automapper_validation/

syntax.rs

//! The MIG syntax check a CONTRL reports: one message against its MIG alone.
//!
//! No Prüfidentifikator, AHB or condition takes part. A CONTRL reports syntax
//! errors only (CONTRL AHB 1.0 §2.1); a code wrong only in its context is an
//! AHB matter and goes to the APERAK. What is checked:
//!
//! - **UNH and UNT** (UCM): their elements, the segment count (UNT 0074, code
//!   29) and the message reference (UNT 0062 against UNH 0062, code 28). An
//!   error here stops the check of that message: nothing else is reported.
//! - **Segments** (UCS): a segment the MIG marks `M`/`R` missing (13), one the
//!   MIG has no place for (15), more data elements than the MIG defines (16),
//!   a segment or group repeating beyond its `MaxRep` (35, 36).
//! - **Data elements** (UCD): an `M`/`R` element or component empty (13), a
//!   value outside the MIG's code list (12), too many components (16), a bad
//!   decimal notation (19), characters outside the syntax level (21), a
//!   letter in a numeric element (37), no digit before the decimal mark (38),
//!   too long (39), too short (40).
//!
//! The codes are those the CONTRL MIG 2.0b allows in UCM/UCS/UCD 0085. Every
//! finding carries its segment counter from UNH = 1 (UCS 0096); a missing
//! segment or group is reported at its predecessor. Element positions follow
//! S011: the segment tag is 0098 = 1, the first component 0104 = 1.
//!
//! The message is placed with the assembler on the MIG with every segment
//! number kept, i.e. as if a Prüfidentifikator allowed all the MIG allows.

use std::collections::{HashMap, HashSet};

use mig_assembly::assembler::{AssembledGroup, AssembledSegment, Assembler, AssemblerConfig};
use mig_assembly::{pid_filter, structure_check, StructureDiagnosticKind};
use mig_types::schema::common::CodeDefinition;
use mig_types::schema::mig::{MigDataElement, MigSchema, MigSegment, MigSegmentGroup};
use mig_types::segment::OwnedSegment;
use serde::{Deserialize, Serialize};

/// Which CONTRL segment reports a finding.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub enum SyntaxScope {
    /// UCM: an error in UNH or UNT.
    Message,
    /// UCS: an error of a whole segment (0085 in UCS).
    Segment,
    /// UCD: an error of a data element in a segment.
    Element,
}

/// One syntax error of a message.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct SyntaxFinding {
    /// The CONTRL segment that reports it.
    pub scope: SyntaxScope,
    /// The syntax error code, DE0085 (`"13"`, `"35"`, …).
    pub code: String,
    /// The segment the error is in; for a missing one, the missing segment
    /// (a missing group: its first segment).
    pub segment_tag: String,
    /// The segment counter within the message, UNH = 1 (UCS 0096). A missing
    /// segment or group is reported at its predecessor.
    pub message_segment_number: u32,
    /// The data element's position in the segment, the tag being 1 (0098).
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub element_position: Option<u32>,
    /// The component's position in its composite, from 1 (0104).
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub component_position: Option<u32>,
    /// What is wrong, for people.
    pub detail: String,
}

/// Values accepted beyond the MIG's own code lists, by MIG segment `Number`,
/// element index and component index (0-based). A MIG's code list can lag
/// its AHBs — a Prüfidentifikator an AHB correction adds, an answer code an
/// AHB takes from a Codeliste the MIG predates — and what an AHB allows is
/// never a syntax error.
pub type AcceptedCodes = HashMap<(String, usize, usize), Accepted>;

/// What an element accepts beyond its MIG code list.
#[derive(Debug, Clone, Default, PartialEq, Eq)]
pub struct Accepted {
    /// Any value: an AHB leaves the element open.
    pub any: bool,
    /// These codes.
    pub codes: HashSet<String>,
}

impl Accepted {
    fn accepts(&self, value: &str) -> bool {
        self.any || self.codes.contains(value)
    }
}

/// The character repertoire of a syntax level (UNB S001 0001).
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Charset {
    /// UNOA: upper-case letters, digits, space and a few signs.
    Unoa,
    /// UNOB: printable ASCII.
    Unob,
    /// UNOC: ISO 8859-1, the energy market's level.
    Unoc,
    /// Any other level: every character but control characters.
    Other,
}

impl Charset {
    /// The repertoire of syntax identifier `id` (`"UNOC"`).
    pub fn from_syntax_identifier(id: &str) -> Self {
        match id {
            "UNOA" => Self::Unoa,
            "UNOB" => Self::Unob,
            "UNOC" => Self::Unoc,
            _ => Self::Other,
        }
    }

    fn allows(self, c: char) -> bool {
        match self {
            Self::Unoa => {
                c.is_ascii_uppercase() || c.is_ascii_digit() || " .,-()/='+:?!\"%&*;<>".contains(c)
            }
            Self::Unob => (' '..='~').contains(&c),
            Self::Unoc => (' '..='~').contains(&c) || ('\u{A0}'..='\u{FF}').contains(&c),
            Self::Other => !c.is_control(),
        }
    }
}

/// The interchange's syntax settings a value is judged by.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct SyntaxSettings {
    pub charset: Charset,
    /// The decimal mark (UNA, default `.`).
    pub decimal_mark: char,
}

impl Default for SyntaxSettings {
    fn default() -> Self {
        Self {
            charset: Charset::Unoc,
            decimal_mark: '.',
        }
    }
}

/// The placements of a MIG the syntax check assembles a message on: the
/// whole MIG kept apart, with its group variants merged, and with them
/// apart. They depend on the MIG alone, and building them is most of the
/// check's cost (about 10 ms for UTILMD), so a caller checking many messages
/// builds them once ([`check_message_syntax_prepared`]).
#[derive(Debug, Clone)]
pub struct MigPlacements {
    unmerged: MigSchema,
    merged: MigSchema,
    apart: MigSchema,
}

impl MigPlacements {
    pub fn new(mig: &MigSchema) -> Self {
        let numbers = all_numbers(mig);
        Self {
            unmerged: pid_filter::filter_mig_for_pid_unmerged(mig, &numbers),
            merged: pid_filter::filter_mig_for_pid(mig, &numbers),
            apart: pid_filter::filter_mig_variant_aware(mig, &numbers),
        }
    }
}

/// Check one message against `mig`.
///
/// `segments` are the message's segments as the assembler takes them for
/// `mig` (`MessageChunk::segments_for_mig`): UNH…UNT, with the envelope
/// before them where the MIG covers it. Only UNH…UNT is judged.
pub fn check_message_syntax(
    segments: &[OwnedSegment],
    mig: &MigSchema,
    settings: SyntaxSettings,
    accepted: &AcceptedCodes,
) -> Vec<SyntaxFinding> {
    check_message_syntax_prepared(segments, mig, &MigPlacements::new(mig), settings, accepted)
}

/// [`check_message_syntax`] with the MIG's placements built beforehand.
/// `placements` must be [`MigPlacements::new`] of `mig`.
pub fn check_message_syntax_prepared(
    segments: &[OwnedSegment],
    mig: &MigSchema,
    placements: &MigPlacements,
    settings: SyntaxSettings,
    accepted: &AcceptedCodes,
) -> Vec<SyntaxFinding> {
    let Some(unh_index) = segments.iter().position(|s| s.id == "UNH") else {
        return Vec::new();
    };
    let unh = &segments[unh_index];
    let unh_number = unh.segment_number;
    let counter = |segment_number: u32| segment_number.saturating_sub(unh_number) + 1;
    let unt_index = segments.iter().rposition(|s| s.id == "UNT");

    // UCM: UNH and UNT. An error here stops the message.
    let mut out = Vec::new();
    for seg in [Some(unh_index), unt_index]
        .into_iter()
        .flatten()
        .map(|i| &segments[i])
    {
        if let Some(slot) = mig.segments.iter().find(|s| s.id == seg.id) {
            check_elements(
                seg,
                slot,
                counter(seg.segment_number),
                settings,
                accepted,
                &mut out,
            );
        }
    }
    match unt_index {
        Some(i) => {
            let unt = &segments[i];
            let count = (i - unh_index + 1).to_string();
            if !unt.get_element(0).is_empty() && unt.get_element(0) != count {
                out.push(finding(
                    SyntaxScope::Message,
                    "29",
                    unt,
                    counter(unt.segment_number),
                    Some((2, None)),
                    format!(
                        "UNT counts {} segments, the message has {count}",
                        unt.get_element(0)
                    ),
                ));
            }
            if !unt.get_element(1).is_empty() && unt.get_element(1) != unh.get_element(0) {
                out.push(finding(
                    SyntaxScope::Message,
                    "28",
                    unt,
                    counter(unt.segment_number),
                    Some((3, None)),
                    format!(
                        "UNT names message '{}', UNH '{}'",
                        unt.get_element(1),
                        unh.get_element(0)
                    ),
                ));
            }
        }
        None => out.push(SyntaxFinding {
            scope: SyntaxScope::Message,
            code: "13".into(),
            segment_tag: "UNT".into(),
            message_segment_number: counter(
                segments.last().map_or(unh_number, |s| s.segment_number),
            ),
            element_position: None,
            component_position: None,
            detail: "UNT is missing".into(),
        }),
    }
    for f in &mut out {
        f.scope = SyntaxScope::Message;
    }
    if !out.is_empty() {
        return out;
    }

    // Place the message on the MIG with every segment number kept. The
    // assembler can merge a group's variants or keep them apart; each loses
    // some messages (merged: the order within a variant, UTILTS SEQ+Z43 DTM
    // RFF; apart: ORDERS' UNS after its LIN groups). A message fits the MIG
    // when some placement fits it, so the one with fewer findings is taken.
    let unmerged = &placements.unmerged;
    let place = |assembly_mig: &MigSchema| {
        let assembler = Assembler::with_config(
            assembly_mig,
            AssemblerConfig {
                strict_code_matching: true,
                skip_unknown_segments: true,
                ..Default::default()
            },
        );
        let (tree, mut diagnostics) = assembler.assemble_with_diagnostics(segments);
        diagnostics.extend(structure_check::check_structure(&tree, unmerged));
        (tree, diagnostics)
    };
    let merged = place(&placements.merged);
    let apart = place(&placements.apart);
    let (tree, diagnostics) = if apart.1.len() < merged.1.len() {
        apart
    } else {
        merged
    };

    let first_segments = group_first_segments(mig);
    let within =
        |n: u32| n >= unh_number && unt_index.map_or(true, |i| n <= segments[i].segment_number);
    let mut reported = HashSet::new();
    for diag in diagnostics {
        let n = u32::try_from(diag.position).unwrap_or(0);
        if !within(n) {
            continue;
        }
        let (code, tag) = match diag.kind {
            StructureDiagnosticKind::MissingRequiredSegment => ("13", diag.segment_id.clone()),
            StructureDiagnosticKind::MissingRequiredGroup => (
                "13",
                first_segments
                    .get(diag.segment_id.as_str())
                    .cloned()
                    .unwrap_or(diag.segment_id.clone()),
            ),
            StructureDiagnosticKind::MaxRepetitionsExceeded => ("35", diag.segment_id.clone()),
            StructureDiagnosticKind::GroupMaxRepetitionsExceeded => (
                "36",
                first_segments
                    .get(diag.segment_id.as_str())
                    .cloned()
                    .unwrap_or(diag.segment_id.clone()),
            ),
            _ => ("15", diag.segment_id.clone()),
        };
        // The assembler and the structure check can both see one defect.
        if !reported.insert((code, tag.clone(), n)) {
            continue;
        }
        out.push(SyntaxFinding {
            scope: SyntaxScope::Segment,
            code: code.into(),
            segment_tag: tag,
            message_segment_number: counter(n),
            element_position: None,
            component_position: None,
            detail: diag.message,
        });
    }

    // UCD: every segment between UNH and UNT against its slot, each group
    // repetition judged by its own MIG variant.
    let mut root: Vec<&AssembledSegment> = tree.segments.iter().collect();
    root.extend(tree.inter_group_segments.values().flatten());
    let mut walk = ElementWalk {
        settings,
        accepted,
        within: &within,
        counter: &counter,
        out: &mut out,
    };
    walk.segments(&root, &unmerged.segments);
    walk.groups(&tree.groups, &unmerged.segment_groups);

    // A segment whose only slot it fills with a wrong qualifier is reported
    // once, as that value (UCD 12), not also as out of place (UCS 15).
    let wrong_codes: HashSet<u32> = out
        .iter()
        .filter(|f| f.scope == SyntaxScope::Element && f.code == "12")
        .map(|f| f.message_segment_number)
        .collect();
    out.retain(|f| {
        !(f.scope == SyntaxScope::Segment
            && f.code == "15"
            && wrong_codes.contains(&f.message_segment_number))
    });

    out.sort_by_key(|f| {
        (
            f.message_segment_number,
            f.element_position,
            f.component_position,
        )
    });
    out
}

fn finding(
    scope: SyntaxScope,
    code: &str,
    seg: &OwnedSegment,
    message_segment_number: u32,
    at: Option<(u32, Option<u32>)>,
    detail: String,
) -> SyntaxFinding {
    SyntaxFinding {
        scope,
        code: code.into(),
        segment_tag: seg.id.clone(),
        message_segment_number,
        element_position: at.map(|(e, _)| e),
        component_position: at.and_then(|(_, c)| c),
        detail,
    }
}

/// Every segment `Number` of `mig`: the filter then keeps all of it.
fn all_numbers(mig: &MigSchema) -> HashSet<String> {
    fn groups(list: &[MigSegmentGroup], out: &mut HashSet<String>) {
        for g in list {
            out.extend(g.segments.iter().filter_map(|s| s.number.clone()));
            groups(&g.nested_groups, out);
        }
    }
    let mut out: HashSet<String> = mig
        .segments
        .iter()
        .filter_map(|s| s.number.clone())
        .collect();
    groups(&mig.segment_groups, &mut out);
    out
}

/// Group id → the tag of its first segment (a missing group is reported as
/// that segment).
fn group_first_segments(mig: &MigSchema) -> HashMap<&str, String> {
    fn walk<'m>(list: &'m [MigSegmentGroup], out: &mut HashMap<&'m str, String>) {
        for g in list {
            if let Some(first) = g.segments.first() {
                out.entry(g.id.as_str()).or_insert_with(|| first.id.clone());
            }
            walk(&g.nested_groups, out);
        }
    }
    let mut out = HashMap::new();
    walk(&mig.segment_groups, &mut out);
    out
}

/// The element check over the assembled tree, by MIG variant.
struct ElementWalk<'a> {
    settings: SyntaxSettings,
    accepted: &'a AcceptedCodes,
    within: &'a dyn Fn(u32) -> bool,
    counter: &'a dyn Fn(u32) -> u32,
    out: &'a mut Vec<SyntaxFinding>,
}

impl ElementWalk<'_> {
    fn groups(&mut self, present: &[AssembledGroup], variants: &[MigSegmentGroup]) {
        for group in present {
            let same_id: Vec<&MigSegmentGroup> =
                variants.iter().filter(|v| v.id == group.group_id).collect();
            for rep in &group.repetitions {
                // A repetition no variant claims is the structure check's
                // finding (UCS 15); its elements have no slots to be judged by.
                // Of several variants with its entry qualifier, the one it
                // fits best.
                let candidates: Vec<&MigSegmentGroup> = match same_id.as_slice() {
                    [only] => vec![*only],
                    _ => rep.segments.first().map_or(Vec::new(), |entry| {
                        same_id
                            .iter()
                            .copied()
                            .filter(|v| {
                                v.entry_qualifier().is_some_and(|q| {
                                    q.tag == entry.tag && q.matches(&entry.elements)
                                })
                            })
                            .collect()
                    }),
                };
                let mut best: Option<Vec<SyntaxFinding>> = None;
                for variant in candidates {
                    let mut found = Vec::new();
                    let mut walk = ElementWalk {
                        settings: self.settings,
                        accepted: self.accepted,
                        within: self.within,
                        counter: self.counter,
                        out: &mut found,
                    };
                    let own: Vec<&AssembledSegment> = rep.segments.iter().collect();
                    walk.segments(&own, &variant.segments);
                    walk.groups(&rep.child_groups, &variant.nested_groups);
                    if best.as_ref().map_or(true, |b| found.len() < b.len()) {
                        best = Some(found);
                    }
                }
                self.out.extend(best.unwrap_or_default());
            }
        }
    }

    fn segments(&mut self, present: &[&AssembledSegment], slots: &[MigSegment]) {
        for seg in present {
            let Some(n) = seg.segment_number else {
                continue;
            };
            if seg.tag == "UNH" || seg.tag == "UNT" || !(self.within)(n) {
                continue;
            }
            let owned = OwnedSegment {
                id: seg.tag.clone(),
                elements: seg.elements.clone(),
                segment_number: n,
            };
            // Of the slots `seg` may fill, the one it fits best: a group can
            // hold several slots of one tag and qualifier (DTM+Z21 by day and
            // by minute), and only one of them need take its values.
            let mut best: Option<Vec<SyntaxFinding>> = None;
            for slot in slots_for(seg, slots) {
                let mut found = Vec::new();
                check_elements(
                    &owned,
                    slot,
                    (self.counter)(n),
                    self.settings,
                    self.accepted,
                    &mut found,
                );
                if best.as_ref().map_or(true, |b| found.len() < b.len()) {
                    best = Some(found);
                }
            }
            self.out.extend(best.unwrap_or_default());
        }
    }
}

/// The slots of `slots` that `seg` may fill: those of its tag whose
/// qualifier it carries, else those of its tag without a qualifier. None
/// when every slot of its tag names qualifiers and `seg` carries none of
/// them: the structure check reports that segment (UCS 15) — unless there is
/// only one slot, whose qualifier is then judged as a value (UCD 12).
fn slots_for<'m>(seg: &AssembledSegment, slots: &'m [MigSegment]) -> Vec<&'m MigSegment> {
    let tagged: Vec<&MigSegment> = slots.iter().filter(|s| s.id == seg.tag).collect();
    let carries = |slot: &MigSegment| {
        qualifier(slot).map(|(e, c, codes)| {
            seg.elements
                .get(e)
                .and_then(|el| el.get(c))
                .is_some_and(|v| codes.contains(v))
        })
    };
    let qualified: Vec<&MigSegment> = tagged
        .iter()
        .copied()
        .filter(|slot| carries(slot) == Some(true))
        .collect();
    if !qualified.is_empty() {
        return qualified;
    }
    let open: Vec<&MigSegment> = tagged
        .iter()
        .copied()
        .filter(|slot| carries(slot).is_none())
        .collect();
    if !open.is_empty() {
        return open;
    }
    match tagged.as_slice() {
        [only] => vec![*only],
        _ => Vec::new(),
    }
}

/// A MIG example code (`XYZ`) stands for any value.
fn is_placeholder(code: &CodeDefinition) -> bool {
    code.name == "Beispielcode" || code.value == "XYZ"
}

/// The real codes of an element; empty when it takes any value.
fn codes(d: &MigDataElement) -> Vec<&str> {
    if d.codes.iter().any(is_placeholder) {
        return Vec::new();
    }
    d.codes
        .iter()
        .map(|c| c.value.as_str())
        .filter(|v| !v.is_empty())
        .collect()
}

/// A slot's qualifier: its first coded element, as (element, component, codes).
fn qualifier(slot: &MigSegment) -> Option<(usize, usize, Vec<String>)> {
    let simple = slot.data_elements.iter().map(|d| (d.position, 0, d));
    let composite = slot.composites.iter().flat_map(|c| {
        c.data_elements
            .iter()
            .map(move |d| (c.position, d.position, d))
    });
    simple
        .chain(composite)
        .filter(|(_, _, d)| !codes(d).is_empty())
        .min_by_key(|(e, c, _)| (*e, *c))
        .map(|(e, c, d)| (e, c, codes(d).into_iter().map(String::from).collect()))
}

fn required(spec: Option<&str>, std: Option<&str>) -> bool {
    matches!(spec.or(std).map(str::trim), Some("M" | "R"))
}

/// Check `seg`'s data elements against `slot`.
fn check_elements(
    seg: &OwnedSegment,
    slot: &MigSegment,
    counter: u32,
    settings: SyntaxSettings,
    accepted: &AcceptedCodes,
    out: &mut Vec<SyntaxFinding>,
) {
    let number = slot.number.as_deref().unwrap_or("");
    let also = |e: usize, c: usize| accepted.get(&(number.to_string(), e, c));
    let defined = slot
        .data_elements
        .iter()
        .map(|d| d.position)
        .chain(slot.composites.iter().map(|c| c.position))
        .max()
        .map_or(0, |p| p + 1);
    let last_filled = seg
        .elements
        .iter()
        .rposition(|e| e.iter().any(|c| !c.is_empty()))
        .map_or(0, |p| p + 1);
    if last_filled > defined {
        out.push(finding(
            SyntaxScope::Segment,
            "16",
            seg,
            counter,
            None,
            format!(
                "{} has {last_filled} data elements, the MIG defines {defined}",
                seg.id
            ),
        ));
    }

    for d in &slot.data_elements {
        let element = seg.elements.get(d.position);
        let at = d.position as u32 + 2;
        let value = element.and_then(|e| e.first()).map_or("", String::as_str);
        if element.is_some_and(|e| e.iter().skip(1).any(|c| !c.is_empty())) {
            out.push(finding(
                SyntaxScope::Element,
                "16",
                seg,
                counter,
                Some((at, None)),
                format!(
                    "{} {} is a simple data element but has components",
                    seg.id, d.id
                ),
            ));
        }
        check_value(
            seg,
            d,
            value,
            counter,
            (at, None),
            settings,
            also(d.position, 0),
            out,
        );
    }

    for c in &slot.composites {
        let at = c.position as u32 + 2;
        let components: &[String] = seg.elements.get(c.position).map_or(&[], Vec::as_slice);
        if components.iter().all(String::is_empty) {
            if required(c.status_spec.as_deref(), c.status_std.as_deref()) {
                out.push(finding(
                    SyntaxScope::Element,
                    "13",
                    seg,
                    counter,
                    Some((at, None)),
                    format!("{} {} ({}) is missing", seg.id, c.id, c.name),
                ));
            }
            continue;
        }
        let defined = c
            .data_elements
            .iter()
            .map(|d| d.position)
            .max()
            .map_or(0, |p| p + 1);
        let last_filled = components
            .iter()
            .rposition(|v| !v.is_empty())
            .map_or(0, |p| p + 1);
        if last_filled > defined {
            out.push(finding(
                SyntaxScope::Element,
                "16",
                seg,
                counter,
                Some((at, None)),
                format!(
                    "{} {} has {last_filled} components, the MIG defines {defined}",
                    seg.id, c.id
                ),
            ));
        }
        for d in &c.data_elements {
            let value = components.get(d.position).map_or("", String::as_str);
            check_value(
                seg,
                d,
                value,
                counter,
                (at, Some(d.position as u32 + 1)),
                settings,
                also(c.position, d.position),
                out,
            );
        }
    }
}

/// Check one simple or component data element's value.
fn check_value(
    seg: &OwnedSegment,
    d: &MigDataElement,
    value: &str,
    counter: u32,
    at: (u32, Option<u32>),
    settings: SyntaxSettings,
    also: Option<&Accepted>,
    out: &mut Vec<SyntaxFinding>,
) {
    let mut push = |code: &str, detail: String| {
        out.push(finding(
            SyntaxScope::Element,
            code,
            seg,
            counter,
            Some(at),
            detail,
        ));
    };
    let name = format!("{} {} ({})", seg.id, d.id, d.name);
    if value.is_empty() {
        if required(d.status_spec.as_deref(), d.status_std.as_deref()) {
            push("13", format!("{name} is missing"));
        }
        return;
    }
    if let Some(bad) = value.chars().find(|&c| !settings.charset.allows(c)) {
        push(
            "21",
            format!("{name} holds '{bad}', which its syntax level does not allow"),
        );
        return;
    }
    if let Some(format) = d.format_spec.as_deref().or(d.format_std.as_deref()) {
        if let Some((code, detail)) = check_format(value, format, settings.decimal_mark) {
            push(code, format!("{name} '{value}': {detail}"));
            return;
        }
    }
    let allowed = codes(d);
    if !allowed.is_empty() && !allowed.contains(&value) && !also.is_some_and(|a| a.accepts(value)) {
        push(
            "12",
            format!(
                "{name} '{value}' is none of the MIG's codes {}",
                allowed.join(", ")
            ),
        );
    }
}

/// A format `an..35`, `an3`, `n..15`, `n5`, `a3`: its kind, its length, and
/// whether the length is a maximum.
fn parse_format(format: &str) -> Option<(&str, usize, bool)> {
    let format = format.trim();
    let kind_end = format.find(|c: char| !c.is_ascii_alphabetic())?;
    let (kind, rest) = format.split_at(kind_end);
    let (upto, digits) = match rest.strip_prefix("..") {
        Some(d) => (true, d),
        None => (false, rest),
    };
    Some((kind, digits.parse().ok()?, upto))
}

/// The UCD code and reason when `value` breaks `format`.
fn check_format(value: &str, format: &str, decimal_mark: char) -> Option<(&'static str, String)> {
    let (kind, length, upto) = parse_format(format)?;
    let counted = match kind {
        "n" => {
            let unsigned = value.strip_prefix('-').unwrap_or(value);
            if let Some(c) = unsigned
                .chars()
                .find(|c| !c.is_ascii_digit() && *c != decimal_mark)
            {
                return Some(match c {
                    '.' | ',' => (
                        "19",
                        format!("'{c}' is not the decimal mark '{decimal_mark}'"),
                    ),
                    _ => ("37", format!("'{c}' in a numeric data element ({format})")),
                });
            }
            let marks = unsigned.matches(decimal_mark).count();
            if marks > 1 || unsigned.ends_with(decimal_mark) {
                return Some(("19", format!("not a decimal number ({format})")));
            }
            if unsigned.starts_with(decimal_mark) {
                return Some(("38", "no digit before the decimal mark".to_string()));
            }
            unsigned.chars().filter(char::is_ascii_digit).count()
        }
        "a" => {
            if let Some(c) = value.chars().find(char::is_ascii_digit) {
                return Some((
                    "37",
                    format!("'{c}' in an alphabetic data element ({format})"),
                ));
            }
            value.chars().count()
        }
        _ => value.chars().count(),
    };
    if counted > length {
        return Some((
            "39",
            format!("{counted} characters, {format} allows {length}"),
        ));
    }
    if !upto && counted < length {
        return Some((
            "40",
            format!("{counted} characters, {format} requires {length}"),
        ));
    }
    None
}

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

    #[test]
    fn formats() {
        assert_eq!(check_format("12345", "n5", '.'), None);
        assert_eq!(check_format("1234", "n5", '.').map(|f| f.0), Some("40"));
        assert_eq!(check_format("123456", "n..5", '.').map(|f| f.0), Some("39"));
        assert_eq!(check_format("-12.5", "n..3", '.'), None);
        assert_eq!(check_format(".5", "n..3", '.').map(|f| f.0), Some("38"));
        assert_eq!(check_format("5.", "n..3", '.').map(|f| f.0), Some("19"));
        assert_eq!(check_format("1,5", "n..3", '.').map(|f| f.0), Some("19"));
        assert_eq!(check_format("1A", "n..3", '.').map(|f| f.0), Some("37"));
        assert_eq!(check_format("Z1", "a..3", '.').map(|f| f.0), Some("37"));
        assert_eq!(check_format("ÄBC", "an3", '.'), None);
        assert_eq!(check_format("AB", "an3", '.').map(|f| f.0), Some("40"));
    }

    #[test]
    fn charsets() {
        assert!(Charset::Unoc.allows('ä'));
        assert!(!Charset::Unoc.allows('€'));
        assert!(!Charset::Unoc.allows('\n'));
        assert!(!Charset::Unoa.allows('a'));
        assert!(Charset::Unob.allows('a'));
    }
}