Changes in version 0.1.2                        

Bug fix release.

  - Fixed a crash in the flow engine when a block's height equalled the
    space left on the page to within floating-point error: split() then
    reported the whole block as fitting (head = <block>, tail = NULL)
    and .place_block() went on to measure the NULL tail ("attempt to
    apply non-function"). .place_block() now (a) treats a block as
    fitting when its height exceeds the remaining space by less
    than 1e-9 pt, matching the tolerance the split() methods already
    use, and (b) stops cleanly when a split returns no tail, instead of
    starting a page for it. Affected block_para(), block_pre() and
    block_table() alike; nmw worked around it. Regression-tested in
    inst/tinytest/test_flow.R.
  - .block_seq() (used by block_table()/block_matrix()) now returns tail
    = NULL rather than an empty sequence block when a split consumes
    every sub-block, which also removes a spurious trailing blank page.

                 Changes in version 0.1.1 (2026-07-30)                  

Documentation-only release for the CRAN resubmission (no code changes).

  - Added the missing \value sections to all exported functions' help
    pages (CRAN review): flow_run(), flow_add(), remaining(),
    running_header()/running_footer(), block_keep(), block_spacer(),
    block_rule(), frame_set(), sp_page(), sp_close(), sp_trace().
  - Expanded the existing \value sections to describe the structure
    (class) and meaning of every return value, e.g. what a "block" is (a
    list with measure/draw/keep/split components consumed by the flow
    engine) and that document-handle returns are environments updated in
    place.

                        Changes in version 0.1.0                        

P0 — Core canvas, measurement, and flow engine (done)

First working slice of the device-based report engine (see
PLAN-simPDF.md).

  - Document / canvas: sp_new() opens a pdf()/cairo_pdf() device with a
    point coordinate system; sp_page(), sp_close(), frame_set().
  - Measurement: sp_width() / sp_height(). AFM-exact for the core fonts
    and linear in size for fractional sizes too (measured at a reference
    size and scaled, working around the pdf() device's integer font-size
    rounding in strwidth).
  - Flow engine: flow_run() / flow_add() place measured blocks,
    advancing the cursor by real height and paginating automatically;
    block_keep() (keep-together), block_spacer(), block_rule(),
    remaining(), running_header() / running_footer().
  - Content blocks: block_para() (word-wrapped, aligned) and block_pre()
    (monospace capture.output() dumps, auto-shrunk to fit the frame
    width).
  - QA: sp_trace() / sp_overlaps() record and detect overlapping text
    bounding boxes. Tests in inst/tinytest/ assert a 200+ line synthetic
    report renders with zero overlap across multiple pages.

This removes the dead-reckoned-coordinate overlap of the nmw diagnostic
reports (see inst/examples/nmw_params.R).

P1 - Tables and matrices (done)

  - block_table(): renders a data.frame/matrix with column widths
    measured from the content, numeric columns right-aligned, a header
    that repeats on every page, a rule under the header, and automatic
    row-boundary page splitting. If the natural width exceeds the frame,
    the font shrinks uniformly to fit. Row names become a leading label
    column.
  - block_matrix(): renders a wide matrix by splitting its data columns
    into groups that each fit the page width (row labels repeated in
    every group), stacked vertically -- replacing hand-placed column
    loops such as PrinTxt(row, i * 8, ...) that overflow for many
    columns.
  - Tracer now records the drawn text of each box, so tests assert
    header-repeat directly. inst/tinytest/test_table.R covers a 200-row
    split table (header repeated once per page) and a 20-column matrix
    split into groups, both with zero overlap.

P1 fixes (from an adversarial review pass)

  - block_table/block_matrix shrink-to-fit now scales the cell padding
    by the same factor as the font and column widths. Previously only
    font and widths were scaled, so at strong shrink (tot > 2 * frameW)
    the last column's text spilled past the frame's right edge (and a
    right-aligned first column past the left). Regression-tested.
  - A 0-row data.frame/matrix now renders its column headers instead of
    a blank band: .tabularize preserves the column count when there are
    no rows, and .natural_widths guards the empty-column max() so header
    = FALSE 0-row input no longer warns. Regression-tested.

P2 - Inline graphics (done)

  - block_plot(expr, height): evaluates a base-R plotting expression
    (scatter, histogram, QQ, ...) inside a reserved band of the flowing
    page via par(fig), then restores the text canvas. Plots sit inline
    with measured text, are kept with their headings via block_keep, and
    participate in pagination. The expression is captured unevaluated
    and run at draw time in the caller's environment.
  - block_image(img, height): embeds a raster, array/matrix, or PNG/JPEG
    file (png/jpeg in Suggests) via rasterImage.
  - Fixed a page-desync bug found in visual verification: the canvas is
    now restored with par(op) alone (no extra plot.new()), so a page
    break after a block_plot advances the physical PDF page in lock-step
    with doc$page_no. Regression-tested by comparing doc$page_no to the
    physical /Type /Page count. See inst/examples/nmw_diagnostics.R.

Authorship, signatures, and diagrams (done)

For downstream packages that build reports on simPDF:

  - block_authorship() and block_signature(): a "Prepared by" metadata
    block and visible approval/signature lines (role caption + dated
    line). The signing rectangles are recorded on doc$sig.
  - sp_add_sig_fields(): a finishing pass that inserts interactive Adobe
    Acrobat digital-signature fields (AcroForm /Sig) into the written
    PDF via a pure base-R incremental update (no external tools) --
    generalised from NonCompart::addSigFieldNCA, and driven by the
    doc$sig rectangles so the clickable fields land exactly on the
    visible lines. The result is one-click-signable in the free Acrobat
    Reader and passes qpdf --check.
  - block_flow_diagram(): a model-development flow diagram where each
    node is a box measured to its text, laid out as a tidy tree
    (children centred under parents) and scaled uniformly to fit the
    frame -- so more models shrink the diagram instead of cluttering it
    (unlike the dead-reckoned legacy Outline).

nmw report migration (drafted + validated)

All 8 nmw diagnostic reports now have simPDF implementations
(S2-Parameters plus the remaining seven: S1-OFV, S3-Predictions,
S4-Residuals, S5-EBE, SA-Input, SB-Output, SC-IndiPK). Each keeps the
parsing/statistics verbatim and replaces the dead-reckoned
pdf()+PrinTxt/split.screen layout with a blocks list + flow_run()
(text/tables) and sp_figure_page() (per-subject / per-eta plot pages).
Validated on the bundled 2006 example (all 8 generate) and across real
runs (97/98 report generations on 14 models); the S3-Predictions
"zero-length labels" crash on subjects with no observations is fixed
(was 11/18, now 14/14). These functions live with the nmw package (which
gains simPDF in Imports); simPDF gained sp_figure_page() to host the
full-page multi-panel diagnostic pages.

Next

  - Land the migrated report_* functions in nmw (replace bodies, drop
    the old PrepPDF/PrinTxt/AddPage/PrinMTxt/ClosePDF from
    pdf_report.R).
  - Optional pagination for very large flow diagrams (currently
    scale-to-fit).