Measure and center the full bounds of multiline equations

[vxdguy]

Describe the bug

MathPainter generates an image that cuts off the right side of the equation.

To Reproduce

Compiled on Ubuntu 24.04.1
.NET 8.0.405

CSProj contents:

<Project Sdk="Microsoft.NET.Sdk">
  <PropertyGroup>
    <OutputType>Exe</OutputType>
    <TargetFramework>net8.0</TargetFramework>
    <ImplicitUsings>enable</ImplicitUsings>
    <Nullable>enable</Nullable>
    <SelfContained>true</SelfContained>
    <PublishSingleFile>true</PublishSingleFile>
    <PublishTrimmed>true</PublishTrimmed>
    <IncludeNativeLibrariesForSelfExtract>true</IncludeNativeLibrariesForSelfExtract>
    <DebugType>embedded</DebugType>
    <ReadyToRun>true</ReadyToRun>
  </PropertyGroup>
  <ItemGroup>
    <PackageReference Include="CSharpMath.SkiaSharp" Version="0.5.1" />
    <PackageReference Include="SkiaSharp" Version="3.116.1" />
    <PackageReference Include="SkiaSharp.NativeAssets.Linux.NoDependencies" Version="3.116.1" />
  </ItemGroup>
</Project>

main.cs contents:

using CSharpMath.SkiaSharp;
using SkiaSharp;


// Render LaTeX to PNG
var painter = new MathPainter
{
    LaTeX = @"\begin{aligned}
\nabla \cdot \vec{\bf E} & = \frac {\rho} {\varepsilon_0} \\
\nabla \cdot \vec{\bf B} & = 0 \\
\nabla \times \vec{\bf E} &= - \frac{\partial\vec{\bf B}}{\partial t} \\
\nabla \times \vec{\bf B} & = \mu_0\vec{\bf J} + \mu_0\varepsilon_0 \frac{\partial\vec{\bf E}}{\partial t}
\end{aligned}",
    AntiAlias = true,
    TextColor = SKColors.Green,
    FontSize = 48,
    PaintStyle = CSharpMath.Rendering.FrontEnd.PaintStyle.Fill,
    GlyphBoxColor = (SKColors.Red, SKColors.Azure),
};

using var png = painter.DrawAsStream(format: SKEncodedImageFormat.Png, quality: 100); 

if (png == null)
{
    throw new InvalidOperationException("Failed to create PNG stream.");
}

using var file = File.OpenWrite("equation.png");

png.CopyTo(file);

Compile, execute, and open the resulting equation.png and you can see the image of the equation.

The canvas is not wide enough for the generated latex image.

Expected behavior

Width of canvas should be as large as the latex image.

Environment (please complete the following information):

• Platform: CSharpMath.SkiaSharp
• Package Version or Commit: 0.5.11
• Device: Desktop PC, Ubuntu 24.04.1

Additional context

See attached image result.

Maintainer triage (2026-08-30)

Reproduced on current master (8bda039) with the aligned example above and with #172's eqnarray example. DrawAsStream creates its surface from Measure, but later rows can extend past that measured/right-aligned bound and are clipped. The eqnarray reproduction also positions the shorter first row and longer second row as though the complete table bounds were not centered as one unit.

This issue now consolidates #172 and covers multiline table environments in both MathPainter and TextPainter.

Acceptance criteria:

• Measure uses the union of every row/cell's horizontal bounds for aligned, split, array/matrix, and eqnarray-style displays.
• Bounds include italic overhang and any negative/positive cell offsets, not only glyph advances or the first row.
• DrawAsStream does not clip any row at either edge.
• TextPainter centers the complete multiline display as one unit at finite and infinite canvas widths.
• Add regression/golden-image tests using the reproductions from this issue and textPainter does not correctly center an equation which contains eqnarray #172.

[vxdguy]

found another one, same behavior (image cuts off equation). latex below shows 3 equations, one per line, aligned on the "=" symbol.

"\begin{split} a^2 + b^2 &= c^2 \ a^2 &= c^2 - b^2\ a &= \sqrt{c^2 - b^2}\ \end{split}"

Only the first equation is centered horizontally. The horizontal extents only adjusts for the first line. The 2nd and 3rd lines do not seem to expand the width of the image.

[Happypig375]

There are also a few cut-off images in CSharpMath.Rendering.Tests. I suspect it's extra width taken up by italics don't get accounted for.

[Happypig375]

Validated against current master (8bda039). The aligned reproduction measures a surface from the table display, but the final row reaches past the right edge and is visibly clipped. I also rendered #172's eqnarray reproduction through TextPainter: its longer second row is shifted/clipped in the same way. This confirms one multiline-bounds tracker is sufficient, so #172 is consolidated here.