soothfast-docs 0.4.4

//! Generated artifacts from markdown code blocks: one integration-test target
//! per package with a module per page (plain rust blocks), and one example
//! per package with a module per capture block. CI compares regenerated
//! content against disk, so stale artifacts fail.

use crate::markdown::{CodeBlock, Doc};

/// First line of every generated file; `docs check` uses it to tell
/// generated artifacts from hand-written ones.
pub const GENERATED_HEADER: &str = "// @generated by `cargo soothfast docs gen-tests`";

/// `docs/measuring.md` → `soothfast_doc_docs_measuring`.
pub fn sanitized_stem(rel_path: &str) -> String {
    let stem = rel_path.trim_end_matches(".md");
    let san: String = stem
        .chars()
        .map(|c| {
            if c.is_ascii_alphanumeric() {
                c.to_ascii_lowercase()
            } else {
                '_'
            }
        })
        .collect();
    san.trim_matches('_').to_string()
}

fn indent(code: &str, pad: &str) -> String {
    code.lines()
        .map(|l| {
            if l.is_empty() {
                String::new()
            } else {
                format!("{pad}{l}")
            }
        })
        .collect::<Vec<_>>()
        .join("\n")
}

/// One generated file, its path relative to the package directory.
#[derive(Debug, Clone, PartialEq)]
pub struct Generated {
    /// Where the file goes, relative to the package directory.
    pub path: String,
    /// The file's full content, header included.
    pub content: String,
}

/// Wraps the generated `mod` lines and dispatch of a target's `main.rs`.
/// rustfmt neither formats a brace-delimited macro body nor follows a `mod`
/// inside one, so `cargo fmt` cannot make a generated file differ from what
/// `docs check` regenerates, whatever its version, edition or config.
const VERBATIM: &str =
    "macro_rules! verbatim {\n    ($($tokens:tt)*) => {\n        $($tokens)*\n    };\n}\n";

/// The one integration-test target the doc tests of a package compile into.
pub const DOC_TESTS_DIR: &str = "tests/soothfast_docs";

/// Where generated files live, as (directory, file-name prefix, replaced by),
/// relative to the package directory. A file there that starts with
/// [`GENERATED_HEADER`] belongs to gen-tests: it is rewritten when expected
/// and removed when not. An entry with an empty prefix is a target directory
/// gen-tests owns outright; a prefixed entry is the per-page or per-block
/// layout of earlier releases and names the target directory that replaced it.
pub const OWNED: &[(&str, &str, Option<&str>)] = &[
    ("tests", "soothfast_doc_", Some(DOC_TESTS_DIR)),
    (DOC_TESTS_DIR, "", None),
    ("examples", "soothfast_capture_", Some(CAPTURE_DIR)),
    (CAPTURE_DIR, "", None),
];

/// The page a generated file was generated from, and the line of the block
/// when it is one block's file, read from its `// source: PAGE[ line N]` line.
pub fn recorded_source(content: &str) -> Option<(&str, Option<usize>)> {
    let source = content.lines().nth(1)?.strip_prefix("// source: ")?;
    match source.rsplit_once(" line ") {
        Some((page, line)) => match line.parse::<usize>() {
            Ok(line) => Some((page, Some(line))),
            Err(_) => Some((source, None)),
        },
        None => Some((source, None)),
    }
}

/// Every doc-test file a package needs: one module per page with testable
/// blocks under [`DOC_TESTS_DIR`], and the `main.rs` that declares them.
/// Empty when no page has a testable block. An error names two pages whose
/// paths sanitize to the same module.
pub fn doc_test_set<'a>(
    docs: impl IntoIterator<Item = (&'a str, &'a Doc)>,
) -> Result<Vec<Generated>, String> {
    let docs: Vec<(&str, &Doc)> = docs.into_iter().collect();
    one_global_allocator(&docs, CodeBlock::is_testable_rust)?;
    let mut pages: Vec<(String, &str, String)> = Vec::new();
    for (rel_path, doc) in docs {
        if let Some(content) = test_file(rel_path, doc) {
            pages.push((
                format!("doc_{}", sanitized_stem(rel_path)),
                rel_path,
                content,
            ));
        }
    }
    pages.sort_by(|a, b| a.1.cmp(b.1));
    for (i, (module, path, _)) in pages.iter().enumerate() {
        if let Some((_, other, _)) = pages[..i].iter().find(|(m, _, _)| m == module) {
            return Err(format!(
                "{other} and {path} both generate doc-test module `{module}`; rename one"
            ));
        }
    }
    if pages.is_empty() {
        return Ok(Vec::new());
    }
    let mods: String = pages
        .iter()
        .map(|(m, _, _)| format!("mod {m};\n"))
        .collect();
    let main =
        format!("{GENERATED_HEADER}\n#![allow(unused)]\n\n{VERBATIM}\nverbatim! {{\n{mods}}}\n");
    let mut set = vec![Generated {
        path: format!("{DOC_TESTS_DIR}/main.rs"),
        content: main,
    }];
    set.extend(pages.into_iter().map(|(module, _, content)| Generated {
        path: format!("{DOC_TESTS_DIR}/{module}.rs"),
        content,
    }));
    Ok(set)
}

/// The blocks of one target share a crate, so at most one of them may install
/// a global allocator, directly or through `soothfast::bench_main!()`.
fn one_global_allocator(
    docs: &[(&str, &Doc)],
    in_target: fn(&CodeBlock) -> bool,
) -> Result<(), String> {
    let mut found = docs.iter().flat_map(|(page, doc)| {
        doc.blocks
            .iter()
            .filter(move |b| in_target(b))
            .filter(|b| b.code.contains("#[global_allocator]") || b.code.contains("bench_main!"))
            .map(move |b| format!("{page} line {}", b.line))
    });
    match (found.next(), found.next()) {
        (Some(first), Some(second)) => Err(format!(
            "{first} and {second} both install a global allocator, and the blocks of a package compile into one crate; mark one `ignore` or `no_run` in another package"
        )),
        _ => Ok(()),
    }
}

/// For `rel_path` (relative to the package directory) in a place gen-tests
/// writes, per [`OWNED`]: the target directory it belongs to, and whether it is
/// a leftover of the layout that target replaced. The file's header decides
/// ownership.
pub fn owning_target(rel_path: &str) -> Option<(&'static str, bool)> {
    let (dir, name) = rel_path.rsplit_once('/').unwrap_or(("", rel_path));
    if !name.ends_with(".rs") {
        return None;
    }
    OWNED
        .iter()
        .find(|(d, prefix, _)| *d == dir && name.starts_with(prefix))
        .map(|(dir, _, replaced_by)| (replaced_by.unwrap_or(dir), replaced_by.is_some()))
}

/// Module content for one markdown page's doc tests, or None when it has no
/// testable blocks (all ignored / non-rust / capture).
pub fn test_file(rel_path: &str, doc: &Doc) -> Option<String> {
    let testable: Vec<&CodeBlock> = doc.blocks.iter().filter(|b| b.is_testable_rust()).collect();
    if testable.is_empty() {
        return None;
    }
    let mut out = format!("{GENERATED_HEADER}\n// source: {rel_path}\n#![allow(unused)]\n");
    for block in testable {
        let body = indent(&block.code, "    ");
        // Blocks that define `fn main` get it as a nested fn, then called.
        let call_main = if block.code.lines().any(|l| top_level_main(l).is_some()) {
            "\n    main();"
        } else {
            ""
        };
        // A `feature=NAME` tag gates the generated test behind that feature (`feature=A,B` behind all of them).
        let cfg = match block.feature_cfg() {
            Some(pred) => format!("#[cfg({pred})]\n"),
            None => String::new(),
        };
        // rustfmt::skip (per-item — the inner-attribute form is unstable):
        // blocks are verbatim from the markdown; formatting the generated fn
        // would make it diverge from its source and fail `docs check`.
        if block.has_tag("no_run") {
            out.push_str(&format!(
                "\n// line {}: compile-only (no_run)\n{cfg}#[rustfmt::skip]\n#[allow(dead_code)]\nfn doc_block_line_{}() {{\n{body}\n}}\n",
                block.line, block.line
            ));
        } else {
            out.push_str(&format!(
                "\n// line {}\n{cfg}#[rustfmt::skip]\n#[test]\nfn doc_block_line_{}() {{\n{body}{call_main}\n}}\n",
                block.line, block.line
            ));
        }
    }
    Some(out)
}

/// The one example target every capture block of a package compiles into.
pub const CAPTURE_DIR: &str = "examples/soothfast_capture";

/// Cargo's name for the [`CAPTURE_DIR`] example.
pub const CAPTURE_EXAMPLE: &str = "soothfast_capture";

/// Exit status of the capture example for a block id it does not contain.
pub const CAPTURE_UNKNOWN: (i32, &str) = (2, "soothfast_capture: unknown block");

/// Exit status of the capture example for a block whose `feature=` tag is
/// not enabled in this build. The stderr line starts with the sentinel.
pub const CAPTURE_DISABLED: (i32, &str) = (3, "soothfast_capture: disabled block");

/// Block ids for one page's capture blocks, in capture order: the argument
/// the capture example takes to run that block.
pub fn capture_ids(rel_path: &str, doc: &Doc) -> Vec<String> {
    let stem = sanitized_stem(rel_path);
    let count = doc.blocks.iter().filter(|b| b.is_capture()).count();
    (1..=count).map(|n| format!("{stem}_{n}")).collect()
}

/// Every capture file a package needs: one `block_<id>.rs` module per capture
/// block under [`CAPTURE_DIR`], and the `main.rs` that runs the block its first
/// argument names. Empty when no page has a capture block.
///
/// A block with a `feature=` tag (`feature=A,B` for all of them) compiles in
/// only under its features; without them the example exits with
/// [`CAPTURE_DISABLED`], so a capture run can keep the recorded output rather
/// than splice in nothing. A block keeps its own top-level `fn main`, whatever
/// it returns or however it is attributed (`#[tokio::main]`); a block without
/// one becomes the body of one. An error names a page whose block uses a
/// crate-level attribute, which a module cannot carry, or two pages whose
/// paths sanitize to the same block ids.
pub fn capture_set<'a>(
    docs: impl IntoIterator<Item = (&'a str, &'a Doc)>,
) -> Result<Vec<Generated>, String> {
    let docs: Vec<(&str, &Doc)> = docs.into_iter().collect();
    one_global_allocator(&docs, CodeBlock::is_capture)?;
    let mut blocks: Vec<(String, &str, Option<String>, String)> = Vec::new();
    let mut stems: Vec<(String, &str)> = Vec::new();
    for (rel_path, doc) in docs {
        let ids = capture_ids(rel_path, doc);
        if ids.is_empty() {
            continue;
        }
        let stem = sanitized_stem(rel_path);
        if let Some((_, other)) = stems.iter().find(|(s, _)| *s == stem) {
            return Err(format!(
                "{other} and {rel_path} both generate capture blocks `{stem}_N`; rename one"
            ));
        }
        stems.push((stem, rel_path));
        let captures = doc.blocks.iter().filter(|b| b.is_capture());
        for (id, block) in ids.into_iter().zip(captures) {
            let body = capture_module(block)
                .map_err(|e| format!("{rel_path} line {}: {e}", block.line))?;
            let content = format!(
                "{GENERATED_HEADER}\n// source: {rel_path} line {}\n{body}",
                block.line
            );
            blocks.push((id, rel_path, block.feature_cfg(), content));
        }
    }
    if blocks.is_empty() {
        return Ok(Vec::new());
    }
    blocks.sort_by(|a, b| a.1.cmp(b.1));
    let mut set = vec![Generated {
        path: format!("{CAPTURE_DIR}/main.rs"),
        content: capture_main(&blocks),
    }];
    set.extend(blocks.into_iter().map(|(id, _, _, content)| Generated {
        path: format!("{CAPTURE_DIR}/block_{id}.rs"),
        content,
    }));
    Ok(set)
}

fn capture_main(blocks: &[(String, &str, Option<String>, String)]) -> String {
    let mut mods = String::new();
    let mut arms = String::new();
    for (id, _, cfg, _) in blocks {
        let run = format!("\"{id}\" => run(block_{id}::main),\n");
        match cfg {
            Some(pred) => {
                mods.push_str(&format!("#[cfg({pred})]\nmod block_{id};\n"));
                arms.push_str(&format!("            #[cfg({pred})]\n            {run}"));
                arms.push_str(&format!(
                    "            #[cfg(not({pred}))]\n            \"{id}\" => exit(CAPTURE_DISABLED, &id),\n"
                ));
            }
            None => {
                mods.push_str(&format!("mod block_{id};\n"));
                arms.push_str(&format!("            {run}"));
            }
        }
    }
    let (unknown, unknown_msg) = CAPTURE_UNKNOWN;
    let (disabled, disabled_msg) = CAPTURE_DISABLED;
    format!(
        "{GENERATED_HEADER}\n#![allow(dead_code, unused_imports)]\n\n\
         use std::process::{{ExitCode, Termination}};\n\n\
         {VERBATIM}\n\
         verbatim! {{\n{mods}}}\n\n\
         const CAPTURE_UNKNOWN: (u8, &str) = ({unknown}, \"{unknown_msg}\");\n\
         const CAPTURE_DISABLED: (u8, &str) = ({disabled}, \"{disabled_msg}\");\n\n\
         fn main() -> ExitCode {{\n    \
             let id = std::env::args().nth(1).unwrap_or_default();\n    \
             verbatim! {{\n        \
                 match id.as_str() {{\n{arms}            \
                     _ => exit(CAPTURE_UNKNOWN, &id),\n        \
                 }}\n    \
             }}\n\
         }}\n\n\
         fn run<T: Termination>(main: fn() -> T) -> ExitCode {{\n    \
             main().report()\n\
         }}\n\n\
         fn exit((code, sentinel): (u8, &str), id: &str) -> ExitCode {{\n    \
             eprintln!(\"{{sentinel}} {{id}}\");\n    \
             ExitCode::from(code)\n\
         }}\n"
    )
}

/// One block as a module whose `pub fn main` the capture example calls.
fn capture_module(block: &CodeBlock) -> Result<String, String> {
    let lines: Vec<&str> = block.code.lines().collect();
    let Some(at) = lines.iter().position(|l| top_level_main(l).is_some()) else {
        return Ok(format!(
            "pub fn main() {{\n{}\n}}\n",
            indent(&block.code, "    ")
        ));
    };
    for (i, line) in lines.iter().enumerate() {
        let next = lines[i + 1..].iter().find(|l| !l.trim().is_empty());
        if line.starts_with("#![") && !is_lint_attribute(line) {
            return Err(
                "a capture block cannot use a crate-level attribute (`#![...]`) other than a lint level"
                    .into(),
            );
        }
        if line.starts_with("#[macro_use]") && next.is_some_and(|l| l.starts_with("extern crate")) {
            return Err("a capture block cannot use `#[macro_use] extern crate`".into());
        }
    }
    let mut out: Vec<String> = lines.iter().map(|l| l.to_string()).collect();
    if top_level_main(lines[at]) == Some(false) {
        out[at] = format!("pub {}", lines[at]);
    }
    Ok(format!("{}\n", out.join("\n")))
}

/// Whether an inner attribute only sets lint levels, which a module file may
/// carry as well as a crate root.
fn is_lint_attribute(line: &str) -> bool {
    ["allow", "warn", "deny", "forbid", "expect"]
        .iter()
        .any(|lint| {
            line.strip_prefix("#![")
                .is_some_and(|a| a.starts_with(&format!("{lint}(")))
        })
}

/// For a line declaring a top-level `fn main` (or `async fn main`), whether
/// it is already `pub`. Indented lines and other items are None.
fn top_level_main(line: &str) -> Option<bool> {
    let restricted = line
        .strip_prefix("pub(")
        .and_then(|rest| rest.split_once(") ").map(|(_, rest)| rest));
    let (public, rest) = match restricted.or_else(|| line.strip_prefix("pub ")) {
        Some(rest) => (true, rest),
        None => (false, line),
    };
    let rest = rest.strip_prefix("async ").unwrap_or(rest);
    let after = rest.strip_prefix("fn main")?;
    (!after.starts_with(|c: char| c.is_alphanumeric() || c == '_')).then_some(public)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::markdown;

    #[test]
    fn generates_tests_and_examples() {
        let md = "```rust\nlet x = 1;\n```\n```rust no_run\nloop {}\n```\n```rust capture-output\nprintln!(\"h\");\n```\n";
        let doc = markdown::scan(md).unwrap();
        let tests = test_file("docs/a.md", &doc).unwrap();
        assert!(tests.contains("#[test]"));
        assert!(tests.contains("compile-only"));
        assert!(!tests.contains("println!")); // capture block excluded
        assert_eq!(capture_ids("docs/a.md", &doc), ["docs_a_1"]);
    }

    #[test]
    fn doc_tests_compile_into_one_target_with_a_module_per_page() {
        let a = markdown::scan("```rust\nlet x = 1;\n```\n").unwrap();
        let b = markdown::scan("```rust\nlet y = 2;\n```\n").unwrap();
        let none = markdown::scan("```rust ignore\nnope();\n```\n").unwrap();
        let set = doc_test_set([
            ("docs/b.md", &b),
            ("README.md", &a),
            ("docs/skip.md", &none),
        ])
        .unwrap();
        let paths: Vec<&str> = set.iter().map(|g| g.path.as_str()).collect();
        assert_eq!(
            paths,
            [
                "tests/soothfast_docs/main.rs",
                "tests/soothfast_docs/doc_readme.rs",
                "tests/soothfast_docs/doc_docs_b.rs"
            ]
        );
        assert!(set[0].content.starts_with(GENERATED_HEADER));
        assert!(
            set[0]
                .content
                .ends_with("verbatim! {\nmod doc_readme;\nmod doc_docs_b;\n}\n")
        );
        assert_eq!(set[1].content, test_file("README.md", &a).unwrap());
    }

    #[test]
    fn no_testable_page_means_no_doc_test_target() {
        let none = markdown::scan("```rust ignore\nnope();\n```\n").unwrap();
        assert!(doc_test_set([("README.md", &none)]).unwrap().is_empty());
    }

    #[test]
    fn two_pages_on_one_module_name_are_an_error_naming_both() {
        let a = markdown::scan("```rust\nlet x = 1;\n```\n").unwrap();
        let e = doc_test_set([("docs/a-b.md", &a), ("docs/a_b.md", &a)]).unwrap_err();
        assert!(
            e.contains("docs/a-b.md") && e.contains("docs/a_b.md"),
            "{e}"
        );
    }

    #[test]
    fn owned_places_cover_both_layouts_and_nothing_else() {
        assert_eq!(
            owning_target("tests/soothfast_doc_readme.rs"),
            Some((DOC_TESTS_DIR, true))
        );
        assert_eq!(
            owning_target("tests/soothfast_docs/main.rs"),
            Some((DOC_TESTS_DIR, false))
        );
        assert_eq!(
            owning_target("examples/soothfast_capture_readme_1.rs"),
            Some((CAPTURE_DIR, true))
        );
        assert_eq!(
            owning_target("examples/soothfast_capture/block_readme_1.rs"),
            Some((CAPTURE_DIR, false))
        );
        assert_eq!(owning_target("tests/integration.rs"), None);
        assert_eq!(owning_target("tests/soothfast_docs/data.json"), None);
        assert_eq!(owning_target("src/soothfast_doc_x.rs"), None);
    }

    #[test]
    fn all_ignored_means_no_test_file() {
        let doc = markdown::scan("```rust ignore\nfictional();\n```\n").unwrap();
        assert!(test_file("README.md", &doc).is_none());
    }

    #[test]
    fn feature_tag_gates_generated_test() {
        let md =
            "```rust feature=risk\nlet x = 1;\n```\n```rust no_run feature=risk\nloop {}\n```\n";
        let doc = markdown::scan(md).unwrap();
        let tests = test_file("docs/r.md", &doc).unwrap();
        assert_eq!(tests.matches("#[cfg(feature = \"risk\")]").count(), 2);
    }

    #[test]
    fn comma_separated_feature_tag_gates_on_all_of_them() {
        // A block whose body needs two features to compile — gating on only
        // the first leaves it enabled, and failing to build, when the second
        // is off.
        let md = "```rust no_run feature=risk,polygon\nloop {}\n```\n";
        let doc = markdown::scan(md).unwrap();
        let tests = test_file("docs/r.md", &doc).unwrap();
        assert!(tests.contains("#[cfg(all(feature = \"risk\", feature = \"polygon\"))]"));
    }

    fn capture(md: &str) -> Vec<Generated> {
        capture_set([("docs/c.md", &markdown::scan(md).unwrap())]).unwrap()
    }

    #[test]
    fn capture_blocks_compile_into_one_example_with_a_module_per_block() {
        let set = capture(
            "```rust capture-output\nprintln!(\"a\");\n```\n```rust capture-output\nprintln!(\"b\");\n```\n",
        );
        let paths: Vec<&str> = set.iter().map(|g| g.path.as_str()).collect();
        assert_eq!(
            paths,
            [
                "examples/soothfast_capture/main.rs",
                "examples/soothfast_capture/block_docs_c_1.rs",
                "examples/soothfast_capture/block_docs_c_2.rs",
            ]
        );
        let main = &set[0].content;
        assert!(main.contains("mod block_docs_c_1;\nmod block_docs_c_2;\n"));
        assert!(main.contains("\"docs_c_2\" => run(block_docs_c_2::main),"));
        assert!(main.contains("_ => exit(CAPTURE_UNKNOWN, &id),"));
        assert!(
            set[1]
                .content
                .contains("pub fn main() {\n    println!(\"a\");\n}")
        );
        assert_eq!(
            recorded_source(&set[1].content),
            Some(("docs/c.md", Some(1)))
        );
    }

    #[test]
    fn no_capture_block_means_no_capture_example() {
        assert!(capture("```rust\nlet x = 1;\n```\n").is_empty());
    }

    #[test]
    fn feature_tag_gates_the_module_and_dispatches_disabled_without_it() {
        let set = capture("```rust capture-output feature=risk,polygon\nprintln!(\"hi\");\n```\n");
        let main = &set[0].content;
        let pred = "all(feature = \"risk\", feature = \"polygon\")";
        assert!(main.contains(&format!("#[cfg({pred})]\nmod block_docs_c_1;")));
        assert!(main.contains(&format!(
            "#[cfg({pred})]\n            \"docs_c_1\" => run(block_docs_c_1::main),"
        )));
        assert!(main.contains(&format!(
            "#[cfg(not({pred}))]\n            \"docs_c_1\" => exit(CAPTURE_DISABLED, &id),"
        )));
        assert!(main.contains(
            "const CAPTURE_DISABLED: (u8, &str) = (3, \"soothfast_capture: disabled block\");"
        ));
    }

    #[test]
    fn a_blocks_own_top_level_main_becomes_pub_and_keeps_its_attributes() {
        let md = "```rust capture-output\n// fn main is below\nuse std::fmt;\n\n#[tokio::main]\nasync fn main() -> Result<(), Box<dyn std::error::Error>> {\n    fn main_helper() {}\n    Ok(())\n}\n```\n";
        let block = &capture(md)[1].content;
        assert!(block.contains("// fn main is below\nuse std::fmt;"));
        assert!(block.contains("#[tokio::main]\npub async fn main() -> Result<"));
        assert!(block.contains("    fn main_helper() {}"));
        assert_eq!(block.matches("fn main(").count(), 1);
    }

    #[test]
    fn an_already_pub_main_is_left_alone() {
        let block = &capture("```rust capture-output\npub fn main() {}\n```\n")[1].content;
        assert!(block.ends_with("\npub fn main() {}\n"), "{block}");
    }

    #[test]
    fn crate_level_attributes_in_a_capture_block_are_an_error() {
        for code in [
            "#![feature(test)]\nfn main() {}",
            "#[macro_use]\nextern crate serde;\nfn main() {}",
        ] {
            let doc = markdown::scan(&format!("```rust capture-output\n{code}\n```\n")).unwrap();
            let e = capture_set([("docs/c.md", &doc)]).unwrap_err();
            assert!(
                e.starts_with("docs/c.md line 1: a capture block cannot"),
                "{e}"
            );
        }
    }

    #[test]
    fn two_pages_on_one_capture_stem_are_an_error_naming_both() {
        let a = markdown::scan("```rust capture-output\nlet x = 1;\n```\n").unwrap();
        let e = capture_set([("docs/a-b.md", &a), ("docs/a_b.md", &a)]).unwrap_err();
        assert!(
            e.contains("docs/a-b.md") && e.contains("docs/a_b.md"),
            "{e}"
        );
    }

    #[test]
    fn recorded_source_reads_page_with_or_without_a_line() {
        let h = GENERATED_HEADER;
        assert_eq!(
            recorded_source(&format!("{h}\n// source: README.md\n")),
            Some(("README.md", None))
        );
        assert_eq!(
            recorded_source(&format!("{h}\n// source: docs/a line b.md line 41\n")),
            Some(("docs/a line b.md", Some(41)))
        );
        assert_eq!(
            recorded_source(&format!("{h}\n// source: docs/a line b.md\n")),
            Some(("docs/a line b.md", None))
        );
        assert_eq!(recorded_source(&format!("{h}\nfn main() {{}}\n")), None);
    }

    #[test]
    fn a_lint_level_inner_attribute_is_allowed_in_a_capture_block() {
        for attr in ["allow(dead_code)", "warn(missing_docs)", "expect(unused)"] {
            let md = format!("```rust capture-output\n#![{attr}]\nfn main() {{}}\n```\n");
            let block = &capture(&md)[1].content;
            assert!(
                block.contains(&format!("#![{attr}]\npub fn main() {{}}")),
                "{block}"
            );
        }
    }

    #[test]
    fn a_restricted_visibility_main_is_the_blocks_own_main() {
        for vis in ["pub(crate)", "pub(super)"] {
            let md = format!("```rust capture-output\n{vis} fn main() {{}}\n```\n");
            let block = &capture(&md)[1].content;
            assert!(
                block.ends_with(&format!("\n{vis} fn main() {{}}\n")),
                "{block}"
            );
        }
    }

    #[test]
    fn only_a_top_level_main_is_called_by_a_doc_test() {
        let md = "```rust\n// fn main is not here\nfn main_loop() {}\nlet s = \"fn main\";\n```\n```rust\nfn main() {}\n```\n";
        let tests = test_file("docs/m.md", &markdown::scan(md).unwrap()).unwrap();
        assert_eq!(tests.matches("    main();").count(), 1, "{tests}");
    }

    #[test]
    fn two_blocks_installing_a_global_allocator_are_an_error_naming_both() {
        let a = markdown::scan("```rust no_run\nsoothfast::bench_main!();\n```\n").unwrap();
        let b =
            markdown::scan("```rust no_run\n#[global_allocator]\nstatic A: X = X;\n```\n").unwrap();
        let e = doc_test_set([("docs/a.md", &a), ("docs/b.md", &b)]).unwrap_err();
        assert!(
            e.starts_with("docs/a.md line 1 and docs/b.md line 1 both install"),
            "{e}"
        );
        assert!(doc_test_set([("docs/a.md", &a)]).is_ok());
    }

    /// Writes `set` under a temp dir and asks rustfmt, in every edition,
    /// whether it would change the target's `main.rs`.
    fn assert_rustfmt_leaves_alone(name: &str, set: &[Generated], root: &str) {
        let dir = std::env::temp_dir().join(format!("soothfast-fmt-{name}-{}", std::process::id()));
        let _ = std::fs::remove_dir_all(&dir);
        for g in set {
            let path = dir.join(&g.path);
            std::fs::create_dir_all(path.parent().unwrap()).unwrap();
            std::fs::write(path, &g.content).unwrap();
        }
        for edition in ["2015", "2021", "2024"] {
            let out = std::process::Command::new("rustfmt")
                .args(["--check", "--edition", edition])
                .arg(dir.join(root).join("main.rs"))
                .output()
                .unwrap();
            assert!(
                out.status.success(),
                "edition {edition}:\n{}",
                String::from_utf8_lossy(&out.stdout)
            );
        }
        let _ = std::fs::remove_dir_all(&dir);
    }

    fn rustfmt_present() -> bool {
        std::process::Command::new("rustfmt")
            .arg("--version")
            .output()
            .is_ok_and(|o| o.status.success())
    }

    #[test]
    fn generated_targets_are_already_rustfmt_clean() {
        if !rustfmt_present() {
            return;
        }
        let long =
            "docs/library/an_unusually_long_page_name_for_a_reference_section_of_the_library.md";
        let mut capture_md = String::new();
        for n in 0..12 {
            let tag = if n % 3 == 0 {
                " feature=risk,polygon,indicators,backtesting"
            } else {
                ""
            };
            capture_md.push_str(&format!(
                "```rust capture-output{tag}\nprintln!(\"{n}\");\n```\n"
            ));
        }
        let md = format!("```rust\nlet x = 1;\n```\n{capture_md}");
        let pages: Vec<(&str, Doc)> = [
            long,
            "docs/b.md",
            "docs/a_10.md",
            "docs/a_2.md",
            "README.md",
        ]
        .into_iter()
        .map(|p| (p, markdown::scan(&md).unwrap()))
        .collect();
        let docs = || pages.iter().map(|(p, d)| (*p, d));
        assert_rustfmt_leaves_alone("tests", &doc_test_set(docs()).unwrap(), DOC_TESTS_DIR);
        assert_rustfmt_leaves_alone("capture", &capture_set(docs()).unwrap(), CAPTURE_DIR);
    }
}