[b84330808]

Per the SVG spec, a T/t that does not follow a Q/q/T/t uses the current point as its control point, and the next T reflects that control point. Both code paths in @remotion/paths got the second T wrong:

• construct() (used by getLength(), getPointAtLength(), getTangentAtLength(), getInstructionIndexAtLength(), evolvePath()) reflected a stale control point instead: the one from an earlier Q, or [0, 0]. Because that point is absolute, the result also changed when the path was translated.
• iterateOverSegments() (used by reduceInstructions(), and through it getBoundingBox(), cutPath(), reversePath(), scalePath(), warpPath(), interpolatePath()) left the tracked control point null after H/L/V/A/Z/M, which was then read as 0. After C/S, it reflected the cubic's second control point.

Path	Chromium getTotalLength()	getLength() before	after
M 0 0 H 10 T 20 10 T 30 0	45.925	44.206	45.925
same after translatePath(path, 100, 100)	45.925	160.758	45.925
M 0 0 C 0 10 10 10 10 0 T 20 10 T 30 0	55.925	54.206	55.925
M 0 0 Q 5 5 10 0 L 20 0 T 30 10 T 40 0	57.403	60.765	57.403

I also compared 2000 random paths mixing every command against Chromium's getTotalLength(). Before: getLength() disagreed on 300 and getLength(serializeInstructions(reduceInstructions(...))) on 296. After: 0 and 0.

The added tests compare each T chain with the equivalent explicit Q path.

I used Claude Code to write this change.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
bugs	Ready	Preview	Oct 30, 2026 9:04am UTC
remotion	Ready	Preview	Oct 30, 2026 9:04am UTC

[pullfrog]

✅ No new issues found.

Reviewed changes

• construct.ts — after processing a T/t, prev_point is now set to the current point when the predecessor is not Q/q/T/t, instead of reflecting a stale control point. This matches the spec and makes chained T lengths translation-invariant.
• iterate.ts — T segment state now only reflects the tracked control point when the previous segment is actually Q or T; otherwise it sets the control point to the current point. Previously a T following C/S reflected cp2x/cpx, corrupting the next T.
• get-length.test.ts — new test asserting a T chain after H equals the explicit Q equivalent, including a translated copy.
• reduce-instructions.test.ts — new test asserting T chains after H and after C reduce to the same instructions as the explicit Q equivalents.

I verified the reflection logic in both code paths against the SVG spec by tracing H 10 T 20 10 T 30 0 and C ... T 20 10 T 30 0, and ran the full @remotion/paths suite: 104 pass, 0 fail.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash ｜ 𝕏}

[JonnyBurger]

Thanks!