Release by pedroac · Pull Request #12 · pedroac/sprat-cli

pedroac · 2026-05-18T07:48:18Z

No description provided.

…_png_to_func`

- Replace stbi_write_png_to_mem with stbi_write_png_to_func in spratunpack_command.cpp to avoid "undeclared identifier" errors in some stb_image_write.h versions. - Reorder CMakeLists.txt to detect dependencies (libarchive, gettext, zopflipng) before defining targets. This fixes "archive.h not found" errors on macOS where headers are in non-standard brew paths. - Remove redundant source file inclusion in executable targets by linking to spratcore library.

- Replace stbi_write_png_to_mem with stbi_write_png_to_func in spratunpack_command.cpp for better portability and fix build errors. - Remove redundant extern "C" around stb_image_write.h include. - Reorder CMakeLists.txt to ensure dependencies (libarchive, gettext, zopflipng) are detected before target definition, fixing macOS build failures. - Optimize build by removing redundant source inclusion in executable targets. - Bump version to v0.2.7 in VERSION file to resolve immutable release error in GitHub Actions.

- Exclude temporary documentation snapshots in .gitignore - Add unit tests for compare_natural() function covering: natural number sorting, mixed alphanumeric, edge cases - All 14 tests passing (2 unit + 12 integration)

- Add -DBUILD_TESTING=ON to CMake configure steps in both jobs - Update ctest invocations to emit JUnit XML test results - Add artifact upload steps for test results (7-day retention) - Add artifact retention policy (30 days for packages) - Add dependency caching for Linux apt packages and macOS Homebrew - Update Linux apt-get to use --no-upgrade flag for cached packages

- Implement --deduplicate flag with FNV-1a content hashing - Implement --gpu-compress dxt1|dxt5 for GPU-native texture formats - Implement --dilate N for color bleeding to prevent GPU filtering artifacts - Add libsquish library integration (optional dependency) - Bump cache format version (2 → 3) - Update CMakeLists.txt with squish detection

Release v0.3.0 includes the new --deduplicate feature with support for: - Exact deduplication (byte-for-byte identical images) - Perceptual deduplication (visually similar images using dHash) - Full cache management and downstream compatibility Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Enhance spratlayout command to support three deduplication modes instead of a boolean flag: - none: No deduplication (default) - exact: Hash-detect byte-for-byte identical sprites (FNV-1a) - perceptual: Detect visually similar sprites (dHash algorithm) Changes: - Updated --deduplicate argument to accept mode value - Modified cache signature to include full mode string - Added validation for mode values with helpful error messages - Updated function signatures to pass mode string instead of boolean This enables the sprat-gui UI to properly use the deduplication modes as expected by its settings UI. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

The spratcore static library includes command implementations that use libarchive functions, but wasn't linking against libarchive. This caused unresolved symbol linker errors on Windows when executables (spratlayout, spratpack, spratunpack) tried to link against spratcore. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Hoist vector/map allocations out of hot loops, replace ostringstream with string concatenation and snprintf, pre-group sprites by atlas index, rewrite prune_free_rects with mark-and-sweep, use region-only copies in dilate, deduplicate pixel extraction in unpack, and replace queue with reusable vector stack in flood fill.

- Add support for negated filter attributes in convert/layout commands - Improve --threads parameter documentation with clearer behavior and defaults - Remove max_combinations from config keys documentation - Update transform definitions

…parsing When input is a list file, the working_folder (the list file path itself) was passed directly as the root for relative path computation, producing incorrect sprite paths like "../frames/b.png". Pass parent_path() instead at both build_layout_output_text call sites. The layout parser also didn't handle the "root" line emitted by spratlayout, causing spratpack to fail with "Unknown line: root ...". Add root field to Layout struct, parse it in layout_parser, and resolve relative sprite/alias paths against it in spratpack.

…ount64 __builtin_popcountll is a GCC/Clang intrinsic not available on MSVC. Add a popcount64() wrapper that uses __popcnt64 via <intrin.h> on Windows and __builtin_popcountll elsewhere.

…placeholders - Replace specialized conditional blocks ([atlas_path], [rotated], [if_animations], etc.) with a general [if ATTR="VALUE"]...[/if] syntax usable at any nesting level - Restructure JSON transform: move sprites to a flat top-level array; group spatial data into nested rect, pivot, and trim objects - Add {{name_css}} placeholder for CSS-safe sprite class names; CSS transform now uses .sprite-{{name_css}} instead of .sprite-{{index}} - Add {{sprite_names}}/{{sprite_names_json}}/{{sprite_names_csv}} placeholders for animation frame display names - Add --transforms-dir and --default-profiles-config diagnostic flags - Add label= field to profile definitions; adjust desktop/legacy default padding; enable rotate in space profile - Fix parallel CMake build race when copying transforms/ by consolidating into a single sprat_copy_transforms custom target - Improve image cache: validate dimensions on save, evict oldest entries when over the size limit instead of refusing to save - Fix sprite path resolution: index all path suffixes so absolute or temp-prefixed paths match layout entries correctly - Capitalize built-in transform names (JSON, CSV, XML, CSS) - Update man page date and document new features

gemini-code-assist

Code Review

This pull request updates the sprat-cli suite to version 0.5.1, introducing sprite deduplication, artifact reduction via dilation, and GPU-native texture compression. It refactors the transform system for better metadata control and optimizes core packing algorithms. Feedback focuses on reducing memory overhead during dilation, adopting modern C++20 idioms, ensuring arithmetic safety against overflows, and improving maintainability by replacing magic numbers and manual hash implementations with named constants and shared utilities.

gemini-code-assist · 2026-05-18T07:50:59Z

+
+    // Two-buffer approach: read_buf holds the previous pass state, atlas is the write target.
+    // After each pass we copy only the affected sprite region back instead of the full atlas.
+    std::vector<unsigned char> read_buf = atlas;


std::vector<unsigned char> read_buf = atlas; creates a full copy of the atlas buffer. For large atlases (e.g., 16k x 16k), this can consume a significant amount of memory (1GB for RGBA). Consider using a smaller temporary buffer only for the region being dilated or optimizing the two-buffer approach to avoid a full atlas copy.

gemini-code-assist · 2026-05-18T07:50:59Z

+#ifdef _MSC_VER
+static inline int popcount64(unsigned long long x) { return (int)__popcnt64(x); }
+#else
+static inline int popcount64(unsigned long long x) { return __builtin_popcountll(x); }
+#endif


Since the project is using C++20 features (like std::ranges::sort), it is better to use std::popcount from the <bit> header instead of manual compiler-specific intrinsics. This improves portability and readability.

Suggested change

#ifdef _MSC_VER

static inline int popcount64(unsigned long long x) { return (int)__popcnt64(x); }

#else

static inline int popcount64(unsigned long long x) { return __builtin_popcountll(x); }

#endif

static inline int popcount64(unsigned long long x) { return std::popcount(x); }

gemini-code-assist · 2026-05-18T07:50:59Z

 #include <unordered_map>
 #include <unordered_set>
 #include <optional>
+#include <numeric>
+#include <cstring>


Please include the <bit> header to support the use of std::popcount.

Suggested change

#include <unordered_map>

#include <unordered_set>

#include <optional>

#include <numeric>

#include <cstring>

#include <unordered_map>

#include <unordered_set>

#include <optional>

#include <numeric>

#include <cstring>

#include <bit>

gemini-code-assist · 2026-05-18T07:50:59Z

+        size_t write = 0;
+        for (size_t ri = 0; ri < free_rects.size(); ++ri) {
+            if (free_rects[ri].w > 0 && free_rects[ri].h > 0) {
+                if (write != ri) {
+                    free_rects[write] = free_rects[ri];
+                }
+                ++write;
            }
        }
-        prune_free_rects(free_rects);
+        free_rects.resize(write);


This manual compaction loop can be replaced with std::erase_if (available since C++20), which is more idiomatic and concise.

Suggested change

size_t write = 0;

for (size_t ri = 0; ri < free_rects.size(); ++ri) {

if (free_rects[ri].w > 0 && free_rects[ri].h > 0) {

if (write != ri) {

free_rects[write] = free_rects[ri];

}

++write;

}

}

prune_free_rects(free_rects);

free_rects.resize(write);

std::erase_if(free_rects, [](const Rect& r) { return r.w <= 0 || r.h <= 0; });

gemini-code-assist · 2026-05-18T07:50:59Z

+                uint64_t fnv = 14695981039346656037ULL;
+                for (size_t bi = 0; bi < nbytes; ++bi) {
+                    fnv ^= px[bi];
+                    fnv *= 1099511628211ULL;
+                }
+                entry_content_hash = fnv;


The PR adds src/core/fnv1a.h which provides a reusable fnv1a_hash function. It should be used here instead of re-implementing the algorithm manually with magic numbers.

Suggested change

uint64_t fnv = 14695981039346656037ULL;

for (size_t bi = 0; bi < nbytes; ++bi) {

fnv ^= px[bi];

fnv *= 1099511628211ULL;

}

entry_content_hash = fnv;

entry_content_hash = sprat::core::fnv1a_hash(px, nbytes);

gemini-code-assist · 2026-05-18T07:50:59Z

+        }
+        sprites = std::move(deduped);
+    } else if (deduplicateMode == "perceptual") {
+        // O(N²) pairwise + union-find dedup


The perceptual deduplication uses an $O(N^2)$ pairwise comparison. While it is gated by sprite dimensions, it could still be a performance bottleneck for large asset sets with many sprites of the same size. Consider using a more efficient spatial indexing structure (like a BK-tree or VP-tree) if performance becomes an issue for large projects.

gemini-code-assist · 2026-05-18T07:50:59Z

+        // Union-find
+        std::vector<size_t> parent(N);
+        std::iota(parent.begin(), parent.end(), 0);
+        std::function<size_t(size_t)> find = [&](size_t x) -> size_t {


std::function has a small overhead due to type erasure. Since the lambda find is only used locally and is implemented iteratively (not recursively), using auto is more efficient.

Suggested change

std::function<size_t(size_t)> find = [&](size_t x) -> size_t {

auto find = [&](size_t x) -> size_t {

gemini-code-assist · 2026-05-18T07:51:00Z

+    size_t compressed_bytes = (format == "dxt1" || format == "DXT1")
+                              ? (width * height) / 2
+                              : width * height;


The calculation of compressed_bytes uses int for width and height. While unlikely given typical atlas limits, width * height could overflow a 32-bit signed integer. It is safer to cast to size_t before multiplication.

Suggested change

size_t compressed_bytes = (format == "dxt1" || format == "DXT1")

? (width * height) / 2

: width * height;

size_t compressed_bytes = (format == "dxt1" || format == "DXT1")

? (static_cast<size_t>(width) * height) / 2

: static_cast<size_t>(width) * height;

gemini-code-assist · 2026-05-18T07:51:00Z

+    DdsHeader header = {};
+    header.magic = 0x20534444;  // "DDS "
+    header.size = 124;
+    header.flags = 0x0001 | 0x0002 | 0x0004 | 0x1000;  // CAPS | HEIGHT | WIDTH | LINEARSIZE


The DDS header flags use magic numbers. Using named constants (e.g., DDSD_CAPS, DDSD_HEIGHT, etc.) would improve readability and maintainability.

On multi-config generators (MSVC, Xcode) CMake places binaries in a per-config subdirectory (e.g. build/Release/), but sprat_copy_transforms was copying transforms/ to the bare CMAKE_RUNTIME_OUTPUT_DIRECTORY or CMAKE_CURRENT_BINARY_DIR without the config suffix. At runtime find_transforms_dir() looks for transforms/ next to the executable and could not find them, causing the convert and output_pattern CI tests to fail on Windows with "Missing transform in list" and "Failed to open transform file". Detect multi-config generators via GENERATOR_IS_MULTI_CONFIG and append /$<CONFIG> to the destination path so transforms land in the same directory as the binaries, matching the behaviour of the old per-target $<TARGET_FILE_DIR:...> commands.

Search order is now: executable directory → user directory → global system directory. The CWD candidate for profiles config is removed. The bare relative "transforms" fallback is replaced with the global install path. Platform user directories are now correct: - Linux: XDG_CONFIG_HOME (profiles) and XDG_DATA_HOME (transforms), defaulting to ~/.config and ~/.local/share respectively - macOS: ~/Library/Application Support/sprat/ for both (was ~/Library/Preferences/ which is for .plist files only) - Windows: %APPDATA%\sprat\ (unchanged)

spratlayout no longer sorts sprites by filename when reading from a folder. The default is now none for all input types, allowing the packer to reorder sprites freely for better packing efficiency. Use --sort name to restore alphabetical ordering. Man page updates: - spratlayout synopsis now shows folder|file|- to document all three input modes (directory, list file, TAR from stdin) - Input modes section explains each mode including the - stdin TAR path and its use with spratunpack output - spratunpack atlas input and output documented as separate sections - Profile config search order updated: beside executable, then user directory (with correct per-platform paths), then global - Removed --max-combinations (feature was removed) - --sort description clarified; default updated to none - Transform file search path documented in spratconvert section - Examples expanded: list file input, full pipeline with conversion, stdin TAR round-trip (spratunpack | spratlayout -), tar cf - pipe

- Parallelize sprite loading and optimize compact packing search. Implement dynamic scheduling (atomic work index) for packing threads. - Merge guided and shelf packing passes and add early-exit pruning. - Enable recursive directory scanning by default in spratlayout. - Add support for .spratlayoutignore and 'exclude' directive in list files. - Introduce 'root' directive for animations, markers, and list files. - Document changes in README.md and man pages. - Add regression tests for recursive scanning and exclusions.

Replace the animation alias h-flip/v-flip token pair with a single flip keyword followed by a value: "h", "v", or "vh". The layout file syntax changes from: alias "run" h-flip v-flip to: alias "run" flip vh The template variable is similarly unified: the two boolean vars h_flip/v_flip are replaced by a single string var flip. Built-in transforms (JSON, CSV, XML) are updated accordingly. Add --sort stable[:<metric>] to spratlayout. This mode sorts sprites deterministically by a geometric metric (area, maxside, height, width, or perimeter; default: area) with natural path order as the tiebreaker. Sources are always sorted by path first, fixing filesystem non-determinism. The stable order is preserved through the packing loop without trying alternative sort modes, so the layout and file assignment are reproducible across runs. Adding or removing sprites only locally disturbs the atlas rather than scrambling the whole ordering. Update spratprofiles.cfg with more sensible defaults: desktop gains explicit max dimensions (8192×8192), legacy switches to optimize=gpu for square POT textures preferred by old hardware and bumps its limit to 2048, css switches from fast to compact mode for better packing quality, and redundant scale=1 entries are removed throughout.

Add --stdin-list option to read image paths from stdin Introduces a new `--stdin-list` flag that allows piping a list of image paths (one per line) directly to spratlayout without requiring a folder argument or a list file on disk. Refactors the list-parsing logic into a shared `parse_list_stream` lambda, eliminating duplication between the ListFile and new StdinList input paths. Both now support the same `exclude` and `root` directives. On Windows, stdin is explicitly set to text mode when using --stdin-list.

Every frame is placed in a uniform cell (max frame width × max frame height), arranged left-to-right, top-to-bottom, so any sprite can be looked up by (col, row) without parsing the layout file. Column count targets a square atlas and widens automatically when --max-width or --max-height are set. Multipack splits the sequence into equal-capacity grid atlases. Tight-bounds trimming is suppressed so cell dimensions stay uniform even with padding > 0. New profile "grid" added to spratprofiles.cfg; man page updated.

- Added new template placeholders: {{unity_y}}, {{pivot_x_norm}}, {{pivot_y_norm}}, {{pivot_y_norm_raw}}, {{name_hash}}, and {{name_hash_hex}}. - Created unity.json and unity.meta transforms for seamless Unity engine integration. - Updated Phaser and Godot transforms to include normalized pivot data. - Updated README.md and man pages with documentation for the new fields. - Added automated test/unity_test.sh to the CMake test suite.

Renames the --output flag to --atlas in both spratconvert and spratpack commands to better reflect its purpose as an atlas path pattern. - Adds -a as the new preferred short flag. - Maintains backward compatibility for --output and -o. - Updates man pages, README, and help messages to reflect new naming. - Updates regression tests to verify both new and legacy flags.

Remove the custom section-based DSL ({{placeholders}}, [meta]/[header]/ [sprites] tags) and replace it with google/jsonnet v0.20.0 as the transform engine. Transforms are now .jsonnet files that receive the full layout data as std.extVar("sprat") and return a JSON object with a content or files field. - Add google/jsonnet via CMake FetchContent, pinned to SHA f45e01d (v0.20.0); link libjsonnet++ and libjsonnet_static for correct static linking on all platforms - Remove ~1400 lines of DSL parser/renderer from spratconvert_command.cpp; add build_sprat_json(), evaluate_transform(), and parse_transform_result() - Add sprat.libsonnet with format_double() (works around a %g bug in Jsonnet v0.20) and consecutive_runs() for non-contiguous frameTags - Replace all 11 .transform files with 14 .jsonnet files: json, csv, xml, css, aseprite, libgdx, godot, phaser-hash, phaser-array, phaser-anims, plist, unity.json, unity.meta, unity.anim - Multi-file transforms (unity.anim) return a files array instead of content and write one file per animation under --output-dir - Update tests to Jsonnet format and add coverage for non-contiguous animations (aseprite frameTags), multipack atlas_index, and empty animations file - Update README: replace DSL placeholder reference with Jsonnet data model and built-in transform table

Grid mode was silently accepting sprites of different sizes and placing them using a cell stride derived from the largest sprite, while each sprite's w/h in the layout remained its own original dimensions. This made the layout inconsistent: coordinates implied one cell size, metadata implied another. Grid mode now validates that all sprites have the same dimensions before packing. If any sprite differs, a diagnostic names the offending file and the expected size, and the layout is rejected. The max-dimension loop is also replaced with a direct cell_w = sprite_w + padding since the sizes are guaranteed equal after validation. The fix applies to both the single-atlas path (pack_grid) and the multipack path (pack_atlases).

pedroac and others added 20 commits March 27, 2026 01:06

Fix build error by replacing stbi_write_png_to_mem with `stbi_write…

4ecc8d3

…_png_to_func`

chore: enhance project organization and unit test coverage

ce1f669

- Exclude temporary documentation snapshots in .gitignore - Add unit tests for compare_natural() function covering: natural number sorting, mixed alphanumeric, edge cases - All 14 tests passing (2 unit + 12 integration)

Update version

6ddb8ee

chore: Bump version to v0.3.2

3a8f84c

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Release v0.3.3 - libarchive linking fixes

2a9147e

Bumb version

82357b7

perf: Optimize layout generation hot paths

c5fa888

Bump version

b4d1441

Release v0.4.0

71e21ac

- Add support for negated filter attributes in convert/layout commands - Improve --threads parameter documentation with clearer behavior and defaults - Remove max_combinations from config keys documentation - Update transform definitions

Fix MSVC build: replace __builtin_popcountll with cross-platform popc…

cfcb00d

…ount64 __builtin_popcountll is a GCC/Clang intrinsic not available on MSVC. Add a popcount64() wrapper that uses __popcnt64 via <intrin.h> on Windows and __builtin_popcountll elsewhere.

gemini-code-assist Bot reviewed May 18, 2026

View reviewed changes

claude added 2 commits May 18, 2026 09:22

pedroac force-pushed the release branch from a9d4d4f to f2ddf59 Compare May 18, 2026 11:35

pedroac and others added 4 commits May 18, 2026 13:09

Merge branch 'main' into release

66364d4

pedroac force-pushed the release branch from 6fba787 to 022995b Compare May 23, 2026 14:23

claude added 5 commits May 25, 2026 10:27

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Release#12

Release#12
pedroac wants to merge 32 commits into
mainfrom
release

pedroac commented May 18, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

gemini-code-assist Bot May 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	std::function<size_t(size_t)> find = [&](size_t x) -> size_t {
	auto find = [&](size_t x) -> size_t {

Conversation

pedroac commented May 18, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 18, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants