• Closes #xxxx

One of checkbox below must be checked.

• I did not use AI to write the code of this patch.
• This PR follows our AI policy

Summary

Drive common/compression/lzma.rs through the pure-Rust xz-core port of
liblzma rather than the xz / xz-sys bindings to the C library, and drop the
android / wasm32 cfg that the C dependency carried.

Three commits:

1. lzma: export FILTERS_MAX and raise ValueError — LZMA_FILTERS_MAX
   becomes pub const FILTERS_MAX, so parse_filter_chain stops restating the
   value in a local const of its own. The over-long filter chain is reported
   as ValueError rather than LZMAError; parse_filter_chain_spec in
   Modules/_lzmamodule.c raises PyExc_ValueError, and CPython 3.14 agrees:

   $ python3 -c "import lzma; lzma.LZMACompressor(format=lzma.FORMAT_RAW, filters=[{'id':999}]*5)"
   ValueError: Too many filters - liblzma supports a maximum of 4
   

   That measurement also pins the ordering: five specs each carrying an invalid
   id report the length, not the id, so the length check stays ahead of the
   parse loop. extra_tests/snippets/stdlib_lzma.py now covers both.

2. common: port the lzma engine to xz-core — the port itself.

3. test_lzma: drop three passing expectedFailure markers —
   test_decompressor_chunks_empty, test_decompressor_chunks_maxsize and
   test_issue21872 pass on the ported engine. An unexpected success ends the
   suite as FAILED, so their markers go with the port. The other seven
   TODO: RUSTPYTHON markers in that file still fail and stay.

test_issue21872 is worth calling out, because its marker read
AssertionError: True is not false. Modules/_lzmamodule.c splits the
avail_in == 0 case in two:

else if (lzs->avail_in == 0) {
    lzs->next_in = NULL;
    if (lzs->avail_out == 0) {
        /* (avail_in==0 && avail_out==0)
           Maybe lzs's internal state still have a few bytes can
           be output, try to output them next time. */
        d->needs_input = 0;
    } else {
        d->needs_input = 1;
    }
}

The ported engine carries that second arm, which is what the test pins.

The [patch.crates-io] pin

xz-core 0.1.0-rc.0 as published registers a lzma12_optmap static
initializer. On MSVC it lands in .CRT$XIB, which _initterm_e walks as
int (__cdecl *)(void) while the initializer's Rust signature returns (), so
a Windows build exits 255 with empty stdout and stderr. That release also
overflows the alone_decoder dictionary-size check in a debug build for
dict_size == 0.

simnalamburt/xz-rs#21 fixes both and has merged, so the pin is the upstream
commit carrying it, 5bf9541, and not a fork. It drops once a release carrying
that commit reaches crates.io.

Validation

Run on macOS aarch64:

• cargo check -p rustpython-common --features lzma
• cargo check -p rustpython-stdlib
• cargo test -p rustpython-common --features lzma — 76 passed, 0 failed
• cargo clippy --all-targets -- -D warnings
• cargo build --release
• ./target/release/rustpython -m test test_lzma -v — run=121, skipped=1,
  7 expected failures, 0 unexpected successes, SUCCESS
• ./target/release/rustpython extra_tests/snippets/stdlib_lzma.py

The Cargo.lock entry is worth checking on review: xz-core must resolve to
git+https://github.com/simnalamburt/xz-rs?rev=5bf9541..., and [[patch.unused]]
must be absent. If either is not so, the published crate is being built and the
Windows failure above is still present.

AI disclosure

The engine port was written by Grok (grok-4.6) from a written specification.
Review covered the Decompressor state transitions against
Modules/_lzmamodule.c, the lc/lp/pb validation and FORMAT_ALONE filter
forwarding added in #8631 (both carried over by the port), the exception type
and message for every catch_lzma_error return value, and the checks listed
above.

Summary by CodeRabbit

• New Features

  • LZMA compression and decompression are now available on Android and WebAssembly targets when enabled.
  • Improved compatibility and reliability across supported platforms.

• Bug Fixes

  • Fixed Windows MSVC initialization issues and debug-build dictionary sizing.
  • Improved handling of LZMA filter chains, including clearer errors when too many filters are provided.
  • Updated filter-chain limits to remain consistent with the compression backend.

• Tests

  • Added coverage for invalid and oversized raw LZMA filter chains.