Known limitations

In the lesson check, DOM.getOuterHTML serialized angle brackets in text without escaping them. Its output alone cannot determine whether input became HTML elements.

A positioned inline used as an absolute-positioning containing block omits its left margin horizontally and uses content-area top plus top border vertically, offsetting elements by 2-5px

The last word wraps to the next line when an absolute/fixed shrink-to-fit container contains atomic inline boxes

A fixed-width display:table-cell in an anonymous table incorrectly uses width as a lower bound on its min-content contribution

APNG displays only its first frame without animation because it is decoded as a still image

A line containing only one atomic inline box and no text omits the root strut descent, making the line box too short

Pure-percentage descendants of a block containing box with definite zero height are treated as auto content height

In a 122.3125px-wide block, KUI wraps `the lazy dog and keeps running on` into two lines while Chrome uses three

BrowserCompat intrinsic sizes for date/time/month/week/file/submit/reset controls differ from Chrome; month/week are laid out as text fields at width 177 versus Chrome 112/147

A column flex item resolves min-width:fit-content against its own width rather than the container cross size

Elements replaced by content:url() bypass replaced-element sizing, so min/max, aspect-ratio, and width transfer from a definite height do not apply; SVG ratios are also fabricated from raster dimensions

getComputedStyle cannot read animation longhands such as `animation-name`, `-duration` and `-iteration-count`, which return empty strings

getComputedStyle() returns empty strings for box-shadow, border-radius, and individual translate, scale, and rotate properties

Wrapping and keyword serialization of mathematical functions in getComputedStyle

During `padding` transitions, side values differ by around 1e-4 and `getComputedStyle('padding')` serializes four values

The public C++ CSSStyleRule::setSelectorText() method changes selector text without reparsing the selector list, so matching does not change

Empty first/last fragments of block-split inlines resolve percentage padding and margins against the outermost viewport width instead of the containing block content width

`<fieldset disabled>` does not disable descendant form controls: `:disabled` does not match, controls remain focusable and existing focus is not moved

Fieldsets with a rendered legend resolve percentage-height children without subtracting legend intrusion and misplace items in `display:flex` fieldsets

An item with min-height:100%;aspect-ratio:1 in a row flex container has zero width because the ratio is not transferred from min-height

Fractional imposed sizes and height measurement for atomic flex boxes still encounter legacy sentinel and line-height boundaries

Flex/grid still treat raster-cache dimensions as natural width/ratio for SVGs with one or no natural axes and an authored aspect-ratio

An auto-width flex item containing an inline-end float is measured with a flex base of about 100,000

Top/bottom percentage margins on children of flex items are calculated on only one side when passing through a chain of BFC wrapper blocks (the ignored margin-left:10% on an inline-flex item in a column flex container has been fixed)

An inline <svg> with only viewBox and no dimensions collapses to 0 by 0 in a flex row, while Chrome fills the remaining width

The main size of row flex items with writing-mode:vertical-* differs from Chrome

When a whole line is pushed below a float because it does not fit beside it, the atomic box ink top rather than the line-box top is aligned to the float bottom

GPU SkiaBackend::readPixels() and saveScreenshot() do not always read the current window's most recently presented frame

Explicit align-items:stretch on a grid container does not stretch replaced elements such as img, canvas, or video or transfer the ratio to their width

A terminal hard line break in anonymous pre-line grid text adds an extra line of height

When a non-auto min-width is smaller than the content, normal Grid layout still uses min-content width for the column minimum contribution instead of min-width

A child block with a percentage-containing calc height inside an auto-row grid item ends with height 0 instead of Chrome's 10

An auto grid row becomes too tall when an item with height:N% contains two or more atomic boxes

A grid item that is a fieldset with a rendered legend omits the legend intrusion from its auto height and makes the row too short

Intrinsic width queries do not cap track contributions at the fixed fit-content limit

A fit-content container with two 0.5fr columns underestimates its natural row height by one line

Remaining issues in the intrinsic-size input contract for grids with fr tracks and its consumption by Flex, Table, and inline entry points

An auto-height grid with two or more fr rows and a definite min-height does not recompute fractions from min-height, placing items from the second row upward

Grid intrinsic queries omit implicit column contributions when explicit and implicit columns are mixed

Stretched img/canvas elements with percentage heights in auto rows make the row too short for pure percentages and about 500,000 pixels tall for calc(percentage + length)

After a block splits an inline box, getClientRects() omits the full-line rectangle wrapping the block and zero-width rectangles for empty fragments with no horizontal edges

Whole-paragraph layout drops the letter-spacing contribution on spaces in Arabic paragraphs

A failed IFC result cached without a backend suppresses a valid paragraph callback installed later

When a float bottom falls inside the first line of a paragraph, whole-paragraph layout does not exclude the float and paints text into its area

A scroll container's natural-content-height floor is applied only when an IFC result cache exists; inline padding/borders make scrollHeight about 1px too small and incremental layout differ from full rebuilding

An inline-block containing only Arabic text with letter-spacing shrinks to min-content width (the longest word), while text remains unwrapped and paints outside the box

The natural width of an inline-block wrapping a Grid still uses generic ceil rounding and is remeasured from geometry after final layout

An inline element larger than its container font, such as a font-size:40px span on later lines, has getClientRects()/background about 8.7px higher than Chrome and a container-height difference of about 0.7px

With a <sup> or vertical-align-shifted child, an inline element background, getClientRects(), and hit area sit about 9px above its own glyphs

When an inline contains an atomic inline box and soft-wraps after it, the line fragment incorrectly includes the trailing-space width

For `display:inline list-item`, no break occurs between the inside marker and first word; intrinsic width also treats marker plus word as one unit

When line height is 0.01-0.024px shorter than inline span content, KUI treats it as negative leading and places the span top 1px above Chrome

When an inline span has a larger line-height than its parent line, the line baseline and glyphs sit about 10px lower than Chrome

Inside markers with a larger font than the body do not increase first-line height; their ink extends above the list item

IntrinsicSizer ignores child min-width values written with calc(), treating them as zero during shrink-to-fit sizing

Intrinsic width counts both spaces when text ending in a space inside an inline element is followed by text beginning with a space

Intrinsic width definitions are inconsistent for an empty table

Intrinsic sizing takes the maximum of a float and an adjacent BFC root instead of adding them, making shrink-to-fit containers too narrow and pushing the BFC below the float

IntrinsicSizer does not clamp a box's own block size by min/max-height when measuring intrinsic width, so ratio transfer for percentage-height descendants uses an unclamped height

With nonzero letter-spacing/word-spacing, intrinsic text width across element boundaries is underestimated and the last word wraps in shrink-to-fit containers

The min-content width of a table with table-layout:fixed and width:auto differs from Chrome

Direct intrinsic queries on tr or row groups take the largest child box as for a block container rather than summing column widths

A `<span> </span>` containing one space between two inline-blocks is measured as zero intrinsic width

Keyframe easing is applied at the wrong stage: alternate/reverse with asymmetric easing and multiple keyframes with non-linear easing produce values different from Chrome

After deleting an @keyframes rule or removing its stylesheet, the animation manager retains a raw pointer to the freed rule and the animation continues

var() in @keyframes declarations does not substitute custom properties: numeric properties become zero and other properties have no effect

With retained surface and layer promotion enabled, promoted elements inside scrolling containers ignore the container clip and paint outside it from the first frame

With retained surface and layer promotion enabled (off by default), paint-only changes do not repaint promoted layers and unchanged hashes skip presentation, leaving stale content; presented frames with the features disabled also differ from full-page rasterization

When the root has display:inline and position:sticky and body has position:fixed, all boxes below body have 0 x 0 geometry while text is still painted

When legacy -webkit-line-clamp truncates, the forced ellipsis on the last visible block replaces the whole line with an ellipsis; direct text preceded by <br> before the block gets no ellipsis

The legacy evaluator treats children with width:0 or calc() widths as auto and measures their content, widening floats and width:fit-content/max-content containers

In segmented layout beside floats, letter-spacing differs substantially from Chrome and boxes are displaced from glyphs by half a spacing increment

A long word that does not fit breaks in the middle with centered -webkit-line-clamp

A line-clamp paragraph with both a float and inline elements is not clamped, and its first-line text paints beneath the float

The hard-line ellipsis path for line-clamp assumes each hard line occupies one visual line, placing the ellipsis on the wrong line or omitting it when earlier soft wraps exist

In the private hard-line painting paragraph for legacy line-clamp, restoring the final cluster under negative letter-spacing offsets centered/right-aligned lines by half or one spacing cell

A text-align:justify line-clamp paragraph with multiple hard lines has no ellipsis

In a soft-wrapped text-align:justify line-clamp paragraph, the final line appends an ellipsis without truncating its prefix, leaving half the ellipsis clipped by the container

The last line of legacy -webkit-line-clamp includes the start of the next word before truncating it

When the line count stays within line-clamp but a long URL or word does not fit on the last line, KUI draws a native ellipsis while Chrome clips without one

With line-clamp and text-overflow:ellipsis, horizontally overflowing nonfinal lines do not receive per-line ellipses

On Windows, once a local bitmap enters the image cache, overwriting the same file in place fails (for example, saving a new avatar or edit result back to the file currently displayed)

Lua Canvas mistakes 100-900px font sizes without a font weight for numeric weights and throws an exception

Lua `MutationObserver:observe(element, ...)` rejects elements with a type mismatch

`::marker{content:none}` still paints and reserves space for the default marker

measureSubstring crashes the process (0xC0000005) when byteEnd falls inside a multibyte UTF-8 character

The scalar rectangle of a multiline inline, combining first-fragment x with bounding width, is still treated as real occupied geometry by scrollIntoView, outline unions, and sticky positioning

Lines rebuilt through the CSS line-box path, including negative half-leading or fallback-font cases, still lose the root strut descent when a baseline-aligned atomic box lowers the baseline

Without a backend, nested percentage-sized atomic inline boxes repeatedly accumulate box edges on warm frames

Auto static positions for fixed elements in flex and absolute/fixed elements in grid differ from Chrome

An outside list bullet beside a float differs from Chrome when the list item has padding/margin or establishes a BFC

Whole-word rescue widens the center/right/RTL alignment area when only some lines in a multiline paragraph overflow

After an atomic box inside an inline element changes to percentage width, selective descent resolves it against the inline parent box width from the previous frame

After text grows under positive line-clamp, a reused zero-width trailing text box retains its old baseline

Style-only animation frames do not refresh ::before/::after styles; pseudo-elements retain old values while host color and other properties transition or animate

Ratio-only images without natural dimensions, including viewBox-only SVGs or dimensionless SVGs with an authored aspect-ratio, use raster sizes in shrink-to-fit, flex, and absolute positioning

create() for img, audio, video, lottie-player, and DocumentFragment retains one extra reference, so these nodes are never destroyed

Percentage left/right on a relatively positioned inline nested in another inline resolve against the parent inline width instead of the containing block width of the nearest block container

With retained surface, partial damage, and layer promotion all enabled, changing dir and laying out again a whole line leaves a region of pixels from the previous frame

With retained surface enabled (off by default), partial replay of fractional-coordinate explicit damage along the hollow edge of a box shadow differs from full-page rasterization by 30 pixels

When a text run following an atomic inline box crosses a hard line break in an RTL paragraph, the atomic box shifts right and overlaps Hebrew text

Inline-box edge placeholders next to Latin runs in RTL paragraphs follow their own bidi level, separating from box content or landing on the wrong side

An atomic inline box wider than an RTL line sits at the left edge instead of overflowing from the end edge on the left

A separately laid-out inline root in segmented layout gets a paragraph box, including line-height and one solid multiline area, instead of an inline box

In segmented RTL paragraph layout, a separately laid-out inline root shifts its entire glyph run rightward outside its box and overlaps following text

With numeric line-height, a smaller-font inline element places the line text and boxes about 2px lower than Chrome

steps(n, jump-none) and steps(n, jump-both) are treated as jump-end and jump-start

A wrapped column flex item stretched in a fixed-height single-line row flex container does not use its stretched height when wrapping into columns

currentColor in border-*-color, text-decoration-color, column-rule-color, and row-rule-color is resolved during computed-value calculation, so a child using inherit receives the parent's color

Four loader threads create the same SVG renderer concurrently without a lock during first SVG decoding, causing a data race

An absolutely positioned child after cells in a display:table does not receive a static horizontal position after the cells

When all table rows have visibility:collapse, KUI also removes the leading and trailing vertical border-spacing, making the table shorter than Chrome

A text-only display:table with percentage padding subtracts padding again from the content width, squeezing text into multiple lines

Table geometry as a row flex item differs from Chrome: collapsed-border tables miss outer edges and separate-border tables gain extra border height

An auto-layout table column containing several inline-block tags is much wider in KUI than in Chrome

A bordered or padded table places its caption inside its own border and padding, drawing the table border above the caption

Columns do not stretch when a caption is wider than the table grid, leaving the grid against one side of the table box: left in LTR and right in RTL

In a positioned inline split by a block child inside a table cell, left:auto uses a static x after the split rather than before it

Percentage widths on col/colgroup use the whole table width without subtracting border-spacing, making columns wider than Chrome

A table with a collapsed column retains its fixed width and leaves blank space while Chrome shrinks the table

A visibility:collapse column loses its track but leaves its border-spacing, making the table one spacing interval too wide

With collapsed borders, equal-width, equal-style vertical borders of adjacent col/colgroup elements incorrectly choose the logically later border

KUI retains fractional border-spacing while Chrome truncates it to integer pixels, making multicolumn tables progressively wider

An inline-block child of a fixed-width display:table is stretched too wide, and table height differs from Chrome when a later static block enters flow

When a spanning cell is taller than the sum of its rows, excess-height distribution omits the crossed vertical border-spacing and makes the table too tall

Text decoration on text truncated by text-overflow continues beneath the ellipsis

Nowrap text-overflow has no ellipsis when the line contains an inside list marker or an out-of-flow child

With a word that does not fit, a soft-wrapped paragraph declaring text-overflow has one fewer line than the same paragraph without it

A translucent rounded box with overflow:hidden (such as a frosted-glass dialog) shows a faint light edge along children at fractional positions during entrance scaling, translation, or centering

Vertical text (writing-mode:vertical-rl and similar modes) draws text-shadow horizontally without rotating it with the text

`visibility:collapse` on non-table elements prunes the entire subtree: explicitly visible descendants are neither painted nor hit-tested, while Chrome treats it as hidden

-webkit-line-clamp on -webkit-box does not count lines across block children

The previous-sibling fallback for an inline containing block still gives an inaccurate origin when per-line fragments are disabled and the previous text box spans multiple lines

Remove hardcoded Windows paths from the advanced-example toolchain

Floating or absolutely positioned `display:inline-block` elements are not blockified; computed display remains inline-block

Temporary BrowserCompat compensation for form-control baselines and listboxes

Review shared ink-bound pre-culling behavior for degenerate borders, thick wavy lines, and transformed glyphs

Clipped replay cannot construct raster windows for lists containing clip-path, masks, backdrop-filter, 3D matrices, or non-blur filters; animation-only tiles intersecting curved primitives still differ by several levels from full-image rendering

Consolidate cross-platform generic font mappings

Implement the CSS Counter Styles capabilities actually used by WPT in stages

border-width does not fall back according to property syntax when var substitution introduces an expression containing percentages

End-to-end CSS zoom semantics for computed style, layout space, paint coordinates, scroll overflow, and hit testing

getComputedStyle for border-image-*: nonzero outset values and gradient/filter() sources still return empty strings

Some categories of element used values and computed forms remain unfinished in the CSSOM reading layer

P2: runtime fold boxes lack independent CSS used-value data, so reads temporarily fall back to computed values

Two approximations in the decorating-box baseline

The gaps produced by `text-decoration-skip-ink:auto` differ from Chrome

Wavy-line shapes and dashed-line endpoint antialiasing differ from Chrome

IB splitting, intrinsic width, and wrapping for inline details/summary and anonymous ::details-content block/contents slots

Details/summary markers, anonymous boxes, inline display, and inherited list/decoration styles