BaselineCorpus.h File Reference

The frozen-baseline corpus: the fixed set of Donner-owned scenes whose pixels and structural counters are captured from the current production renderer and committed under baselines/. More...

#include <cstdint>
#include <string>
#include <string_view>
#include <vector>
#include "donner/gpu/tests/BaselineScene.h"

Include dependency graph for BaselineCorpus.h:

This graph shows which files directly or indirectly include this file:

Classes
struct	donner::gpu::baseline::CorpusScene
	One named scene: a draw-ordered list of filled paths over a transparent background. More...
struct	donner::gpu::baseline::AxisCounters
	Backend-independent per-axis encode counters, mirroring the CPU encoder's own statistics. More...
struct	donner::gpu::baseline::PathCounters
	Backend-independent structural counters for one encoded path. More...
struct	donner::gpu::baseline::SceneCounters
	Structural counters for one scene. More...

Namespaces
namespace	donner
	Top-level Donner namespace, which is split into different sub-namespaces such as donner::svg and donner::css.

Functions
std::vector< CorpusScene >	donner::gpu::baseline::Corpus ()
	The corpus, in a stable order. Adding, removing, or reordering a scene changes the frozen set and requires re-running both the capture binary and the counter regeneration command.
std::vector< SceneCounters >	donner::gpu::baseline::ComputeCorpusCounters ()
	Encodes every corpus path through the production CPU encoder and returns its counters. Pure CPU work: no GPU device, no windowing system, and no platform driver is involved.
std::string	donner::gpu::baseline::CorpusCountersJson ()
	Serializes ComputeCorpusCounters() as the committed manifest text, byte-for-byte.

Variables
constexpr uint32_t	donner::gpu::baseline::kCorpusSize = tests::kBaselineSize
	Render target size shared by every corpus scene (square RGBA8, transparent background).

Detailed Description

The frozen-baseline corpus: the fixed set of Donner-owned scenes whose pixels and structural counters are captured from the current production renderer and committed under baselines/.

Every scene is authored here from literal geometry, so the corpus never depends on external content and re-encoding it on any host reproduces the same numbers. The corpus is the shared input for three consumers: the capture binary that writes the committed PNGs, the pixel check-mode test that re-renders and diffs against them, and the counter freshness test that re-derives the structural numbers with no GPU at all.