gridcraft/crates/xlsx/src/passthrough.rs

//! Pass-through of package content GridCraft does not model, so that open → edit → save keeps it.
//!
//! On read, the raw bytes of unmodelled parts are kept with their content type and their
//! relationship graph, per owner: the package root (custom document properties, ribbon
//! customisation, add-in task panes), the workbook (custom XML, links to external workbooks with
//! their `<externalReferences>`, data connections, XML maps) and each sheet (query tables, cells
//! mapped to XML elements, form controls, ActiveX controls, OLE objects, the sheet background and
//! printer settings, with the sheet's `<picture>`, `<oleObjects>` and `<controls>` XML, the `r:id`
//! of `<pageSetup>`, the legacy VML shapes those controls need and the pictures of the headers and
//! footers, `&G`, with their `<legacyDrawingHF>`). The workbook also keeps its
//! `<fileVersion>` and the calcFeatures extension, which Excel needs to trust the saved values.
//!
//! On write, kept parts get fresh part names where needed and new relationship ids; the `r:id`s in
//! kept XML are rewritten to match. Kept content is conservative:
//! - Anything whose relationship graph reaches a part GridCraft reads and rewrites (worksheets,
//!   drawings, charts, tables, PivotTables, comments, slicers…) is not kept: it would point at
//!   content that no longer exists in the same form.
//! - External relationship targets (the path of a linked workbook) are kept as text and never
//!   opened, followed or refreshed. A relative path to a linked workbook is rewritten when the
//!   workbook is saved into another folder ([`rebase_links`]), so it still names the same file.
//! - Sheet content lives on the sheet: deleting the sheet drops it, Copy Sheet doesn't copy it.
//!   Cell references inside kept parts (a control's linked cell, a query table's range, a cell
//!   mapped to an XML element) are kept as they were, not adjusted to inserted or deleted rows.
//! - Strict Open XML files are not passed through (their namespaces differ from what we write).
//! - Sizes are bounded: per part and per workbook.

use std::borrow::Cow;
use std::collections::{HashMap, HashSet, VecDeque};
use std::fmt::Write as _;
use std::path::{Component, Path};
use std::sync::Arc;

use gridcraft_model::{Preserved, PreservedPart, PreservedRel, PreservedSheet, PreservedXml, Sheet, Visibility, Workbook};
use quick_xml::events::attributes::Attribute;
use quick_xml::events::{BytesStart, Event};
use quick_xml::name::QName;
use quick_xml::{Reader, Writer};

use crate::IoError;
use crate::package::{self, normalize};
use crate::read::Ctx;
use crate::write::{NS_REL, Out, Rels, XML_DECL};
use crate::xml::{self, esc_attr};

mod read;
mod rebase;
mod write;

pub use read::KeptReader;
pub use rebase::{SheetKept, rebase_links};
#[cfg(test)]
pub(crate) use write::sheet_names_in;
pub use write::{KeptOut, SheetIds, WorkbookKept, workbook_elements, write_package_rels, write_workbook};

/// Most bytes kept for one workbook.
pub const MAX_KEPT_TOTAL: usize = 256 * 1024 * 1024;
/// Largest single kept part.
pub const MAX_KEPT_PART: u64 = 64 * 1024 * 1024;
/// Most parts kept for one workbook.
const MAX_KEPT_PARTS: usize = 4096;
/// Most relationships read from one rels part.
const MAX_RELS: usize = 10_000;
/// Largest relationships, content-types or VML part read here.
const MAX_META: u64 = 16 * 1024 * 1024;
/// Longest chain of relationships followed from an owner.
const MAX_CHAIN: usize = 16;
/// Largest kept XML fragment of a sheet.
const MAX_FRAGMENT: usize = 4 * 1024 * 1024;
/// Deepest element nesting accepted in a kept fragment.
const MAX_FRAGMENT_DEPTH: usize = 256;

#[cfg(test)]
thread_local! {
    /// A smaller kept-bytes allowance for tests on this thread.
    pub(crate) static KEPT_TOTAL_FOR_TEST: std::cell::Cell<usize> = const { std::cell::Cell::new(MAX_KEPT_TOTAL) };
}

fn kept_total() -> usize {
    #[cfg(test)]
    {
        KEPT_TOTAL_FOR_TEST.with(std::cell::Cell::get)
    }
    #[cfg(not(test))]
    {
        MAX_KEPT_TOTAL
    }
}

const NS_REL_STRICT: &str = "http://purl.oclc.org/ooxml/officeDocument/relationships";
const CT_VML: &str = "application/vnd.openxmlformats-officedocument.vmlDrawing";

/// Relationship kinds (last segment of the type) kept from the package root.
const PACKAGE_ROOTS: &[&str] = &["custom-properties", "extensibility", "webextensiontaskpanes"];
/// Relationship kinds kept from the workbook without a workbook element.
const WORKBOOK_ROOTS: &[&str] = &["customXml", "connections", "xmlMaps"];
/// Relationship kinds kept from a sheet without a sheet element.
const SHEET_ROOTS: &[&str] = &["queryTable", "tableSingleCells"];
/// Sheet relationship kinds that mean the sheet XML may hold elements worth keeping.
const SHEET_ELEMENT_KINDS: &[&str] = &["oleObject", "ctrlProp", "control", "image", "package", "printerSettings"];
/// Attributes of `<fileVersion>` (CT_FileVersion).
const FILE_VERSION_ATTRS: &[&str] = &["appName", "lastEdited", "lowestEdited", "rupBuild", "codeName"];
/// The workbook extension listing the calculation features the saved values were computed with.
const CALC_FEATURES_URI: &str = "{B58B0392-4F1F-4190-BB64-5DF3571DCE5F}";
const NS_CALC_FEATURES: &str = "http://schemas.microsoft.com/office/spreadsheetml/2018/calcfeatures";
/// Kept sheet elements, in schema order (all after `<legacyDrawing>`, before `<tableParts>`).
const SHEET_SLOTS: &[&str] = &["legacyDrawingHF", "picture", "oleObjects", "controls"];

/// Relationship kinds GridCraft reads and rewrites, or that depend on content it rewrites:
/// kept content whose graph reaches one of them is dropped.
const REWRITTEN: &[&str] = &[
    "officedocument",
    "core-properties",
    "extended-properties",
    "worksheet",
    "chartsheet",
    "dialogsheet",
    "xlmacrosheet",
    "drawing",
    "chart",
    "chartex",
    "chartstyle",
    "chartcolorstyle",
    "chartusershapes",
    "table",
    "pivottable",
    "pivotcachedefinition",
    "pivotcacherecords",
    "comments",
    "vmldrawing",
    "threadedcomment",
    "person",
    "sharedstrings",
    "styles",
    "theme",
    "sheetmetadata",
    "calcchain",
    "vbaproject",
    "rdrichvalue",
    "rdrichvaluestructure",
    "rdrichvaluetypes",
    "rdrichvaluewebimage",
    "rdarray",
    "rdsupportingpropertybag",
    "rdsupportingpropertybagstructure",
    "richvaluerel",
    "slicer",
    "slicercache",
    "timeline",
    "timelinecache",
    "powerpivotdata",
];

/// Relationship kinds a chart sheet reaches that are otherwise rewritten: its drawing, chart and
/// their style, colour and shape parts are kept byte for byte (the chart sheet is not modelled).
const CHART_SHEET_GRAPH: &[&str] = &["chartsheet", "drawing", "chart", "chartex", "chartstyle", "chartcolorstyle", "chartusershapes"];

/// Part-name prefixes (lower case) GridCraft's writer never produces: kept parts there keep their
/// original names. Anything else is renamed so it can't collide with written parts.
const OWN_DIRS: &[&str] = &[
    "customxml/",
    "docprops/custom.xml",
    "customui/",
    "xl/externallinks/",
    "xl/connections.xml",
    "xl/xmlmaps.xml",
    "xl/querytables/",
    "xl/ctrlprops/",
    "xl/embeddings/",
    "xl/activex/",
    "xl/webextensions/",
    "xl/printersettings/",
];

fn kind(rel_type: &str) -> &str {
    rel_type.rsplit('/').next().unwrap_or("")
}

fn is_kind(r: &PreservedRel, kinds: &[&str]) -> bool {
    let k = kind(&r.rel_type);
    kinds.iter().any(|x| x.eq_ignore_ascii_case(k))
}

fn rewritten(r: &PreservedRel) -> bool {
    !r.external && REWRITTEN.contains(&kind(&r.rel_type).to_ascii_lowercase().as_str())
}

fn rels_part_name(part: &str) -> String {
    let (dir, file) = package::split_dir(part);
    if dir.is_empty() { format!("_rels/{file}.rels") } else { format!("{dir}/_rels/{file}.rels") }
}