1.8 KiB
Screenshot dimension validation
RX_VK_SHOT_W and RX_VK_SHOT_H accept positive decimal integers. Missing variables
retain the1280x720 default. Empty, signed, whitespace-containing, partially numeric,
zero and overflowing values disable screenshots for that process and report the
invalid request once. They do not fall back to a different requested capture size.
Before any screenshot allocation, the dimensions must fit the device's
maxImageDimension2D, the RGBA readback's CPU addressable size, and the existing
PNG encoder. That encoder writes one IDAT with stored deflate blocks: the checked
size includes its zlib header, five bytes per block and Adler32. The
PNG chunk limit is2^31-1 bytes.
This is a format/capability bound; it does not add a memory budget or guarantee
that a valid request can be allocated on a particular GPU.
Rejected requests never enter the capture producer. Window presentation continues; headless presentation keeps its existing offscreen policy and produces no capture for an invalid size. The fixed-size binary frame logger is independent. Valid standard, ultrawide, odd and small dimensions retain their requested sizes.
tests/test_capture_dimensions.py compiles production setup, the real capture
entry/presentation conditions, and the PNG writer. A Vulkan allocation model checks
rejection before calls, reports once, and preserves window/headless routing. A
small real PNG validates chunk CRCs, decoded scanlines and stored-block sizing.
Large dimension/format boundaries use handles only, never large allocations.
Resource readiness in #1472 remains separate. In a combined source tree its sliced setup test must include the actual dimension guard and initialized physical-device limits before exercising allocation failures; it must not stub the guard away.