….40.0-787-g0f94c19331)

… freetype compatibility."

This reverts commit bfccc58.

….40.0-795-gcbc694239b)

….40.0-1240-gb57e04d1c7)

…, and removed.

It was deprecated in m139, and gone in m140.

Milestone 139
-------------
  * `SkFontMgr_New_FontConfig` with 1 parameter has been deprecated and will be removed in a future
    release. Clients will need to call the other version providing an SkFontScanner (e.g.
    `SkFontScanner_Make_FreeType()`)

Conflicts:
	src/skia/Font.cpp

…favor of the `SkRecorder*` version.

Milestone 140
-------------
  * `SkImage::isValid(GrRecordingContext*)` has been deprecated in favor of the `SkRecorder*` version.
    To migrate do something like `image->isValid(ctx->asRecorder())`.

    `SkImage::makeSubset(GrDirectContext*, ...)` has been deprecated in favor of the `SkRecorder*`
    version. To migrate, do something like `image->makeSubset(ctx->asRecorder, ..., {})`

    `SkImage::makeColorSpace(GrDirectContext*, ...)` has been deprecated in favor of the `SkRecorder*`
    version. To migrate, do something like `image->makeColorSpace(ctx->asRecorder, ..., {})`

    `SkImage::makeColorTypeAndColorSpace(GrDirectContext*, ...)` has been deprecated in favor of the
    `SkRecorder*` version. To migrate, do something like
    `image->makeColorTypeAndColorSpace(ctx->asRecorder, ..., {})`

    In the case you are working with CPU-backed images, `skcpu::Recorder::TODO()` should work until
    a `skcpu::Context` and `skcpu::Recorder` can be used properly.

…atrix.inverse()

m139:
    bool setPolyToPoly(const SkPoint src[], const SkPoint dst[], int count);
m140:
    bool setPolyToPoly(SkSpan<const SkPoint> src, SkSpan<const SkPoint> dst);

Before m140, only (deprecated):
    [[nodiscard]] bool invert(SkMatrix* inverse) const;
Added in m140:
    std::optional<SkMatrix> invert() const;

Old:
    bool getSegment(SkScalar startD, SkScalar stopD, SkPath* dst, bool startWithMoveTo);
Newly added:
    bool getSegment(SkScalar startD, SkScalar stopD, SkPathBuilder* dst, bool startWithMoveTo);

Old:
    bool transform(const SkMatrix& matrix, SkRRect* dst) const;
Newly added:
    std::optional<SkRRect> transform(const SkMatrix& matrix) const;

Old:
    bool getBoundaryPath(SkPath* path) const;
Newly added:
    SkPath getBoundaryPath() const;

was:
    bool applyToPath(SkPath* dst, const SkPath& src) const;
m140:
    bool applyToPath(SkPathBuilder* dst, const SkPath& src) const;

This reverts commit cd8af0c.

Should not be needed in m140.

See pybind/pybind11#5694

…canvaskit/0.40.0-3020-g2ab8add5be)

SK_DISABLE_LEGACY_NONCONST_ENCODED_IMAGE_DATA is gone in m144:

--- skia-m144/include/core/SkImage.h	2025-12-08 00:15:16.000000000 +0000
+++ skia-m145/include/core/SkImage.h	2026-01-18 04:38:15.000000000 +0000
@@ -739,11 +739,7 @@

         example: https://fiddle.skia.org/c/@Image_refEncodedData
     */
-#if defined(SK_DISABLE_LEGACY_NONCONST_ENCODED_IMAGE_DATA)
     sk_sp<const SkData> refEncodedData() const;
-#else
-    sk_sp<SkData> refEncodedData() const;
-#endif

     struct RequiredProperties {
         bool fMipmapped = false;

…pendFloatArray

The prototype has changed:

-    void appendFloatArray(const char* owner,
-                          const char* name,
-                          const std::vector<float>& value);
+    void appendFloatArray(const char* owner, const char* name, SkSpan<const float> value);

commit 9cd8adae4d5253da322f04ceef1dafaa3e6847a8
Author: Ben Wagner <bungeman@google.com>
Date:   Wed Jan 7 15:23:18 2026 -0500

    [pdf] Value owning AttributeList::appendTextString

    The existing appendTextString is efficient in that it does not make a
    copy of the `value` string, but also add a version that takes an
    SkString so that ownership can be passed.

    Change appendFloatArray and appendNodeIdArray to take SkSpan instead of
    std::vector&.

    Document that for all `const char*` parameters the attribute will hold
    the pointer and not make a copy and for how long the pointer must remain
    valid. Document that all other information is copied into the attribute.

…Format and VkImageLayout

Signed-off-by: Hin-Tak Leung <htl10@users.sourceforge.net>

Signed-off-by: Hin-Tak Leung <htl10@users.sourceforge.net>

src/skia/Matrix.cpp:1562:13: error: return type 'Tuple<int, int>' must match previous return type
      'Tuple<float &, float &>' when lambda expression has unspecified explicit return type
 1562 |             return py::make_tuple(-1, -1);
      |             ^
1 error generated.

Signed-off-by: Hin-Tak Leung <htl10@users.sourceforge.net>

…0.0-3021-gfba326b882)

Revert "Make artefacts available even when pytest fails."

This reverts commit a057e3c.

The name of the patch was auto-generated from its content
(and the change has indeed been applied once, reverted, and relanded upstream)
from `git format-patch -1 ...`. IMHO it should be kept as is, and
the content should not be converted to forward-apply. I'll just remove
the "-Reland" part to make copilot happier.

This reverts commit 300f730.

Fixes skia-python#359

Most of them done programmatically with something like:
    perl -pi -e 's/\&pts\[0\]/pts.data()/g' skia/*

Copilot points out that the latter is safer. Many(all?) of them were
touched by the recent SkSpan changes, plus a few more.

…inimize-download.patch

The example code in the Canvas Creation section of the tutorial requires the `sys` module to run correctly, but the `import sys` statement was missing.

This change adds the import sys to resolve the error.