Module xz 

The xz family (xz, unxz, xzcat, lzma, unlzma, lzcat): stuffr behaving as XZ Utils 5.8.4’s xz.

Ported from XZ Utils 5.8.4 src/xz/ by Lasse Collin and contributors (0BSD; the licence ships as LICENSE-xz), one module per C file: args.rs (args.c, with the parts of coder.c and util.c the options reach), options.rs (options.c, and liblzma’s string_conversion.c for --filters), suffix.rs (suffix.c), hardware.rs (hardware.c, with tuklib_physmem/tuklib_cpucores), message.rs (message.c, mytime.c, main.c’s exit status, tuklib_exit.c; signals.c’s handlers are in compat/host.rs), coder.rs (coder.c), file_io.rs (file_io.c), list.rs (list.c, util.c’s number formatting), index.rs (liblzma’s file_info.c, index_decoder.c and block_header_decoder.c, in pure Rust so that --list is the same on both tiers) and this module (main.c). Every string is in messages.rs.

Names (R2): all six map to this family, and the mode comes from the invoked name by args.c’s substring rule, so unxz decompresses, xzcat and lzcat decompress to standard output, and the lzma names use the .lzma format.

Hardware (R3): the real process reads the machine (physical memory, processor threads, resource limits); a sandboxed run ([run_in_counted]) sees two threads and 4 GiB, so fuzzing is deterministic, and its decoder is limited to 256 MiB (Ruling P7).

Known deviations (the source of truth: the README’s “The xz family” and the spec mirror this list):

• --version’s first line names stuffr (xz (stuffr X.Y.Z; XZ Utils 5.8.4 compatible)); its second line and --robot --version are the reference’s, on both tiers.
• The -v line’s speed and elapsed fields vary from run to run, as they do between two runs of xz.
• The output is written to a temporary file beside it and published when complete (file_io.rs); an input that is also the output is never lost, even where xz would lose it (spec deviation 4).
• The header probe (coder.rs): on the C tier it is xz’s own call (lzma_code with no output room), so the output is opened, refused or kept exactly as xz does. On the pure tier only: corruption in the first LZMA chunk’s properties byte or its range-coder start (.xz), or in the range-coder start (.lzma), is said before the output is opened, where xz finds it after opening it. Without -f the corruption is reported where xz reports that the output exists. With -f an existing output survives where xz removes it.
• Closed standard streams: run with standard input or output closed (xz <&-, or xz -c a >&-), xz says “(stdin): Read error: Bad file descriptor” or “(stdout): Write error: Bad file descriptor” and exits 1; the Rust runtime reopens a closed standard descriptor on /dev/null before main, so stuffr reads an empty input (compressing it and exiting 0, or, decompressing, saying “File format not recognized”), or writes to nowhere and exits 0.
• A parent that ignores SIGPIPE is not seen (the Rust runtime ignores it before main): writing to a closed pipe still kills stuffr, where xz says nothing and exits 0.
• The -v indicator’s one-second timer is polled against the clock after each coder call rather than set with alarm(2) (message.c’s own way where there is no SIGALRM). The lines and when they appear are xz’s; SIGALRM, SIGINFO (where the system has it) and SIGUSR1 still each ask for a line.
• No sandbox: xz’s capsicum, pledge and Landlock confinement (sandbox.c) is not ported. xz 5.8.4 on macOS has none either, and nothing it changes elsewhere is observable.
• The memory figures xz prints (-vv, --list, the memory-limit messages) and every decision against a limit come from liblzma’s formulas modelled on a configure-built xz 5.8.4 for the host (Ruling P5), on both tiers.
• Pure tier: a stream that declares a dictionary larger than its data needs commits that memory, where xz only reserves it; -M bounds it. Where the size is declared (.lzma with a known size, a .xz Block’s uncompressed size) the window is capped at that size. A sandboxed run caps the decoder at 256 MiB (Ruling P7).
• Pure tier: a .lzma whose header gives a known size and that also carries an end-of-payload marker is refused (liblzma accepts it; no xz encoder writes one).
• Pure tier: with threads and a --block-size smaller than the dictionary, the Block boundaries differ from xz’s (for a lone LZMA2 chain and no --block-list, the threaded encoder’s Blocks are at least the dictionary size).
• Pure tier: the -v line’s compressed size and ratio are those of the pure encoder’s output, which is compatible with xz’s, not equal to it.
• Pure tier: output written before a decode error may be a strict prefix of xz’s (Ruling P9: lzma-rust2 drops what its failing call had decoded), and -v’s figures at that point differ with it; the messages, exit code and file tree are xz’s. Near the end of a raw, .lzma or .lz stream, corruption can be reported as truncation (Ruling P6).

Functions

run

Entry point for the six names; name is the canonical invoked name, argv0 the program as invoked (every message’s name, as xz’s progname) and args the arguments after it.