haematite-cli 0.21.3

//! An answer's names without bodies: rows that carry only their origin.
//!
//! When the text is `.` or `collection("…")` alone, the service pages the
//! origins itself from a source of revision entries, and no worker starts.
//! The page keeps the same row, byte and records-read bounds as a page of
//! values, stops at its limit as one does, and its `next` always starts just
//! after the last place it read, so no record is read twice.

use std::num::NonZeroU64;

use haematite_query::Refusal;
use haematite_query::handle::Error;
use haematite_query::page::{Allowance, Origin, RecordSource, Resume};
use serde::{Deserialize, Serialize};

/// What each row of an answer carries.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Deserialize, Serialize)]
#[serde(rename_all = "lowercase")]
pub enum Rows {
    /// Each row's value beside its origin.
    #[default]
    Values,
    /// Each row's origin alone.
    Origins,
}

/// A row with no value.
#[derive(Debug, Serialize)]
pub struct Named {
    pub origin: Origin,
}

/// One page of origins.
#[derive(Debug)]
pub struct Page {
    pub rows: Vec<Named>,
    pub next: Option<Resume>,
    /// The records given, on this page and every earlier one.
    pub records_read: u64,
    counts: Vec<u64>,
}

fn protocol(error: impl std::fmt::Display) -> Error {
    Error::new(Refusal::QueryProtocol, error.to_string())
}

/// A `next` that starts just after `origin`.
const fn after(origin: Origin, records_read: u64, passed_counts: Vec<u64>) -> Resume {
    Resume {
        records_read,
        pending: origin,
        discard_outputs: 0,
        passed: true,
        passed_counts,
    }
}

/// One page of origins from `source`, `read_before` records already given.
///
/// # Errors
///
/// `query_value_too_large` for a row that cannot fit an empty page; any
/// refusal of the source.
pub fn page(
    source: &mut impl RecordSource,
    allowance: &Allowance,
    resume: Option<&Resume>,
    limits: &[NonZeroU64],
) -> Result<Page, Error> {
    let read_before = resume.map_or(0, |resume| resume.records_read);
    let counts = resume.map_or_else(
        || vec![0; limits.len()],
        |resume| resume.passed_counts.clone(),
    );
    if counts.len() != limits.len()
        || counts
            .iter()
            .zip(limits)
            .any(|(count, limit)| *count > limit.get())
    {
        return Err(protocol(
            "origins continuation counters do not match the limit stages",
        ));
    }
    if resume.is_some_and(|resume| !resume.passed || resume.discard_outputs != 0) {
        return Err(protocol(
            "origins continuation must start after a passed place",
        ));
    }
    let mut page = Page {
        rows: Vec::new(),
        next: None,
        records_read: read_before,
        counts,
    };
    // The rows array's brackets.
    let mut bytes = 2_u64;
    loop {
        if page
            .counts
            .iter()
            .zip(limits)
            .any(|(count, limit)| *count == limit.get())
        {
            return Ok(page);
        }
        let given = u64::try_from(page.rows.len()).map_err(protocol)?;
        if given == allowance.rows.get() {
            return Ok(page.cut());
        }
        let Some(record) = source.next_record()? else {
            page.next = source
                .passed()
                .map(|origin| after(origin, page.records_read, page.counts.clone()));
            return Ok(page);
        };
        let row = Named {
            origin: record.origin,
        };
        let size =
            u64::try_from(serde_json::to_vec(&row).map_err(protocol)?.len()).map_err(protocol)?;
        if size.saturating_add(2) > allowance.bytes.get() {
            return Err(Error::new(
                Refusal::QueryValueTooLarge,
                format!(
                    "the origin of {}/{} needs more than {} bytes",
                    row.origin.collection, row.origin.id, allowance.bytes
                ),
            ));
        }
        let separator = u64::from(given > 0);
        let grown = bytes.saturating_add(separator).saturating_add(size);
        if grown > allowance.bytes.get() {
            // Not given: the next page reads it again.
            return Ok(page.cut());
        }
        bytes = grown;
        page.records_read = page
            .records_read
            .checked_add(1)
            .ok_or_else(|| protocol("records read overflow"))?;
        for count in &mut page.counts {
            *count = count
                .checked_add(1)
                .ok_or_else(|| protocol("limit count overflow"))?;
        }
        page.rows.push(row);
    }
}

impl Page {
    /// Cut after the last row given; a full page has at least one.
    fn cut(mut self) -> Self {
        self.next = self
            .rows
            .last()
            .map(|row| after(row.origin.clone(), self.records_read, self.counts.clone()));
        self
    }
}