Blender Git Loki

Fix line width broken for consecutive line draw calls

D9054 did multiple consecutive `immBegin()`/`immEnd()` draw calls to draw
multiple lines at varying thickness. This would only work for the first line,
then they'd all get a 1px thickness (at least on macOS).

Issue was that `wide_line_workaround_end()` called `immBindShader()` directly
to restore the old shader (which the workaround overrides). However this
doesn't set `imm->builtin_shader_bound` which has to be done for the workaround
to work on the next `immBegin()` call. Instead `immBindBuiltinProgram()` can be
called.

Differential Revision: https://developer.blender.org/D10520

Reviewed by: Cl�ment Foucault

Fix T85939: Eevee Specular BSDF shader compile error

Introduced by 7f7e6830991b.

Fix T85930: Custom Property Error: Wrong Subtype

Regression in 08dbc4f996e4e95f3ab64f7bb3e1193700c585f5

Unfortunately lambda functions don't work with postponed annotations.

Cleanup: spelling

Docs: clarify radial/cycle use for BMesh loops

Also add minor corrections clarifications elsewhere.

Geometry Nodes: Move node UI storage function to BKE

This function will be used in another file with attribute search,
so it makes sense to share from the UI storage header.

GPencil: Fix typo error in previous commit

GPencil: Improve Flip algorithm for Interpolate

If the lines cross, but the angle is very sharp, check the distance between the extremes.

Merge branch 'blender-v2.92-release'

Fix T85844: high pressure at start of line.

m_tabletInRange is no longer set for Wintab after 2e81f2c01abd21fdbc
reverted Wintab changes. This reverts most button processing to
behavior present in 2.91.2.

Left in place is a bugfix for Windows Ink: button events while a
Windows Ink pen is in range should still be processed. Events processed
by Windows Ink and not passed to DefWindowProc do not create WM_*BUTTON
events, but button events from e.g. tablet pad express keys do create
WM_*BUTTON events and should be handled.

Revert Automatic Tablet API dynamic fallback.

Removes fallback to Windows Ink when Wintab reports no devices present.
Returns to old behavior of using only Wintab if Wintab.dll exists.

Revert high fequency mouse input for Windows.

Windows mouse history function GetMouesMovePointsEx has well documented
bugs where it receives and returns 32 bit screen coordinates, but
internally truncates to unsigned 16 bits. For mouse (relative position)
input this is not a problem as motion events and the resulting screen
coordinates reliably fit within 16 bit precision.

For tablets (absolute position) the 16 bit truncation results in
corrupt history when tablet drivers use mouse_event or SendInput from
the Windows API to move the mouse cursor. Both of these functions take
absolute mouse position as singed 32 bit value on the range of 0-65535
(or 0x0-0xFFFF) inclusive. Values larger than 0x7FFF (the largest
signed 16 bit value) are reliably corrupt when retrieved from
GetMouesMovePointsEx history. This is true regardless of whether mouse
history is retrieved using display resolution (GMMP_USE_DISPLAY_POINTS)
or high resolution points (GMMP_USE_HIGH_RESOLUTION_POINTS), the latter
of which should return points in range 0-65535.

Reviewed By: brecht

Maniphest Tasks: T85874

Differential Revision: https://developer.blender.org/D10507

WM_MOUSEMOVE cleanup.

Fix T85844: high pressure at start of line.

m_tabletInRange is no longer set for Wintab after 2e81f2c01abd21fdbc
reverted Wintab changes. This reverts most button processing to
behavior present in 2.91.2.

Left in place is a bugfix for Windows Ink: button events while a
Windows Ink pen is in range should still be processed. Events processed
by Windows Ink and not passed to DefWindowProc do not create WM_*BUTTON
events, but button events from e.g. tablet pad express keys do create
WM_*BUTTON events and should be handled.

Revert Automatic Tablet API dynamic fallback.

Removes fallback to Windows Ink when Wintab reports no devices present.
Returns to old behavior of using only Wintab if Wintab.dll exists.

Revert high fequency mouse input for Windows.

Windows mouse history function GetMouesMovePointsEx has well documented
bugs where it receives and returns 32 bit screen coordinates, but
internally truncates to unsigned 16 bits. For mouse (relative position)
input this is not a problem as motion events and the resulting screen
coordinates reliably fit within 16 bit precision.

For tablets (absolute position) the 16 bit truncation results in
corrupt history when tablet drivers use mouse_event or SendInput from
the Windows API to move the mouse cursor. Both of these functions take
absolute mouse position as singed 32 bit value on the range of 0-65535
(or 0x0-0xFFFF) inclusive. Values larger than 0x7FFF (the largest
signed 16 bit value) are reliably corrupt when retrieved from
GetMouesMovePointsEx history. This is true regardless of whether mouse
history is retrieved using display resolution (GMMP_USE_DISPLAY_POINTS)
or high resolution points (GMMP_USE_HIGH_RESOLUTION_POINTS), the latter
of which should return points in range 0-65535.

Reviewed By: brecht

Maniphest Tasks: T85874

Differential Revision: https://developer.blender.org/D10507

Fix T85895: Viewing value passes in Image Editor displays the value as alpha

There was a similar report prior to the introduction of the Image
Engine, see T74586.
This was fixed by rB6a5bd812b569 at that time, but got lost in the
refactor it seems.

Above commit introduced the `ED_space_image_get_display_channel_mask`
function that will determine
the valid bitflags for the display channel of a given ImBuf.

But since the refactor, this is not called anymore (`draw_image_main` is
not called anymore)
Now it seems we can safely reuse that said function
`ED_space_image_get_display_channel_mask` also for the Image Engine.

Maniphest Tasks: T85895

Differential Revision: https://developer.blender.org/D10510

Cleanup: Remove unecessary double negative

Annotations: Add layer Opacity paramater

This option was removed in the refactor of grease pencil and never was set again.

Geometry Nodes: improve accessing attribute meta data

This allows accessing attribute meta data like domain and data type
without having to create a `ReadAttribute`. I kept the `attribute_names`
method for now to keep the patch more self contained.

Differential Revision: https://developer.blender.org/D10511