The legacy decoders (lib/legacy/zstd_v01.c .. zstd_v07.c) are next in
the Rust migration. Each of those files is a frozen snapshot of the
FSE/Huff0 entropy coders and frame logic of one historical release, so
their ports must not reuse the modern Rust entropy modules and must not
share code with each other: outputs and error codes have to stay
byte-identical to the frozen C forever. This commit installs the
build-system scaffolding so seven per-version ports can land
independently, each adding only its own module file plus a one-line
registration in rust/src/legacy/mod.rs.
Cargo grows features legacy-v01 .. legacy-v07. They are never default
features: the C build defaults differ per build system, so each build
system passes the list explicitly, derived from its own legacy
configuration:
- lib/Makefile and programs/Makefile map ZSTD_LEGACY_SUPPORT=N to the
features for versions N..7 (0 disables legacy), mirroring the
ZSTD_LEGACY_FILES selection in lib/libzstd.mk.
- tests/Makefile always enables all seven features because its
ZSTDLEGACY_FILES wildcard compiles every lib/legacy/*.c regardless of
the dispatch level.
- build/meson maps legacy_level exactly like the makefiles; build/cmake
enables all seven whenever ZSTD_LEGACY_SUPPORT is ON because it
always compiles all seven C files (ZSTD_LEGACY_LEVEL only selects the
C dispatch).
Every build system also encodes the legacy selection in the Rust target
directory name (e.g. c1-d1-default-legacy5), for the same reason the
HUF mode is encoded there: a cached archive built for one configuration
must never be linked into a build expecting another. In tests/Makefile
the legacy level additionally flows into the existing HUF C-mode stamp,
so the flat C test objects (which bake -DZSTD_LEGACY_SUPPORT into the
dispatch) are rebuilt whenever the level changes. In programs/Makefile
the compress-only, decompress-only, and CLI archives keep
level-independent directories (RUST_HUF_MODE) because they are only
linked into ZSTD_LEGACY_SUPPORT=0 program variants and carry no legacy
features.
A feature whose version has not been ported yet gates nothing: the
module registration in rust/src/legacy/mod.rs is added by each port,
so enabling e.g. legacy-v05 today simply leaves that decoder in C.
This is what makes mixed C/Rust legacy levels link cleanly while the
seven ports land in any order.
Test plan:
- cd rust && cargo fmt --check && cargo clippy --all-targets
-- -D warnings && cargo test --all-targets
- cargo clippy with --no-default-features --features
decompression,legacy-v01 and with all seven legacy features
- make -C tests fuzzer && ./tests/fuzzer -i1 --no-big-tests
- make -C tests test-rust-lib-smoke; make -C tests test-legacy
- make -C lib libzstd.a with ZSTD_LEGACY_SUPPORT=0, 1 and default (5)
- cmake configure and meson setup (including -Dlegacy_level=1) emit the
expected --features lists and legacy-suffixed target directories
The dictionary-builder sources (lib/dictBuilder) are about to start moving
to Rust, beginning with divsufsort. The Rust crate previously only modeled
the compression/decompression module split plus the forced-HUF decoder
modes, so no build could express "this C configuration includes (or
excludes) dictBuilder" to Cargo. Without that, a Rust archive could carry
dictBuilder modules into a build whose C side disabled them, or worse,
omit a migrated implementation from a build whose C shims require it.
Add a `dict-builder` cargo feature and thread it through every build that
consumes the Rust static archive, mirroring exactly how each build system
already gates the dictBuilder C sources:
- rust/Cargo.toml: new `dict-builder` feature, included in the default
set because the C library builds dictBuilder by default
(ZSTD_LIB_DICTBUILDER ?= 1). The feature is empty until the first
dictBuilder module lands.
- lib/Makefile: RUST_CARGO_FEATURES gains dict-builder when
ZSTD_LIB_DICTBUILDER is enabled, following the existing
ZSTD_LIB_COMPRESSION/ZSTD_LIB_DECOMPRESSION pattern. The archive
directory naming grows a matching `b<0|1>` dimension
(c1-d1-b1-default etc.) so differently configured archives never
collide; the repeated config prefix is factored into
RUST_MODULE_CONFIG.
- programs/Makefile: the full-featured archives now request
compression,decompression,dict-builder (equal to the default set, so
the target directory stays shared with tests). The partial-library
variants gain the `b0` name dimension, and zstd-dictBuilder gets its
own lib-c1-d0-b1 archive because it compiles the dictBuilder C sources
without decompression; it previously shared the compression-only
archive, which will lack the migrated dictBuilder symbols.
- tests/Makefile: no flag change needed since tests use the crate default
feature set; a comment now records that dict-builder arrives that way.
- build/cmake/lib/CMakeLists.txt: ZSTD_BUILD_DICTBUILDER now adds the
dict-builder feature and a `b<0|1>` component in the Rust build-config
directory name, in lockstep with the DictBuilderSources gating.
- build/meson/lib/meson.build: meson compiles the dictBuilder sources
unconditionally, so the feature list and config name gain dict-builder
unconditionally (c1-d1-b1-<huf-mode>).
The `dict-builder` feature deliberately does not imply `compression`.
lib/Makefile forces ZSTD_LIB_DICTBUILDER=0 when compression is disabled,
but CMake does not couple the two options, so encoding the C-side
constraint in Cargo would make the Rust archive diverge from the C source
list in that (already unsupported) CMake configuration.
Test plan:
- cd rust && cargo build --release
- cargo build --release --no-default-features \
--features compression,decompression
- cargo build --release --no-default-features \
--features compression,dict-builder
- Full validation (fuzzer, smoke tests, dictionary byte-identity) runs
with the follow-up commit that ports divsufsort onto this scaffolding.
Build a configuration-matched Cargo archive from Meson and flatten its object
members into static libzstd. This gives C consumers one archive even though
the migrated Rust objects and remaining C code refer to each other.
Shared libraries whole-archive the Cargo output to retain Rust-only ABI
exports. The implementation supports Meson 0.50's generated-archive linking
rules, matching HUF mode and 32-bit configuration, and documents cross-build
target and archiver selection.
Test Plan:
- fresh Meson both-build smoke consumers and invalidDictionaries
- modern static/shared/both, forced-HUF, i686, install, and fuzzer paths
- real Meson 0.50 static and shared smoke builds
- cargo clippy, cargo clippy --benches, cargo clippy --tests, and nightly fmt
Refs: build/meson/README.md archive and cross-build documentation
Partial, Meson-only implementation of #2976 for non-MSVC builds.
Due to the prevalence of private symbol reuse, linking to a shared
library is simply utterly unreliable, but we still want to defer to the
shared library for installable applications. By linking to both, we can
share symbols where possible, and statically link where needed.
This means we no longer need to manually track every file that needs to
be extracted and reused.
The flip side is that MSVC completely does not support this, so for MSVC
builds we just link to a full static copy even where
-Ddefault_library=shared.
As a side benefit, by using library inclusion rather than including
extra explicit object files, the zstd program shrinks in size slightly
(~4kb).
After merging #2951 I realized that we will want to explicitly disable
assembly when we aren't including the assembly source file. Otherwise,
if some non clang/gcc compiler does support assembly, it would fail to
build.
Regression from commit a5f2c45528. It is
not possible to unconditionally add the asm sources, since not all
compilers understand the .s file extension.
Specifically for meson, only compilers inheriting from the GNU mixin
will allow a .s file at configure time.
zstd doesn't support asm for MSVC for the same basic reason; if/when
Windows asm support is added, it would involve preprocessing with nasm,
most likely.
This matches the Makefile build. Due to one private xxhash symbol in use
by the program, it recompiles a private copy of xxhash.
Due to the test binaries making extensive (?) use of private symbols, it
doesn't even attempt to link to shared libzstd, and instead, all of the
original object files are added to libtestcommon itself for private
linkage. This, too, matches the Makefile build.
Ref. #2261
`zstd_errors.h` and `zdict.h` are public headers, so they deserve to be
in the root `lib/` directory with `zstd.h`, not mixed in with our private
headers.
Instead of providing a default no-op implementation, check the symbols
for `NULL` before accessing them. Providing a default implementation
doesn't reliably work with dynamic linking. Depending on link order the
default implementations may not be overridden. By skipping the default
implementation, all link order issues are resolved. If the symbols
aren't provided the weak function will be `NULL`.
this API is deprecated, for a loong time now,
all related symbols will be removed in a future version (likely v1.5.0)
and the header file `zbuff.h` doesn't compile from `include/` anyway,
because it needs to be positioned one directory below `zstd.h`.
Also removed `cover.h` from `cmake` installer,
as it should have never been part of this list to begin with.
ddidderr
Yann Collet