site calls mem.dynamic_arena_init(&site.arena) (Odin's default alignment suffices)
• Config loading (flags + JSON) uses the arena allocator explicitly
• site_allocator(site) returns the arena allocator for callers
• destroy_site frees the arena
• main.odin sets context.logger = log.create_console_logger() — without this, all log.* calls are silently dropped
• context.allocator is set to site_allocator(&site) in the main loop
• context.temp_allocator freed per watch-loop iteration via defer free_all

Spall profiling

Optional, compiled out by default. Enabled with -define:SPALL=true:

odin build . -define:SPALL=true -o:speed -out:thor-prof
./thor-prof -drafts  # generates thor.spall

Uses core:prof/spall with @(instrumentation_enter)/@(instrumentation_exit) hooks — every function auto-instrumented, no manual annotation needed.

Building

Local development

nix develop
# From blog root:
odin run ./thor -- -drafts
# Assets (CSS/JS/fonts) are copied/minified automatically by thor
caddy run  # serves public/ on blog.localhost

No CSS build step — main.css is static Tufte-based CSS, no preprocessor or compiler needed.

Production build

nix build  # runs thor, outputs to ./result/

Tests

cd thor
odin test .                  # main package tests (site, frontmatter)
odin test . -all-packages    # includes mustache specs, lambdas, pipes, markdown tests

Mustache engine

Spec-compliant implementation at mustache/. See mustache/SPEC.md for the implementation specification and mustache/EXTENSIONS.md for non-standard extensions (pipes).

Files

File	Responsibility
mustache.odin	Public API (parse, render, Template), parser (parse_section with allocator threading), renderer (render_nodes), template inheritance (merge_block_overrides), delete_template/delete_partials
tokenizer.odin	Tokenizer (template string → []Token), standalone whitespace detection
data.odin	Reflection-based data model: base_value (peels union/any/nested-any layers), lookup_in (structs + maps, handles Type_Info_Any value kind in maps), resolve_name, is_truthy, any_to_string, list_info, extract_list_element (unwraps [dynamic]any element types so downstream lookups see the real value), call_interp_lambda/call_section_lambda
pipes.odin	Pipes extension: Pipe_Filter AST, parse_pipeline, apply_pipeline, apply_filter (switch dispatch), apply_group_by. Stored on Node.filters; render-scoped results in temp allocator.
spec_test.odin	JSON spec test runner — loads spec/specs/*.json, runs each test case
lambda_test.odin	Spec lambda tests
pipes_test.odin	Pipe filter tests

Architecture

parse(source) → tokenize → trim_standalone_whitespace → parse_section → Template
render(tmpl, data, partials) → render_nodes (walks flat node array against context stack) → string

• Two-phase API: parse() produces a reusable Template, render() walks it against data. Templates parsed once, rendered many times.
• Flat [dynamic]Node array with first_child/child_count indices — pre-order layout.
• Context stack: ^[dynamic]any with append/pop for section push/pop.
• base_value peels Named/Distinct/Union layers (including json.Value). Also unwraps nested any-of-any (which occurs when map[string]any values are read via runtime map internals).
• lookup_in resolves keys on structs (via reflect.struct_field_value_by_name with allow_using = true) and maps. Detects Type_Info_Any value kind in maps and reads the inner any directly to avoid double-wrap.
• Template inheritance: {{<parent}} loads parent from partials, {{$block}} defines overridable sections. merge_block_overrides propagates overrides through multi-level chains.
• Dynamic partial names: {{>*key}} resolves partial name from data context at render time.

Lambdas

Spec-compliant. Stored as any values in the data context.

• Interpolation lambdas: proc() -> string, proc() -> int, proc() -> bool — called via call_interp_lambda, result stringified and escaped.
• Section lambdas: proc(string) -> string, proc(string) -> int, proc(string) -> bool — called via call_section_lambda with the raw section text (node.content). String result is re-parsed as mustache and rendered against the current context stack.

Pipes

{{#key | op args…}}…{{/key}}. Stored as [dynamic; MAX_PIPES]Pipe_Filter on each Node (fixed-cap inline storage, no per-tag heap allocation at parse time). Applied in the renderer via apply_pipeline before truthiness check. Currently only group_by <field> is implemented (returns [dynamic]Group where Group{key, items}). See mustache/EXTENSIONS.md.

Not implemented

• Set delimiters ({{= =}}, delimiters.json)

Known limitations

• cmark allocates via C malloc, not the arena. HTML output leaks until process exit (problematic in watch mode — see TODOS.md).
• CSS/JS cache busting uses manual ?v=N query params instead of content hashing.
• Tree-sitter grammar/query paths for dynamic grammars hardcoded in treesitter/treesitter.odin (Nix store hashes, Helix-version-dependent). HTML/CSS are statically linked.
• map[string]any only works through lookup_in's special-case handling; thor otherwise uses structs.
• format_f64 in mustache brute-forces shortest float representation.
• Content directory not mounted in VFS (modules can ship templates/assets but not content packs yet).

Design decisions

You may never, ever remove TODO: or FIXME: comments. Those are for humans, not machines. See HUGO.md for analysis of why thor doesn't need Hugo's shortcode context isolation. See mustache/PARTIAL_INDENT.md for whitespace handling analysis. See mustache/SPEC.md for the original implementation specification. See mustache/EXTENSIONS.md for non-standard extensions (pipes).

TODO

See TODOS.md for the full list.