drawCircleHelper and fillCircleHelper wrote pixels without startWrite,
so direct calls drew nothing on SPI displays (#437). They now open
their own write, and fillCircle, drawRoundRect and fillRoundRect close
theirs first so writes never nest.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Deleting a canvas or display through an Adafruit_GFX pointer called
only the base destructor, so the canvas buffer leaked (#369). AVR is
left out because a virtual destructor links malloc and free into
display sketches that never use the heap (+602 bytes on an Uno).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
writePixels and writeColor used hwspi._spi without checking the connection,
so displays on software SPI hung on the first fill.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Each 20000-pixel pass took about 1.65 s with no yield on a 3.5" HX8357,
which tripped the soft watchdog. 5000 pixels is about 0.4 s.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
yield() panics when drawing from a timer or Ticker callback (#285).
optimistic_yield(10000) still yields on long draws so the watchdog
stays fed, and does nothing where yielding is not allowed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
delay(1) ran on every writeColor call, making small fills about 3x
slower (#479). delay(0) no longer yields on core 3.x and trips the
soft WDT, and yield() panics when drawing from SYS context.
optimistic_yield(10000) yields only after 10 ms and is a no-op where
yielding is not allowed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
If malloc fails in a GFXcanvas1/8/16 constructor, buffer is NULL and
fillRect, lines and text write through it and crash the board. Set the
canvas size to 0x0 instead so every draw call clips out, and sketches
can check width()/height() or getBuffer().
Based on #441 by @aremmell and the 0x0 idea from @jepler.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Measuring at the button position predicts a line wrap that print()
never does, so a wide label on a button near the right edge moved
off its button. Measure at (0,0) and offset by the returned x/y.
The charBounds virtual change is unrelated to this fix, so it is
dropped.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
GFXfont stores first and last as uint16_t, but the custom-font paths read them as bytes. Use word-width reads while retaining the existing 8-bit character API, so the stored range is interpreted correctly without changing the font format or adding UTF-8 behavior.
Constraint: Keep the public 8-bit character API and limit the patch to issue #490\nRejected: UTF-8/API refactor | broader scope is outside the reported portability bug\nConfidence: high\nScope-risk: narrow\nReversibility: clean\nDirective: Keep GFXfont first/last accesses word-width when reading the stored range\nTested: MinGW GCC 8 -Wall -Wextra -Werror with an endian-aware host fixture; WSL GCC AddressSanitizer and UBSan; baseline fixture fails with 0x0 bounds while the fix passes with 5x7\nNot-tested: Physical Arduino board hardware
Adafruit_GFX_Button's constructor only initializes _gfx, leaving
currstate/laststate as indeterminate memory. justPressed()/
justReleased() read these before the first press() call, so they can
report a spurious press/release transition on a button that was never
actually touched.
Fixes#401
Signed-off-by: 94xhn <87560781+94xhn@users.noreply.github.com>
* GFXcanvas: Allow subclass to take charge of buffer allocations
In Adafruit_dvhstx, the low level driver takes charge of the framebuffer
allocations. Consequently, any allocation by the GFXcanvas class would be
an unused waste of memory.
Introduce a new constructor parameter `bool allocate_buffer=true`. The
default preserves the old behavior of allocating the buffer in the
constructor and freeing it in the destructor, while passing in `false`
skips buffer allocation & freeing.
In this case, the subclass must allocate the buffer before the first
drawing operation (such as in its constructor or begin method), and
free it in its own destructor or end method.
The subclass can also freely change the buffer pointer at runtime, for
instance in a swap method if the underlying framebuffer is double-buffered.
* experimentally run a different clang-format
This runs version 19.1.7: uvx clang-format -i *.cpp *.h
the changes look similar to what was shown during CI
* undo one edit that ci doesn't like
Mikey Sklar
Tyeth Gundry
Adabot
94xhn
BillMerryman
dforsi
tyeth-ai-assistedandAdabot
Bob Moss
DS-Aditya-928
Carter Nelson
Liz
Jeff Epler