A collection counts each piece of code a closure or frame keeps: its instructions, tables,
names, nested functions and its source text, each shared buffer once. Code that eval and the
Function constructor compile at run time is charged when it is made, after a heap reservation.
Function.prototype.toString reserves and charges the text it copies.
ToNumber on a hexadecimal string folded its digits into a double one at a
time, so a string past 2^53 could come out one unit off the literal with
the same digits. The lexer kept digits exactly in a u128 but rounded per
digit once the value passed 128 bits. Both now call one helper that keeps
the digits exactly while they fit, records whether any later digit is
set, and rounds once.
Declaration binding instantiation reserves an upper bound of what a script's,
eval's or function's variables and function declarations add to the heap
before making them, and each declaration and parameter takes a unit of fuel.
A long declaration list now ends with the heap limit without growing the
estimate past it.
Realm::set_limits recomputes when the next collection runs, so a lowered heap
limit applies to the next run instead of after the old threshold.
ToNumber skips the same white space and line terminators as the lexer, so
U+180E is no longer white space in a numeric string.
A new property or binding now adds its name's code units to the heap
estimate when it is inserted, so names declared by eval or Function
source move the estimate toward a collection. Function parameter and
catch bindings no longer count their entries twice.
A host function's error message reserves its size in the heap before
the error object is made, so a message larger than the heap ends the
run with the heap limit.
The Function constructor sums its arguments' lengths and checks the
source limit before flattening or copying any of them, and builds the
source from their code units, so a lone surrogate in an argument is
kept.
In script and eval code a finally block that ends normally leaves the completion value of its try or catch block (ECMA-262 5.1 section 12.14), so try { 1; } finally { 2; } completes with 1. A new GetCompletion instruction saves the value when the finally block starts, and SetCompletion sets it back when the block ends normally; a break, continue or throw out of the finally block still replaces it.
A string returned by a host function is charged to the heap, and one that does not fit after a collection ends the run with the heap limit.
A collection counted each property or binding at a fixed 64 bytes and
never counted the name's own buffer. Once the string that produced a long
name became garbage, the name kept its buffer alive while the estimate
dropped it, so a script could hold megabytes of names past the heap
limit. trace_object and trace_env now add each name's buffer, counting a
buffer that a flat string value or another name shares once.
The constructor compiles its source as (function anonymous(...) {...}), and the
compiled function bound anonymous to itself as a named function expression does.
That binding shadowed a global variable of the same name. ES5.1 section 15.3.2.1
creates the function with the global environment as its scope and no name binding,
so the constructed function now leaves the name unbound.
Listing a String object's index names for for-in built one name per UTF-16 unit before any
limit check, so a string within the heap limit could make the loop allocate many times that
limit. The heap budget for all the names is now checked first, after a collection if needed,
and each name takes a unit of fuel. A wrapper made by ToObject stays rooted while the names
are listed.
Realm::run compiles the script with the realm's Options::strict instead of always as sloppy code, so a realm made with strict: true rejects with statements and calls functions with an undefined this. Indirect eval and Function keep their own strictness (ES5.1 10.4.2, 15.3.2.1).
Function.prototype.apply drops the temporary GC roots it holds for the argument values when an index getter throws, as it already did when reading succeeds. Before, a caught throw left them rooted until the run ended, so a loop of such calls could reach the heap limit.
Realm::run now resets the exception that bind_last_thrown binds, so a run
that completes, reaches a limit or does not parse leaves it undefined. The
script source, the text built by Function and the strings given to eval are
checked against the run's remaining max_source_units before they are encoded
or flattened, so an oversized source is rejected without being copied.
vectorcraft-js gains a Realm that runs ECMAScript 5.1 scripts. The parsed
script compiles to stack-machine bytecode, a verifier checks every jump
target, handler, constant index and stack depth before the code runs, and
the interpreter keeps JavaScript calls on its own frame stack. Closures,
this, arguments, with, direct and indirect eval, new Function,
try/catch/finally, labeled break and continue, and for-in in insertion
order follow sections 10 to 14 of ECMA-262 5.1. Variable references are
resolved before the right-hand side of an assignment runs (11.13).
Objects and environments live in an arena with a mark-and-sweep
collector and index handles. Built-ins are limited to Object and
Object.prototype's basic methods, Function with call and apply, eval
and the error constructors.
Every run is bounded by fuel, heap bytes, frames, nested host calls and
an interrupt flag another thread can set, with smaller defaults in the
wasm build. A run that reaches a limit ends with a limit error that a
script's try cannot catch. Source passed to eval and Function counts
against Options::max_source_units.
cargo xtask test262 also runs the suite's ES5 test/language tests that
need only these built-ins and prints their counts apart from the parse
counts. Nothing in the app calls the interpreter yet.
Current test262 (ECMA-262 2017 Annex B.3.6) accepts `for (var x = 1 in o)` only outside strict
mode code. The parser now rejects the initializer in strict code, with the error at `in`.
New crate vectorcraft-js reads ECMAScript 5.1 source into a syntax tree
with source positions: a UTF-16 lexer (line and column tracking, automatic
semicolon insertion, the regexp and division goals), preprocessor lines
read as directive tokens, a recursive-descent parser to an arena tree, and
the ES5 early errors, strict mode included. Nesting deeper than
Options::max_depth (200 by default) gives a SyntaxError with its line and
column rather than overflowing the stack. Nothing runs scripts yet.
cargo xtask test262 downloads test262 at a pinned commit, checks its
SHA-256, unpacks it into the gitignored corpus/ folder and runs the ES5
syntax tests. A differential test in the engine compares the lexer with
scriptcheck on line counts, directives and unterminated literals.
The legacy reader skipped `TV`, `Tv`, `Ty` and `TY` of section 15.3.4 of
the version 7 file format specification, so type set with `1 TV` opened
horizontal.
- `TV` sets a text object's writing direction. VectorCraft keeps one
direction for each text object, the one in force at its first
character, with a warning when a later character has another.
- `Ty` marks a Kumi block as a Tate-chu-yoko run, which is VectorCraft's
own mapping, and `TY` does the same in horizontal type. A `TY` block in
vertical type opens unmarked. Two blocks with nothing between them open
as one block.
- `Tv` sets Standard Vertical Roman Alignment.
All four are modal, and an operand other than 0 or 1 leaves the value as
it was.
Docs: docs/file-format-parity.md.
After Select All in threaded text, the Justification dialog kept the edited
frame's selected range when that frame was no longer threaded at OK. A thread
broken after the frame moves the rest of the story into it, so the paragraphs
past the old range kept their values. The dialog now drops the range and
changes every paragraph in that frame.
With mixed indents, the Tabs panel sends one text.setFormat per paragraph,
addressed by frame and byte offset. Threaded text reflows after each step,
so an indent that made earlier paragraphs wrap onto more lines moved text
into the next frame, and the later steps landed on the wrong paragraphs or
on none. The steps now run from the story's last paragraph to its first, so
each step's offset comes before every change already made.
After Select All in threaded text, the Tabs panel's indent markers read
the indents of the edited frame alone. When another frame's paragraphs
had a different left indent, a marker drag changed them by the edited
frame's values, or left them alone. The markers now take every
paragraph of every frame in the story, so each paragraph keeps its own
left indent or first-line position.
The Justification dialog kept the frames of the story from the moment
it opened. OK now acts on the frames in the edited frame's thread when
it is pressed, so a frame that left the thread while the dialog was
open is not changed, and a frame that joined it is.
After Select All in threaded text, the panels read the frames from the
list saved at Select All. A frame that had left the thread still took
Paragraph, Tabs, Justification and Text Styles changes, a frame added
later did not, and when the first saved frame had left, Character,
OpenType and font changes went to that frame alone.
text_story now reads the frames of the edited frame's thread from the
document. Locked and hidden frames stay in the story.
An object read as its drawn look counted in the import's "came in as
their drawn look" note when it was on the page or in a kept symbol, but
not when it was in a pattern's art or a brush's art. A file whose only
such objects were in a pattern or a brush opened without the note.
- A kept pattern's drawn looks count, once for each pattern name: a
definition that a later one of the same name replaces no longer counts.
- A brush's art counts once a brush that is kept uses it; art no kept
brush uses adds nothing.
Q restores the painting state but not the transparency (Xy): the writer
sets Xy only when the next object's state differs from the last one it
wrote. Restored at Q, type after a clipping group following a Screen glow
drew in Screen and nearly vanished on a light background.
A blend whose spine was never edited has only its keys in the editing
data, no spine part. Its spine is the straight lines between the keys'
centres in the record's order; read as such, the blend comes in live
instead of as its drawn steps.
Illustrator 24 writes its symbol definitions ahead of %AI9_BeginArtStyles,
so a styled object inside a symbol found no style and came in as its drawn
look. The pre-pass that finds swatch names now also keeps the first
top-level art styles section, which the reader parses before the art.
set_refuses_a_recovery_or_plugins_folder_it_cannot_use takes write or
list permission away from two folders. A guard now gives each folder mode
0o755 when it is dropped, also when an assertion panics, so the test
folder is removed afterwards.
prefs.set did not check the folder given as recoveryFolder or pluginsFolder,
beyond the automation roots check, so a control or MCP caller could save a
folder that does not exist, a file, or a folder the app cannot use. Only the
Preferences dialog checked the folder itself.
The dialog's folder checks move to the engine, and prefs.set runs them after the
automation roots check. A folder that fails returns the same error the dialog
shows, and no preference changes. An empty value, an omitted value and a folder
the preference already holds are not checked. The Wasm build skips the check, as
the dialog did. The dialog's OK now gets the check from prefs.set, and Choose…
calls the moved check directly.
Enter and Escape counted their key repeats as presses. Holding Enter with
a selection and a selection tool opened the Move dialog, confirmed it,
and opened it again at the key's repeat rate. After a drag, which gives
Move an offset, each round moved the object again.
The app now records what Enter or Escape went down on: the open dialog,
the command search, Edit Toolbar's drawer or Presentation Mode. When that
changes, the key's presses are removed from the input until it is let go.
The tool keys, the dialog's OK and the dialog's Escape also ignore
repeats, so holding Escape with the place cursor loaded discards one file.
A new press still acts, and Enter held in the Type tool still repeats line
breaks.
This also fixes these related faults:
- Ctrl+Enter or Cmd+Enter opened Move and confirmed it in the same frame.
- Holding the Enter that confirmed a dialog typed line breaks into the
text being edited.
- A command bound to Enter reopened its dialog after the dialog's OK.
- Holding Enter on a value a field rejected confirmed the dialog.
- A command search that opened while Enter was down ran its first result.
- Holding Escape over a dialog's Tips closed the dialog as well.
- Holding Escape over a dialog went on to close Edit Toolbar's drawer,
leave isolation mode or run a command bound to Escape.
The check tested one box per source segment for every candidate hole or
island. It now searches a tree of the segments' bounding boxes, built once
when the first contour is checked. The answer for each contour is the same,
so the output paths do not change. MAX_OFFSET_SEGMENT_PAIRS does not change.
With the Artboard tool, the canvas context menu lists the entries of the
public Use the artboard context menu page in its order, in place of the
entries for objects and the view. Lock Artboards and Unlock Artboards are
two submenus. An entry that cannot run is left out, and Undo and Redo stay
dimmed. A right-click on an artboard that is not selected selects it alone
first, and the menu measures the art for Delete Empty Artboards and Fit to
Artwork Bounds once, at the right-click.
Cut, Copy, Duplicate Artboards, Rename, Delete Artboards, Lock Selected,
Unlock Selected, Fit to Artwork Bounds, Switch Orientation and Export act on
the artboards the tool has selected, else the active one. docs/mcp.md lists
what the other entries act on.
- artboard.rename renames in one undo step and numbers several names. The
menu's Rename opens a dialog that runs it, and the Artboards panel's name
field runs it too.
- artboard.switchOrientation swaps width and height and keeps the top-left
corner, skipping square and locked artboards.
- artboard.selectAll chooses the tool and selects every artboard.
- edit.paste takes artboardsAt, and the menu's Paste centers copied
artboards on the right-click's point.
- The menu's Cut and Copy run artboard.cut and artboard.copy. In the desktop
app those commands publish to the system clipboard, and the menu shows the
Cut and Copy shortcuts for them.
- artboard.fitToArt and the panel's name field record no undo step when
nothing changes.
artboard.duplicate, copy, cut, rename and switchOrientation take their
artboards from the tool's selection when a call names none. Each now records
them in the journal as selectedArtboards, which a replay reads and a recorded
action leaves out. The copy commands also record art. switchOrientation
records scaleArt, and it takes patterns and lockedAndHidden and records what
its artboard.setProps step resolved for them.
While the Artboard tool is in use, a command chosen from any menu or the
command palette moves the tool's selection as the Artboards panel does:
artboard.new makes the new artboard active, artboard.duplicate selects the
copies, and artboard.delete and artboard.deleteEmpty make a remaining
artboard active.
The fitter accepted a piece at depth 14, or with fewer than six points,
whatever its error, so a traced outline could stay farther than the
tolerance from its points. Synthetic 2 px block noise at Paths 100 and
Corners 0 kept a piece 2.7 px off at a tolerance of 0.1.
Such a piece is now accepted only when each point is within the tolerance
of the curve. Otherwise a short run is cut at its worst point, down to
two points, and a run at the depth limit is cut in the middle, which keeps
the recursion shallow. Every other piece is unchanged. Noisy art can get
more anchors.
a_traced_pixel_that_straightens_to_no_area_... now traces at Paths 80. At
Paths 90 that pixel's no-area result came from pieces over the tolerance.
An image over the pixel cap was left out without reading its data, so
the program went on to read that data as PostScript and the file failed
to open (Ghostscript's eps2write writes a page with a large inline image
this way). The data is now skipped: the filters that find their own end
(ASCIIHexDecode, ASCII85Decode, SubFileDecode) read the file up to it,
without decoding what they pass.
cvi and cvr read a string as the PostScript syntax reads a number
(PLRM 3rd ed., cvi, cvr), so (16#20) cvi is 32. The reader parsed strings
as decimal only. Ghostscript's eps2write prolog undoes a name's #20 escape
that way, so its files with a spot colour named with spaces stopped
reading at that colour.
The count of a SubFileDecode filter bounds what it passes through, not
what it allocates, but the reader refused counts over 4 MiB as a
rangecheck. Ghostscript's eps2write files read each content stream with
currentfile Length () /SubFileDecode filter, so a page whose stream is
longer than 4 MiB failed to open.
bind makes each procedure it goes into read-only and ignores read-only
procedures (PLRM 3rd ed., bind). The reader bound them again, so a prolog
that binds a procedure holding an earlier-bound one, after defining a name
such as M as an operator, turned that procedure's own variable M into
setmiterlimit. Ghostscript's eps2write prolog does this, and its files with
sampled functions (transfer, black generation) failed to open.
document.save's params text, which file.saveAs, file.saveCopy and
file.saveAsTemplate share, now says that the warnings also name each key
given in `options` that the format doesn't read.
save_file's note for a key in `options` that the written format doesn't
read named every .ai file a PDF-compatible .ai, also when the save wrote
a version 7 file (`aiVersion: "7"`) or one without PDF content
(`pdfCompatible: false`).
The save itself now writes that note, beside its note on a top-level
`compress`, `version`, `preview` or `precision`, and names the file it
writes, including when it reuses the options the document was last saved
with. save_file returns the save's reply unchanged, and
`run_command document.save` and the other save commands give the same
notes.
The reader put an opacity mask's art in a group every time, so a mask
made from one path came back from a saved .ai as a group holding the
path, while Make Opacity Mask keeps one object as the art. This also
changes how existing files with single-object masks open, because
the editing data of a .ai, an EPS and a PDF goes through this reader:
the art is the object.
A hidden object or a guide stays in a group, because the canvas draws
the art's root whether it is shown or not and whether guides show. A
clipping path alone stays a clipping group.
The editing copy writes a plain group of one object as a group (u, the
object, U), so it reads back as a group. Its other plain groups are
still written as their members. Hidden mask art is written shown,
because the canvas draws it and a group read back would not.
The panel menu's item added the artboard and left the active artboard
where it was, while the panel's button, the Control bar and Properties
made the new one active. All four now call one helper. With the Artboard
tool, the new artboard is then the only one selected.
When the link-watch worker unwound before storing its stamps,
poll_link_scan returned None on every frame and left link_scan set. The
app then asked for a repaint every 100 ms and start_link_scan returned
early, so the app did not look for changed linked files in the
background again for the rest of the session.
poll_link_scan now clears a look that has ended without stamps. The new
ended_without_stamps helper holds the slot's lock across its read of the
stamps and its read of the ended flag, so a look whose worker stores its
stamps during the check is applied. The app goes back to its two-second
wake, and the next look starts on schedule. A look still running and a
look that stored its stamps behave as before.
Clipping paths in a compound path that holds a drawn look become one
compound path. It took the paths' shared stroke, opacity and blend mode
but dropped the knockout shape flag. It now has the flag when every
clipping path has it.
An `Xy` whose shape operand is set now reaches an image (embedded or
linked) and a gradient mesh when it is written before the object, and a
mesh or blend when it is written after the object's end. A live blend
takes the flag from its drawn steps. Before, the reader dropped the flag
on these objects, so an image saved with Opacity & Mask Define Knockout
Shape lost it on reopening, with no warning.
A styled object takes the flag written on its drawn look. For a text
object, the flag goes to the object when the look draws only its type,
and otherwise to the group that holds the look and the type. An opaque
look that draws more than the type and sets the flag now goes into that
group as well.
Area type whose open frame ends where it starts was written as text and
read back in a closed frame. It is now written as its outlines, alone or
as a frame of a thread.
The reader gives an Xy right after a compound path or a text object to
that object. The writer now writes XR before the Xy of type, images and
guides, so type next to type, a compound path before type, image or
guide, and type before an image or guide keep their own opacity, blend
mode and knockout shape flag.
- Type keeps its Isolate Blending and Knockout Group state in a state
block after its text object, as its outlines did on their group.
- The reader no longer gives a layer the Xy before its last text object.
- The ASCII85 end marker of the text document stays on one line.
- The warning and docs/file-format-parity.md also list area type in an
open frame, type on an open path whose ends meet, type of only white
space and threads of type on a path as written as outlines.
- README.md, ROADMAP.md, docs/gaps.md, docs/roadmap.md and
docs/target-app-parity.md describe the editing copy a saved .ai holds.
- clip_path's doc comment is back on clip_path.
- import_fuzz reaches threaded area type and vertical type.
The editing copy of a PDF-compatible .ai wrote every text object as its
outlines. Point type, area type and type on a path are now written as
text: a text object in the layers that names its story, and the text
document after the document data (%AI11_BeginTextDocument) holding the
stories.
- Each story holds its characters (a paragraph ends in \r, a forced line
break is \x03), its character runs (font, size, leading, tracking,
baseline shift, horizontal and vertical scale, solid fill and stroke
with its weight, capitals, Tate-chu-yoko) and its paragraph runs
(alignment, indents, space before and after, hyphenation, right-to-left
direction).
- Point type keeps its point and, when it is turned, its matrix. Type
scaled by one factor, such as all type on a large canvas, has its scale
in its sizes, frame and path. Area type keeps its frame, rows,
columns, gutters and first baseline, type on a path its path and where
it starts and ends, and vertical type its orientation. A threaded story
is one story through its frames in order. Area type and type on a path
carry their frame's path in the layers as well.
- Type the text document can't hold is written as its outlines, with a
warning: type in opacity masks, pattern tiles and art cut to a clipping
object, mirrored type, turned or unevenly scaled area type and type on
a path, frames of several parts, area type that flows around a text wrap,
threads whose frames differ in orientation or include a frame written
as outlines, type and stories past the sizes the reader reads, and type
with inline graphics, ruby, gradient or pattern paint or an appearance
or opacity mask of its own. Other character and paragraph attributes are left out, with
a warning.
Tracking is written as a whole number of thousandths of an em, and
solid paints as simple paints; tracking that isn't a whole number is
rounded, with the warning.
Each construct is written in the form VectorCraft's .ai reader reads
(crates/eps/src/import/ate.rs at 36ef34dfdd: Texts::read :407, story
:501, frame :456, para :562, style :588, paint :622, AreaFrame::read
:284; the text object in import/ai/read/art.rs:15 and its path in
import/ai/frames.rs:26), and the tests read the written data back
through that reader. The scripting
reference documents the same attributes on the reading side (text frame
kinds, columnCount, columnGutter, rowCount, startTValue, endTValue,
orientation, and the character attributes):
https://web.archive.org/web/20160410101607/https://wwwimages.adobe.com/content/dam/Adobe/en/devnet/illustrator/sdk/CC2014/Illustrator%20Scripting%20Reference%20-%20JavaScript.pdf
The reader now closes a text object's frame path painted with a closing
operator (n, f, s, b, h), so area type's frame comes back closed; places
point type whose story keeps no composed lines at its frame's point;
keeps a story's last empty paragraph (only the \r that ends the story is
dropped); gives type made from the text document its text object's lock,
as it gives it its visibility; and gives type its text object's knockout
shape flag, as it gives it its opacity and blend mode.
Tests: the round trip of every kind of type and its state, a second
write of what is read back giving the same bytes, the text document and
stubs, scaled point type, a story split between threaded frames, the
fallbacks and the warnings, and the engine's editing copy with every save
option; a fuzz case for damaged text documents and stubs.
Docs: docs/file-format-parity.md.
a_pan_moves_the_marks_without_building_them covered Direct Selection
only. It now runs with both tools.
The module comment and MAX_PAN's comment now say what is kept as
triangles and as document points, and when a pan builds the marks again.
A test label now says points rather than px.
The timings in docs/target-app-parity.md are measured again with 10 fill
colors on the 2,000 circles. The pixel sentence there also names the
rounding errors of kept vertices.
Kept squares and handles hold document points, and each frame maps them from the view's center as Xf::to_screen does. A pan's distance comes from the difference of the two view centers.
A frame that repeats a view now draws the marks as one mesh. frame_output drops the kept marks first, so the tests that read outlines and anchors from the frame's shapes keep working.
dotCooCoo