steep/crates/steep-server/src/bot_actions/review/mod.rs

//! The pull request review action.
//!
//! [`exec_review`] drives one review end to end. The work is split by concern:
//! - [`diff`] parses the pull request diff into the lines it changes,
//! - [`prompt`] renders those changes for the model,
//! - [`anchor`] checks that the model's reviews land on changed lines,
//! - [`render`] builds the markdown summary posted on the pull request.

mod anchor;
mod diff;
mod prompt;
mod render;

#[cfg(test)]
mod fixtures;

use tracing::{info, instrument, warn};

use crate::{
    consts::{BOT_PROCESS_MSG, REVIEW_PROMPT, SANDBOX_SYSTEM_PROMPT},
    domain::review::ReviewResult,
    event_bus::Reporter,
    gitea::{GiteaAPI, ReviewPayload},
    metrics,
    open_router::{OpenRouterClient, Tool},
    sandbox::{Sandbox, SandboxConfig, agent},
};

use self::{
    anchor::resolve_review_sides,
    diff::{drop_generated_files, parse_changed_lines, resolve_filenames},
    prompt::format_changes,
    render::review_result_to_markdown,
};

/// The services a review needs, borrowed from the bot for the duration of one job.
pub struct ReviewContext<'a> {
    pub gitea: &'a GiteaAPI,
    pub open_router: &'a OpenRouterClient,
    pub sandbox: &'a SandboxConfig,
    pub reporter: &'a Reporter,
}

pub struct ReviewOutcome {
    pub markdown: String,
    pub cost: Option<f64>,
    pub comments: u64,
}

impl ReviewOutcome {
    pub fn empty() -> Self {
        Self {
            markdown: String::new(),
            cost: None,
            comments: 0,
        }
    }
}

#[instrument(
    skip_all,
    fields(repo = %payload.repository.full_name, pr = payload.pull_request.number)
)]
pub async fn exec_review(
    ctx: &ReviewContext<'_>,
    tools: Vec<Tool>,
    payload: &ReviewPayload,
) -> anyhow::Result<ReviewOutcome> {
    info!(action = %payload.action, "Starting review");

    let full_name = &payload.repository.full_name;
    let index = payload.pull_request.number;

    let progress_comment = ctx
        .gitea
        .comment(
            &BOT_PROCESS_MSG.replace("{model}", ctx.open_router.model()),
            full_name,
            index,
        )
        .await?;

    let review = review_pull_request(ctx, tools, payload).await;

    // The progress comment is housekeeping: failing to update it must neither turn a
    // posted review into a failure nor hide the error that made the review fail.
    match review {
        Ok((review_result, markdown)) => {
            if let Err(err) = ctx
                .gitea
                .delete_comment(full_name, progress_comment.id)
                .await
            {
                warn!(%err, "Failed to delete the progress comment");
            }

            Ok(ReviewOutcome {
                markdown,
                cost: review_result.cost,
                comments: review_result.reviews.len() as u64,
            })
        }
        Err(err) => {
            if let Err(edit_err) = ctx
                .gitea
                .edit_comment(
                    &format!("Error while reviewing: {err}"),
                    full_name,
                    progress_comment.id,
                )
                .await
            {
                warn!(err = %edit_err, "Failed to report the error on the progress comment");
            }

            Err(err)
        }
    }
}

/// Runs the review and posts it, returning the result and the markdown that was
/// posted.
async fn review_pull_request(
    ctx: &ReviewContext<'_>,
    tools: Vec<Tool>,
    payload: &ReviewPayload,
) -> anyhow::Result<(ReviewResult, String)> {
    let full_name = &payload.repository.full_name;
    let index = payload.pull_request.number;

    let git_diff = ctx.gitea.pull_request_diff(full_name, index).await?;

    // The file list only refines the paths and describes the changes: a failure is
    // not fatal, the diff is the source of truth for the lines.
    let files = match ctx.gitea.pull_request_files(full_name, index).await {
        Ok(files) => files,
        Err(err) => {
            warn!(%err, "Failed to list the pull request files");
            Vec::new()
        }
    };

    let mut changed_lines = parse_changed_lines(&git_diff);
    drop_generated_files(&mut changed_lines);
    resolve_filenames(&mut changed_lines, &files);

    let changes = format_changes(&files, &changed_lines);

    let bot_request = REVIEW_PROMPT
        .replace("{subject}", &payload.pull_request.title)
        .replace("{comment}", &payload.comment.body)
        .replace("{changes}", &changes);

    let (mut review_result, cost) = run_sandboxed_review(ctx, tools, payload, &bot_request).await?;

    resolve_review_sides(&mut review_result, &changed_lines);

    review_result.cost = cost;
    if let Some(cost) = review_result.cost {
        metrics::openrouter_cost_usd(cost);
    }

    let markdown = review_result_to_markdown(&review_result);

    ctx.gitea
        .post_pull_request_review(&review_result, &markdown, full_name, index)
        .await?;

    Ok((review_result, markdown))
}

async fn run_sandboxed_review(
    ctx: &ReviewContext<'_>,
    tools: Vec<Tool>,
    payload: &ReviewPayload,
    bot_request: &str,
) -> anyhow::Result<(ReviewResult, Option<f64>)> {
    let sandbox = Sandbox::create(
        &ctx.sandbox.runtime,
        &payload.repository.clone_url,
        ctx.gitea.token(),
        payload.pull_request.number,
        ctx.reporter,
    )
    .await?;

    let result = agent::run(
        ctx.open_router,
        &sandbox,
        tools,
        SANDBOX_SYSTEM_PROMPT,
        bot_request,
        ctx.sandbox.max_iterations,
        ctx.reporter,
    )
    .await;

    if let Err(err) = sandbox.cleanup().await {
        warn!(%err, "Failed to clean up sandbox container");
    }

    let result = result?;
    info!(
        iterations = result.iterations,
        cost = ?result.cost,
        "Sandboxed review finished"
    );

    Ok((result.answer, result.cost))
}