Templates (askama 0.16)

Templates are compiled into the binary at build time, so cargo build reports template errors as Rust type errors pointing at the #[derive(Template)] struct. Each page's struct sits next to its handler in src/web/*.rs.

Layouts

base.html                  blocks: title, page_class, subheader, content
├── layout_repo.html       repository header and tabs; blocks: page_title, content
└── layout_settings.html   account settings side nav; block: settings

• Every page struct needs ctx: Ctx, because base.html reads ctx.user, ctx.flash and ctx.path.
• Repo pages also need r: RepoCtx and tab: &'static str. tab is one of code, commits, branches, pulls or settings and selects the active tab.
• Settings pages need tab (profile or keys).
• In repo pages, page_title is a prefix ending in ·, e.g. {% block page_title %}Branches · {% endblock %}.
• Never create a file named base.html in a subdirectory. askama resolves {% extends %} relative to the current template's directory first, so repo/*.html extending "base.html" would pick up repo/base.html and fail with a cyclic dependency. That's why section layouts are named layout_*.html and kept at the root.

Partials ({% include %}, which share the including template's scope)

Bind them with {% let %} before including, e.g. {% let file_ref = commit.id.clone() %}, or {% let repos = recent %} to reuse the repo list.

Calling Rust from templates

Helpers live in src/web/fmt.rs and are called by full path:

{{ crate::web::fmt::icon("git-branch")|safe }}            {# names: match arms in src/icons.rs #}
{{ crate::web::fmt::avatar(user.username, 20)|safe }}
<time title="{{ crate::web::fmt::datetime_full(c.committer_time) }}">{{ crate::web::fmt::ago(c.committer_time) }}</time>
{{ crate::web::fmt::plural(view.commit_count, "commit") }} {# "1 commit", "3 commits", "2 branches" #}
{{ base }}/tree/{{ crate::web::fmt::urlpath(view.refname) }}/{{ crate::web::fmt::urlpath(entry.path) }}
/login?next={{ crate::web::fmt::urlencode(ctx.path) }}

• askama passes struct fields by reference (&c.committer_time) but method-call results and literals by value. Numeric helpers therefore take impl fmt::Num<i64> / impl fmt::Num<usize>, which accepts both. Follow that pattern for new numeric helpers. String helpers take &str.
• Methods on view structs are fine too (r.base(), r.pull_url(pr.number), f.display_path(), l.class()).
• Use urlpath for ref names and file paths in URL paths (keeps /, encodes spaces, # and so on). Use urlencode for query-string values.
• Only mark output |safe when it's HTML we generate: icon, avatar, highlighted lines, and markdown::render output. Never put |safe on user-provided strings.

Syntax that works here

• {% if let Some(x) = opt %}…{% endif %}, {% else if %}, {% for (a, b) in pairs %}, loop.index, loop.last.
• {% match value %}{% when crate::web::code::BlobContent::Code with { lines } %}…{% endmatch %}. Name enum variants by full path.
• Comparing an Option<enum>: data.mergeability == Some(crate::git::Mergeability::Clean).
• Comparing a loop item against a field: {% if *b == view.base %}.

Markup conventions

• There is no inline JavaScript (the CSP forbids it). Use the data-* hooks from static/app.js, which are listed in the root CLAUDE.md. Destructive forms get data-confirm="…".
• Dropdowns are <details class="dropdown"> with a <summary> and a .dropdown-menu, so they work without JS.
• Show empty states with <div class="blankslate"> (an icon, an <h3> and a <p>).
• Show errors on re-rendered forms with <div class="flash flash-error">. For messages after a redirect, use redirect_flash in the handler instead.
• Forms POST to same-origin paths, and every page must still work with JS disabled.