- #2587: a frame turned by anything but a quarter turn keeps its size about its turned
centre; the box of its two mapped corners collapsed it (a 10 x 10 square turned 45° became
14.14 x 0, and the OLE header with it) for EXPLODE, REFEDIT, paste and array placement
(austin62jake, CodeRabbit). Moves, scales, mirrors and quarter turns keep the exact
corner mapping. `recase` reads only the rewritten digits and a few after them, not the
whole 310 line (austin62jake). The PR's rotated-frame test expects the kept size.
- #2585: an `output` that contradicts the path's extension is refused, as a device that
does is; it lost silently to the extension.
- #2608: the window's state is dropped when another dialog took its place (it missed a UI
pass), so reopening reads the drawing instead of an old unsent edit (CodeRabbit); TEXT and
ATTDEF get a one-line field (Enter commits), so a line break can't turn into a space on
save; the module comment says the shortcut menu opens it.
- #2584: a tilted LWPOLYLINE/HATCH run is flattened to at most a million vertices, then
its arcs stay chords (a hostile file of arcs asked for gigabytes; austin62jake). The
file-format-parity.md hunk is dropped: maintainers update the parity docs.
- #2471: another program's proxy objects are in the DWG loss report as its proxy entities
are (left out before 2013, or when their class can't be named); they went silently.
- #1626: a drawing saved at a view keeps it in the desktop app and when it has xrefs: the
app's first frame and the xref pass both fitted the extents over it. OPEN's result says
`saved_view`; one `view::saved_view` replaces the copy of `usable_view` in files.rs.
A misplaced doc comment goes back to its test.
- #2028: a plot rotation (73) outside 0-3 is written back as 0, not as read.
- #1026: the binary code-page retry drops the first pass's tags before it runs (the peak
doubled near the tag limit); a float spelling of an integer group clamps to what the
width holds unsigned too, so 65535.0 reads as 65535 does, not 32767.
- #2658: the reference names Undo by the last command ("Undo Group of commands"); Undo
matched nothing and went to the end of Edit. Ordering and its test share one rule.
- #2612: its test's Insert gets `clip: None` (XCLIP landed after it was written).
render_look's DARK repeated the rasteriser's default background; both now
read RasterOptions::DARK_BACKGROUND. ECHO and LISTED_TABLES are renamed
MAX_ECHO_CHARS and MAX_LISTED_TABLES. No change in behaviour.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
main added view twist. Kept from both: a drawing opens on the view it was saved at, with its saved twist; one that stores none fits its extents through main's twisted box. The VPORT reader takes main's twisted view centre and keeps this PR's first-*ACTIVE view; the writer writes the saved view, else main's twisted fit, with the fallback still fitting the width through the record's aspect.
A plot style alone drew black lines on the dark canvas: the tables that
change colours print colour 7 (and, in monochrome.ctb, every colour)
black. With `plot_style` and no `background`, the render now draws on
white, as a plot prints on paper; a given background still wins.
`plot_style: "none"` (or "") draws every object with its own
properties, also in a layout that shows its own plot styles. An unknown
table's error lists the built-in tables and at most ten of the
drawing's own. The app's ui.render test now draws a red line, so it
fails if the plot style isn't passed on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The MCP render tool and ui.render, in the app and headless, always drew
on the dark canvas with no plot style, so a PNG placed on a page needed
the SVG export instead.
They now take `background` ("dark", the default; "white", "paper" or
"#rrggbb") and `plot_style` (a built-in plot style table such as
monochrome.ctb or grayscale.ctb, or one the drawing holds). Colour 7 is
drawn black on a light background and white on a dark one, as on a plot.
An unknown value is refused with what is accepted. Both copies of
ui.render read the options through one function in the engine
(render_look). The render tool's description also says it returns an
image block, not text.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The renderer draws a zero-width frame as a line, so leaving it out of the hit geometry made a
visible object unpickable. Also corrects where the corner tolerance switches to ulps (4 ulps
pass 1e-6 near 1.1e9).
A matrix every frame accepts (a rectangular array, plain translations) now needs no
look at the objects, and an empty matrix list never touches the drawing. ARRAYPATH
scans the selection once instead of once per item.
Covered by the existing refusal tests; copy_all is not reachable without a drawing
through any command, so there is no no-drawing test.
Putting the frame in prims() also made it a TRIM/EXTEND edge, an object snap
target and dimension geometry. prims() is back to nothing for it, the extents
arm is restored, and select::hit_polylines adds the frame's outline for picking
(none for a frame with no width or height or a bad corner).
Tests: clicking_an_ole_frame_edge_selects_it (now also a frame with no width),
ole_frame_is_not_a_trim_edge_or_snap_target.
copy_all scanned the selection for an OLE frame once per matrix before copying.
It now looks at the objects once and then only tests each matrix.
Covered by the existing refusal tests (classic_polar_array_refuses_a_turned_frame_but_not_an_upright_one).
An OLE2FRAME is drawn but prims() gave nothing for it, so hit testing never saw
it and a click on its outline picked nothing. prims() now returns the frame's four
edges from the same corner helper the extents use, so the extents arm is gone.
The renderer keeps its closed-polyline arm (one joined outline, not four segments).
Test: clicking_an_ole_frame_edge_selects_it.
ALIGN builds its matrix from two angles that agree only to rounding, so a large
scale leaves a rotation part of about scale * 1e-16 that an absolute 1e-12 bound
refused as a turn. The bound now grows with the scale; a real turn is still refused.
Test: ole_frame_accepts_a_large_scale_with_rounding_noise_in_its_rotation.
main's #2673 runs the DWG writer in a child process. The kept version now goes with the request
(encode_kept_noted, worker protocol 2), so the worker reads the bridge DXF as that version too.
The proxies proxy_classes leaves out are now counted in Unwritten, as ACAD_PROXY_OBJECT and
ACAD_PROXY_ENTITY.
- objects are found through a handle index when reading and writing clips, and removed in one pass (no scan per reference)
- the writer's handle seed starts past the clip objects, so a drawing with many clipped references gets no duplicate handles
- a clipped copy that runs out of the block budget marks the list truncated, like an unclipped one
- clipped block contents keep their own layers in plots
Review fixes from CodeRabbit on #2748:
- DVIEW TWist and view.set refuse to turn a locked active viewport ("The viewport is locked.")
- the untiled *Active VPORT record fits the extents rotated by the twist it writes, with the twist normalised
Review fixes from CodeRabbit on #2749:
- the image file is read through read_capped, so a file that grows after the size check can't exhaust memory
- an image cut by a layout viewport widens the sheet's bounds only by the part inside it, as other primitives do
- uploaded tiles of closed drawings are dropped also when the next list has no images
The heartbeat opened <instance>.lock with a create-and-truncate that followed symlinks, and fell back to std::fs::write, so another account sharing SAVEFILEPATH could plant a link and have the app truncate a file of the victim's. The lock is now opened with O_NOFOLLOW|O_NONBLOCK and must be a plain file, and unlocked writes replace the file through the atomic writer. Autosave files are created 0600 regardless of umask. Lock pruning parses the folder's autosave names once instead of once per lock.
- Every engine twist (DVIEW, view.set, opening a drawing, the untiled view) goes through norm_angle, so a tiny negative twist is 0, not a full turn.
- VIEW Window in a twisted view stores the window's size along the view's axes (View::twisted_box), so a restore shows all of it.
- A deferred tangent or perpendicular snap on an arc in a twisted layout viewport turns the arc's angles with its centre.
- The active viewport's cached drawing is rebuilt when its twist changes; moving and scaling it can't turn it.
AUDIT checked every layer's linetype and every style's text style by scanning the table; LAYERSTATE restore and restoring a named view's layer snapshot scanned the layer table for every saved layer; CAL scanned the variable map for every case-insensitive reference. Each now builds its name index once, keeping the same first-match and exact-match rules.
* MINSERT: a block reference repeated in rows and columns
MINSERT asks INSERT's questions, then the grid's as AutoCAD 2024 does
(measured with accoreconsole): the numbers of rows and columns (1 to
32767), the distance between rows or a unit cell when there is more than
one row, the distance between columns when there is more than one column,
then the attribute values once. A distance along an axis with one copy is
0, and one row by one column is a plain reference.
EXPLODE leaves a grid whole ("The object is minserted.") instead of
exploding its first copy and dropping the rest. MIRROR mirrors a block
reference as AutoCAD does, negating the X scale factor and a grid's
column distance, so a mirrored grid's rows stay on the right side. An
attribute with no default is asked without "<>".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* MINSERT JSON: refuse a distance that is not a number
`rowSpacing` and `columnSpacing` went through `f64_or`, which takes any
value that is not a number as the default, so `"rowSpacing": "15"` stored 0
and put every copy on top of the others without an error. They are now
checked as `rows` and `columns` are: absent or null is 0, a finite number is
kept, anything else is refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
INSERT of a file, IMPORT, paste and COMPARE export each checked every incoming layer, linetype, style and block name by scanning the growing destination vector, and restarted the anonymous block name search from 1 at every collision, so a compact drawing with many distinct names made the command quadratic. The four copies of that code become one helper (cmd/merge.rs) that looks names up in a set built once per table and continues each generated prefix from the last number it allocated. Xref loading does the same for its dependent symbols, and no longer rescans the block table after a fetch that imported nothing.
Dylan M. Taylor