Files
pdfcraft/crates/forms
mtlcl410andClaude Opus 5.5 6918d0be21 M6: a static XFA form whose data fills thousands of fields opens in seconds (#4638)
Opening a static XFA form gives its fields the values in its datasets
(a form filled in by another application). Each value went through
forms::set_value, which reads the whole field tree to find the field,
writes it, and recalculates, reading the tree again. With n values that
is about 2n readings of n fields: quadratic. A public 136-page form with
about 6,000 fields took 261 s to open, a 95-page one with about 4,000
took 112 s; 8,000 synthetic fields took over ten minutes.

forms::set_values fills a batch as set_value fills each in turn, reading
the form once and recalculating once after the last value (the
calculation order, ISO 32000-2 Table 224 /CO, runs once over the final
values). Fields with the same fully qualified name (12.7.4.2) still share
the value. A name the batch already wrote, or a twin of it, makes the
form be read again first, so each write starts from what the earlier one
left, as one by one. Each value reports its own result; a failed
recalculation fails the batch, as it fails set_value. The engine collects
the changed values and fills them with set_values, finding fields by name
through a map rather than a scan.

The two forms now open in 4 s and 3 s, with the same fields, values,
warnings and document information as before; 8,000 synthetic fields open
in 1.5 s.

Tests: set_values matches set_value one by one (values, twins, the
calculated total, every appearance, saved; refusals for an unknown name,
a read-only field, an unknown option and clearing a no-toggle-off radio
group the batch selected); it reads the form a fixed number of times; the
engine takes 59 data values on a 60-field static form exactly as filling
them one by one does (the read-only one refused); and opening 800 fields
takes under 16 times as long as 100 (about 4x now, about 48x before).


Claude-Session: https://claude.ai/code/session_01CBVyQEwtmEw3mS1MoAiwrK

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-25 10:02:35 -04:00
..

pdfcraft-forms

Interactive forms (AcroForm, ISO 32000-2 §12.7). Layer L3; depends on pdfcraft-cos and pdfcraft-fonts.

API

let all: Vec<Field> = fields(&doc);                         // terminal fields with widgets, options, flags, /DA
let snapshot = fields_with_adopted(&doc);                     // fields and page-only adoption count in one pass
let all = snapshot.items;
let adopted = snapshot.adopted_page_fields;
set_value(&mut doc, "name", &FieldValue::Text("Ada".into()))?;
set_value(&mut doc, "agree", &FieldValue::Check(true))?;
set_value(&mut doc, "size", &FieldValue::Radio(Some("L".into())))?;
set_value(&mut doc, "toppings", &FieldValue::Choice(vec!["Ham".into()]))?;
// Many values: as set_value one by one, but reading the form once and recalculating once.
let results: Vec<Result<(), FormError>> = set_values(&mut doc, &[("name".into(), FieldValue::Text("Ada".into()))])?;
reset(&mut doc, None)?;                                      // Clear form (back to /DV)

// Prepare a form
let name = add_field(&mut doc, 0, [72.0, 700.0, 216.0, 722.0], &NewField::Text { multiline: false }, None)?; // "Text1"
set_props(&mut doc, &name, &FieldProps { name: Some("email".into()), required: Some(true), ..Default::default() })?;
// A partial appearance edit keeps each widget's other styles and custom /DA font.
set_props(&mut doc, "email", &FieldProps {
    appearance: Some(LookPatch { width: Some(3.0), fill: Some(None), ..Default::default() }),
    ..Default::default()
})?;
delete_field(&mut doc, "email")?;

// A table of text fields (view space, y down), laid out and named before any of it is added:
// refused whole for a cell under MIN_FIELD_SIDE or a name with a period, repeated or taken.
let names = all.iter().map(|f| f.name.as_str());
let base = table::free_base(names.clone());                    // "Table1", or the next one free
let mut grid = table::FieldTable::new([40.0, 240.0, 240.0, 360.0], &base);
grid.set_columns(2, &base);
let cells: Vec<table::Cell> = grid.cells_for(names)?;      // "Table1_1", "Table1_c2_1", …

Behaviour

  • Inheritable attributes (/FT, /Ff, /V, /DV, /DA, /Q, /MaxLen) are resolved down the field tree; the form's /DA and /Q are the last fallback.
  • Text and choice fields get new appearance streams (/Tx BMC … EMC): auto-size (0 Tf), comb (cells parted in the border colour), multiline wrapping, password masking, quadding, /MK background and border, list-box selection. The /DA font is used when /DR has it as a simple font or a composite font with a predefined Unicode CMap; otherwise Helvetica (WinAnsi). A value or button caption the font can't hold (Chinese in a Helvetica field, a character beyond U+FFFF in a UCS-2 CMap) is kept, as Acrobat keeps it: /V is stored as typed, the appearance draws ? for those characters, the widget is marked with the first one (pdfcraft_fonts::UNDRAWN_MARK, cleared once all of it can be drawn) and the form sets /NeedAppearances, so other viewers draw the field from its value with fonts that have them. Not in a document that declares PDF/A, which forbids that flag (declares_pdfa). undrawn_fields lists the marked fields; the engine reports each edit that draws one, and flattening refuses to make that ? permanent.
  • Displaying a form that leaves its appearances to the viewer (/NeedAppearances, or a text or choice widget with no /AP /N): with_field_appearances returns a copy with those widgets drawn from their values, as typing them would draw them; the engine shows and prints that copy and saves the file as it was (#3470). with_field_appearances_cached keeps each widget's drawing while the objects it was drawn from are the same (FieldAppearanceCache), so typing in one field redraws that field, not the form. Check boxes and radio buttons with no appearance for their state are drawn from /MK (border, background, and when on the /MK /CA ZapfDingbats symbol in the /DA colour), and push buttons with none from /MK, /BS and their caption.
  • Check boxes and radio buttons switch /V and each widget's /AS. Widgets with no appearances get PdfCraft's own (a drawn check mark or dot; no symbol font needed).
  • Imported /MaxLen 0 is treated as unlimited (also clearing an inherited limit), while the original entry is preserved. Positive limits still apply.
  • Values are validated (options, MaxLen, single vs multi-select, NoToggleToOff, read-only) and a failed call changes nothing.
  • FieldProps::appearance applies a LookPatch independently to each widget. None keeps a property; Some(None) removes a border/fill colour. Unchanged /DA, custom font resources, widget-specific /DA overrides, indirect /MK and /BS contents, dash arrays and unknown dictionary keys are preserved. It is mutually exclusive with the complete look replacement. Use the engine's Edit::Batch to change several fields atomically in one undo step.
  • /RV (rich text) is removed when a value is set so it can't contradict /V.
  • New fields follow Acrobat: names Text1, Check Box1, Group1 (radio groups), Dropdown1, List Box1, Button1, Date1, Signature1; /Helv is added to /DR; appearances are drawn at once. Radio buttons joining a group become kid widgets with their own export state. Renaming changes only the last part of a dotted name.
  • Renaming a field renames it in every Calculate script that reads it (af::rename_in_calculation): AFSimple_Calculate field lists, simplified field notation (the BVCALC comment and its translation), and getField calls whose argument is the name as one string literal. The rest of each script is kept as written. Names built at run time, a script stream that isn't UTF-8, other events' scripts and a name another field still has are left alone. calculations_changed compares two readings of a form to say which formulas an edit rewrote.
  • The calculation order (/CO, §12.7.3): calculation_order lists the calculated fields in the order recalculate runs them, with the fields each formula reads (Calculate::reads, af::calculation_reads; names built at run time aren't known) and what is wrong with the order: a field reading one that calculates later, a loop, a field /CO leaves out, and /CO entries that calculate nothing. CalculationOrder::fixed is Fix order: a stable topological sort that keeps the current order wherever it can and leaves loops as they are. set_calculation_order writes /CO afresh. A field that gets a calculation joins the end of the order; one whose calculation changes keeps its place.
  • Widget::caption reads a push button's normal label (/MK /CA, including indirect values). FieldProps { caption: Some((widget_index, text)), ..Default::default() } changes or clears that label, independently of the field name. The index is 0-based. A label-only edit redraws just that widget, copying shared/indirect /MK and appearance dictionaries before editing. Other widgets, actions, icons, rotation and unknown entries are preserved. Existing icon-only layouts keep hiding the caption; caption rendering retains the existing Helvetica/WinAnsi limits and reports unavailable characters as above.

Not yet

Rich-text values and exact appearance metrics/layout remain incomplete. JavaScript execution and FDF/XFDF exchange are integrated through the engine; this crate provides field actions and native AF rules. /MK /R rotation (0, 90, 180, 270) is drawn into the appearance.