[package]

name = "qsv"

version = "24.0.0" #:version

authors = ["Joel Natividad <joel@datHere.com>"]

description = "A Blazing-Fast Data-wrangling toolkit."

documentation = "https://github.com/dathere/qsv#qsv-ultra-fast-csv-data-wrangling-toolkit"

homepage = "https://qsv.dathere.com"

repository = "https://github.com/dathere/qsv"

keywords = ["csv", "geocode", "data-engineering", "etl", "opendata"]

categories = ["command-line-utilities", "parser-implementations"]

license = "MIT"

autotests = false

edition = "2024"

rust-version = "1.99"

resolver = "3"

autobins = false

include = [

"src/**/*",

"LICENSE-MIT",

"COPYING",

# Third-party attribution. THIRD_PARTY_NOTICES.md is the inventory; the

# vendored MIT-licensed assets must ship with their copyright notices, as

# their licenses require. (src/cmd/assets/LICENSE-DataTables.txt is already

# covered by "src/**/*" above, but luadate's is not — date.lua was being

# packaged without its notice.)

"THIRD_PARTY_NOTICES.md",

"resources/luau/vendor/luadate/LICENSE.txt",

"resources/luau/vendor/README.md",

"README.md",

"CHANGELOG.md",

"resources/luau/vendor/luadate/date.lua",

"resources/describegpt_defaults.toml",

# Profile engine resources — bundled via include_str! in

# src/cmd/profile/profile_spec.rs::EMBEDDED and

# src/cmd/profile/external_validate.rs::EMBEDDED_RESOURCES. These

# patterns only affect the `cargo package`/`cargo publish` tarball;

# they have no effect on builds from a git checkout (CI, local dev,

# the binary-release workflows), where the files are always present

# on disk. They're kept so the tarball stays complete (the

# include_str! resources must ship with it) in case we run

# `cargo package` or resume crates.io publishing. NOTE: we do not

# currently support `cargo install qsv` — [patch.crates-io] forks

# are stripped on publish, yielding an untested dependency tree —

# so the crates.io build path is not exercised today. The geoconnex

# paths are tracked here unconditionally — feature gating happens at

# the Rust level, not at packaging time, and the ~10KB of

# geoconnex/SHACL Turtle plus the small YAML is harmless to anyone

# who doesn't compile the feature.

"resources/profiles/*.yaml",

"resources/profiles/README.md",

"resources/dcat-us-v3/**/*",

"resources/dcat-ap-v3/shacl/*",

"resources/geoconnex/shacl/*",

]

[[bin]]

name = "qsv"

test = true

bench = false

doctest = false

path = "src/main.rs"

required-features = ["feature_capable"]

[[bin]]

name = "qsvmcp"

test = true

bench = false

doctest = false

path = "src/main.rs"

required-features = ["qsvmcp"]

[[bin]]

name = "qsvlite"

test = true

bench = false

doctest = false

path = "src/mainlite.rs"

required-features = ["lite"]

[[bin]]

name = "qsvdp"

test = true

bench = false

doctest = false

path = "src/maindp.rs"

required-features = ["datapusher_plus"]

[[test]]

name = "tests"

path = "tests/tests.rs"

[lints.clippy]

cast_possible_truncation = "allow"

cast_possible_wrap = "allow"

cast_sign_loss = "allow"

# things are often more readable this way

needless_raw_string_hashes = "allow"

cast_lossless = "allow"

module_name_repetitions = "allow"

type_complexity = "allow"

zero_prefixed_literal = "allow"

# correctly used

enum_glob_use = "allow"

result_unit_err = "allow"

# not practical

similar_names = "allow"

too_many_lines = "allow"

struct_excessive_bools = "allow"

# preference

unnecessary_wraps = "allow"

# false positive

needless_doctest_main = "allow"

# noisy

missing_errors_doc = "allow"

use_self = "allow"

cognitive_complexity = "allow"

option_if_let_else = "allow"

implicit_clone = "allow"

# pub(crate) documents crate-internal intent; equivalent to pub only because the module is private

redundant_pub_crate = "allow"

# warned

missing_asserts_for_indexing = "warn"

# code identifiers in doc comments should be backticked.

# prose-y proper nouns are exempted via doc-valid-idents in clippy.toml

doc_markdown = "warn"

[profile.release]

codegen-units = 1

debug = false

lto = true

opt-level = 3

panic = "abort"

strip = true

# luau (mlua) requires the unwind panic strategy: when a Luau callback raises an

# error (e.g. qsv_cumsum overflow, an invalid qsv_shellcmd), mlua propagates it

# across the Lua/C boundary via a Rust unwind. With panic = "abort" that unwind

# hits a non-unwinding context and aborts the process instead of surfacing the

# error (see https://github.com/dathere/qsv/issues/3937). luau is only enabled

# in the main `qsv` binary (all_features / distrib_features), so that binary

# MUST be built with this profile. The other binaries (qsvlite, qsvdp, qsvmcp)

# do NOT enable luau and keep panic = "abort" via [profile.release].

[profile.release-luau]

inherits = "release"

panic = "unwind"

[profile.release-samply]

inherits = "release"

debug = true

panic = "unwind"

strip = false

[profile.dev]

debug = "line-tables-only" # or 0 if you rarely use a debugger

[profile.dev.package."*"]

debug = 0 # no debuginfo for deps

opt-level = 1 # optional: faster test runtime for heavy deps (polars etc.), slower first build

# Fast "near-release" profile for perf iteration

[profile.release-dev]

inherits = "release-luau"

lto = "thin"

codegen-units = 16

incremental = true

[dependencies]

anstream = { version = "1.0", optional = true }

arboard = { version = "3.6.1", default-features = false, optional = true }

atoi_simd = "0.18"

base62 = { version = "2.2", optional = true }

base64-simd = { version = "0.8", optional = true }

bitvec = "1"

blake3 = { version = "1.8", features = ["rayon", "mmap"] }

bytecount = "0.6"

bytemuck = { version = "1.24", features = [

"latest_stable_rust",

], optional = true }

byteorder = "1.5"

bytes = "1"

cached = { version = "4", features = ["ahash", "redb_store", "redis_store"] }

calamine = { version = "0.36", features = ["chrono"] }

censor = { version = "0.3", optional = true }

chrono = { version = "0.4", default-features = false }

chrono-tz = "0.10"

console = { version = "0.16", optional = true }

const_format = "0.2"

cpc = { version = "5", optional = true }

crc32fast = { version = "1.4", optional = true }

crossbeam-channel = "0.5"

crossterm = { version = "0.29", optional = true }

csv = "1.4"

csv-diff = "0.1"

csv-index = "0.1"

csv-nose = { version = "1", default-features = false, features = [

"runtime-dispatch-simd",

] }

csvlens = { version = "0.15", optional = true, default-features = false, features = [

"clipboard",

] }

csvs_convert = { version = "0.13", default-features = false, features = [

"converters",

], optional = true }

dns-lookup = { version = "4", optional = true }

directories = "6.0"

dotenvy = "0.15"

dunce = "1"

dynfmt2 = { version = "0.4", default-features = false, features = ["curly"] }

# only used to validate --default-encoding labels up front (fetch/fetchpost).

# reqwest's charset feature does the actual decoding, but Encoding::for_label()

# silently falls back to UTF-8 on an unknown label, so we reject typos early.

# No compile cost: encoding_rs is already built via calamine/csv-nose/reqwest.

# Being a direct dep also lets the `nightly` feature turn on its nightly-only

# simd-accel (+ std, which multiversions more x86 code paths) for the whole dep tree.

encoding_rs = "0.8"

eudex = { version = "0.1", optional = true }

ext-sort = { version = "0.1", default-features = false, features = ["memory-limit"] }

fake = { version = "5", features = [

"derive",

"chrono",

"time",

"uuid",

"random_color",

], optional = true }

fast-float2 = "0.2"

flate2 = { version = "1", optional = true }

# same inode/file-id check for the --output-is-input guard (util::check_output_is_not_input)

# and `denull --apply`; already in the tree via walkdir. NON-optional on purpose: the guard

# prevents silent data loss and must be present in every binary, including qsvlite/qsvdp,

# which do not enable `feature_capable`.

same-file = "1.0.6"

foldhash = "0.2"

glob = { version = "0.3", optional = true }

file-format = { version = "0.29", features = ["reader"] }

filetime = "0.2"

flexi_logger = { version = "0.31", features = [

"async",

"compress",

"dont_minimize_extra_stacks",

], default-features = false }

futures = "0.3"

futures-util = "0.3"

gender_guesser = { version = "0.2", optional = true }

geojson = { version = "1", default-features = false, optional = true }

geosuggest-core = { version = "0.8.5", features = ["geoip2"], optional = true }

geosuggest-utils = { version = "0.8.5", optional = true }

geozero = { version = "0.15", features = [

"with-csv",

"with-shp",

], optional = true }

governor = { version = "0.10", optional = true }

grex = { version = "1.4", default-features = false }

gzp = { version = "2", default-features = false, features = ["snappy_default"] }

hostname-validator = "1.1"

human-panic = "2"

iana-time-zone = "0.1"

indexmap = { version = "2.14", features = ["serde"] }

indicatif = "0.18"

# already in the tree as a transitive dependency of qsv_currency; taken directly so describegpt

# can validate an ISO-4217 alpha-3 code (`Currency::from_code`) and viz can render its symbol.

iso_currency = "0.7"

itertools = "0.15"

itoa = "1"

jaq-core = "3"

jaq-json = { version = "2", features = ["serde"] }

jaq-std = "3"

jsonschema = { version = "0.58", features = [

"resolve-file",

"resolve-http",

"tls-aws-lc-rs",

], default-features = false }

libc = "0.2"

log = "0.4"

magika = { version = "1.1", optional = true }

memchr = "2"

memmap2 = "0.9"

mimalloc = { version = "0.1", default-features = false, features = [

"extended",

], optional = true }

minijinja = { version = "3.0", features = [

"json",

"loop_controls",

"serde",

"speedups",

"stacker",

"urlencode",

] }

minijinja-contrib = { version = "3.0", features = [

"datetime",

"pycompat",

"rand",

"timezone",

"unicode_wordwrap",

"wordcount",

"wordwrap",

] }

mlua = { version = "0.12", features = [

"luau",

"luau-jit",

"serialize",

], optional = true }

num_cpus = "1"

# object_store backs the `get_cloud` sub-feature (s3://, gs://, az:// sources for

# `qsv get`). default-features=false; only the cloud providers are enabled. Already

# compiled transitively via polars in distrib/qsvmcp/datapusher_plus builds, so the

# marginal cost there is ~zero. See `qsv get --help`.

object_store = { version = "0.14", default-features = false, features = [

"aws",

"azure",

"gcp",

], optional = true }

odht = "0.3"

# opener: open the generated chart/Data Schematic in the OS default app (browser for

# HTML, image viewer for PNG/SVG/...) for `qsv viz --open`. See the `viz` feature.

opener = { version = "0.9", optional = true }

# ort is needed by magika for ONNX runtime; download-binaries fetches prebuilt binaries

ort = { version = "2.0.0-rc.12", default-features = false, features = [

"download-binaries",

"tls-rustls",

], optional = true }

phf = { version = "0.14", features = ["macros"] }

# plotly backs the `viz` command (charts + auto-dashboard). default-features=false

# keeps the base `viz` build lean; `plotly_embed_js` (via the `viz` feature) makes the

# HTML self-contained, and `static_export_default` (via `viz_static`) adds PNG/SVG/PDF

# export through a webdriver-managed headless browser. NO polars dependency. See #302.

plotly = { version = "0.14.1", default-features = false, optional = true }

# viz UI localization. Locale YAMLs live under src/cmd/locales/ (NOT a top-level

# locales/ dir) so the existing "src/**/*" package include covers them; rust-i18n

# resolves the path relative to CARGO_MANIFEST_DIR at compile time.

# NOTE: the `i18n!` init MUST be invoked at a crate root (src/main.rs) because the

# `t!` macro expands to `crate::_rust_i18n_t!`. See src/cmd/viz_i18n.rs.

rust-i18n = { version = "4.2.1", optional = true }

pragmastat = "14.0"

polars-readstat-rs = { version = "0.24.2", optional = true }

polars = { version = "0.55", features = [

# "approx_quantile" powers the APPROX_QUANTILE SQL aggregate in sqlp. It is

# NOT in any umbrella qsv enables (upstream keeps it in docs-selection/full

# only), so without it polars-sql cfg-gates the function out entirely and

# sqlp rejects it as "unsupported function 'approx_quantile'".

"approx_quantile",

"asof_join",

"avro",

# "avx512" is x86_64-only — enabled via the

# [target.'cfg(target_arch = "x86_64")'.dependencies] table below

# (AVX-512 is an x86 ISA extension; inert on aarch64 et al.)

# "aws",

"binary_encoding",

# "bitwise",

"business",

# "cloud",

"coalesce",

"cross_join",

"cse",

"csv",

"decompress",

"diagonal_concat",

# "dtype-full" replaced by the specific dtypes qsv uses. The small-int

# dtypes are required: util.rs builds Polars schemas with UInt8/UInt16 for

# small non-negative integer columns, and sqlp's editable .pschema.json

# documents Int8/Int16/UInt8/UInt16 as valid cast targets. These kernels

# are feature-gated in Polars, so they must be enabled explicitly (u8/u16

# were only surviving transitively; i8/i16 were lost with dtype-full).

"dtype-i8",

"dtype-i16",

"dtype-u8",

"dtype-u16",

"dtype-date",

"dtype-datetime",

"dtype-time",

"dtype-duration",

"dtype-categorical",

"dtype-decimal",

# "extract_jsonpath",

"iejoin",

"ipc",

"json",

"lazy",

"list_eval",

"streaming",

# "object",

"parquet",

"performant",

"pivot",

"rank",

"semi_anti_join",

"serde-lazy",

"strings",

# "string_normalize",

# "string_pad",

"sql",

"timezones",

], optional = true }

pyo3 = { version = "0.29", features = ["auto-initialize"], optional = true }

# yaml_serde: serde_yaml fork maintained by The YAML Organization

# https://github.com/yaml/yaml-serde - used by the `profile` command to parse

# CKAN scheming specs (e.g. dataset-druf.yaml).

yaml_serde = { version = "0.10", optional = true }

qsv-dateparser = "0.19"

qsv_docopt = "1.10"

qsv-stats = "0.56"

qsv_currency = "0.8"

qsv-tabwriter = "2"

qsv_vader_sentiment_analysis = { version = "0.2", optional = true }

rand = "0.10"

rand_hc = "0.5"

rand_xoshiro = "0.8"

rayon = "1.12"

redis = { version = "1", features = ["ahash"], default-features = false }

regex = "1"

reqwest = { version = "0.13", features = [

"blocking",

"brotli",

"cookies",

"deflate",

"gzip",

"http2",

"json",

# query: RequestBuilder::query(), used by `geocode geoclient`. Listed EXPLICITLY for the same

# reason as charset below: all_features enables it by accident through other dependents of

# reqwest, so a local all_features build compiles while narrower feature sets (e.g. CI's

# Linux job) fail with "no method named `query`".

"query",

"rustls",

"stream",

"zstd",

# charset: makes Response::text() honor the Content-Type charset= parameter

# (via encoding_rs) and strip BOMs, instead of a blind from_utf8_lossy.

# Listed EXPLICITLY on purpose: with all_features this already arrives by

# accident, because cpc (via `apply`) and plotly_static (via `viz_static`)

# depend on reqwest WITHOUT default-features=false, and reqwest's defaults

# include charset. qsvlite/qsvdp/qsvmcp get it from nowhere else, and

# dropping `apply`/`viz_static` would otherwise silently regress .text()

# back to lossy UTF-8. Pulls encoding_rs + mime, both already in the tree.

"charset",

# system-proxy: hyper-util/client-proxy-system - resolves OS-level proxy

# configuration (macOS SystemConfiguration, Windows registry) in addition

# to the HTTP_PROXY/HTTPS_PROXY environment variables. Same accidental

# unification story as charset (cpc, plotly_static, and polars-io).

"system-proxy",

], default-features = false }

rfd = { version = "0.17", optional = true }

rmp-serde = { version = "1.3", optional = true }

rust_decimal = { version = "1.42", default-features = false }

sanitize-filename = { version = "0.6", optional = true }

simd-json = "0.18"

self_update = { version = "1.2", features = [

"archive-zip",

"compression-zip-deflate",

"github",

"progress-bar",

"reqwest",

"rustls",

"signatures",

], default-features = false, optional = true }

semver = "1"

serde = { version = "1", features = ["derive"] }

serde_json = { version = "1", features = ["preserve_order"] }

serde_urlencoded = { version = "0.7", optional = true }

sha2 = { version = "0.11", optional = true }

simdutf8 = "0.1"

spreadsheet-ods = { version = "1", optional = true }

toml = "1.1"

toon-format = { version = "0.7", default-features = false }

smallvec = "1"

snap = "1"

strsim = { version = "0.11", optional = true }

strum = { version = "0.28", features = ["phf"] }

strum_macros = "0.28"

sysinfo = "0.39"

tempfile = "3.27"

terminal-colorsaurus = { version = "1.0", optional = true }

textwrap = { version = "0.16", features = ["terminal_size"] }

thousands = { version = "0.2", optional = true }

threadpool = "1.8"

tikv-jemalloc-ctl = { version = "0.7", optional = true }

tikv-jemallocator = { version = "0.7", optional = true }

time = { version = "0.3.55", optional = true }

titlecase = { version = "3", optional = true }

tokio = { version = "1", features = ["parking_lot", "rt-multi-thread", "time"] }

unicode-width = { version = "0.2", optional = true }

uuid = { version = "1", features = ["v4", "v7"] }

url = "2.5"

webdriver-downloader = { version = "0.17", optional = true }

whatlang = { version = "0.18", optional = true }

xxhash-rust = { version = "0.8", features = ["xxh3"] }

zip = "8"

zmij = "1.0"

zstd = { version = "0.14", optional = true }

[dev-dependencies]

actix-governor = "0.10"

actix-web = { version = "4.13", default-features = false, features = [

"compress-brotli",

"compress-gzip",

] }

assert-json-diff = "2.0"

# for inflating the gzip+base64 payloads viz HTML embeds (see tests/test_viz.rs inflate helpers)

base64-simd = "0.8"

# for verifying the pinned plotly.js CDN Subresource Integrity digest against the bundle qsv

# actually embeds (see cmd::viz plotly_cdn_sri_matches_the_embedded_bundle). A dev-dependency, not

# the optional `profile` sha2, so the oracle runs for EVERY viz test feature set rather than only

# the ones that happen to enable `profile`.

sha2 = "0.11"

flate2 = "1"

newline-converter = "0.3"

# disable these dev dependencies for testing the `to` command

# as they are expensive and slow down the build/CI tests

# postgres = "0.19"

# rusqlite = { version = "0.32", features = ["bundled"] }

quickcheck = { version = "1", default-features = false }

serial_test = { version = "4.0", features = ["file_locks"] }

[patch.crates-io]

# polars-readstat-rs depends on these directly; redirect them onto the same git polars

# revision qsv pins, otherwise a second, incompatible polars tree resolves from crates.io.

polars-core = { git = "https://github.com/pola-rs/polars", rev = "099c28632d62ce41e941f88c4210a300650f7188" }

polars-arrow = { git = "https://github.com/pola-rs/polars", rev = "099c28632d62ce41e941f88c4210a300650f7188" }

# Polars (since py-2.0.0) calls HashTable::prefetch_{get,insert}, which exist only in its

# hashbrown fork (https://github.com/rust-lang/hashbrown/pull/727). Polars' own workspace

# [patch] doesn't propagate to dependents, so mirror it here; keep the rev in sync with the

# [patch.crates-io] section of the pinned polars rev's Cargo.toml. Drop once #727 is released.

hashbrown = { git = "https://github.com/pola-rs/hashbrown", rev = "d76a6f81abd8f388676aff15068236a8d21c7e1a" }

# use our tweaked fork of csv crate

# the csv crate underpins a lot of qsv's functionality, so every perf tweak helps

# for details, see https://github.com/dathere/qsv/wiki/Tuned-csv-Fork-Benchmarks

csv = { git = "https://github.com/dathere/rust-csv", branch = "qsv-tuned" }

csv-core = { git = "https://github.com/dathere/rust-csv", branch = "qsv-tuned" }

csv-index = { git = "https://github.com/dathere/rust-csv", branch = "qsv-tuned" }

# csv = { path = "../rust-csv" }

# csv-core = { path = "../rust-csv/csv-core" }

# csv-index = { path = "../rust-csv/csv-index" }

# use latest upstream now that our PR has been merged

csvlens = { git = "https://github.com/YS-L/csvlens", rev = "6acf060" }

# use our patched fork of csvs_convert to bump dependencies until our PR is merged

csvs_convert = { git = "https://github.com/jqnatividad/csvs_convert", branch = "replace-streaming-stats-redux", optional = true }

# use our patched fork of whatlang to bump hashbrown until our PR is merged

whatlang = { git = "https://github.com/jqnatividad/whatlang-rs", branch = "bump-hashbrown-to-16" }

# Our PRs have been merged upstream! We can now use the official plotly crate instead of our fork

# until the next plotly.rs release.

plotly = { git = "https://github.com/plotly/plotly.rs", rev = "00fe0512e9f924bd0fc2edfc6cc617a1604f4e0d" }

# use our patched fork until our PR is merged with bumped deps and quick-xml RUSTSEC fixes

spreadsheet-ods = { git = "https://github.com/jqnatividad/spreadsheet-ods", branch = "deps-bump-202607" }

# Polars has a much higher release tempo for its Python bindings compared

# to its underlying Rust library. See https://github.com/pola-rs/polars/releases

# It's qsv's policy to use the latest upstream of polars/py-polars

# to take advantage of Polars' latest unreleased fixes and features.

# Based on what's available at the time of qsv's release, we may need to pin polars to a py-polars tag

# or a specific commit if more revisions have been made since the latest polars/py-polars release.

# When pinning to an unreleased rev, re-add a `polars = { git = ..., rev = ... }`

# entry here AND make sure that rev's own Cargo.toml version matches the

# `version = "x.y.z"` requirement in [dependencies] below — otherwise Cargo

# silently ignores the patch (it shows up as [[patch.unused]] in Cargo.lock) and

# resolves polars from crates.io instead.

# BUILD NOTE: Be sure to set QSV_POLARS_REV to the commit short hash or tag

# of polars/py-polars before building qsv. This allows us to show the polars rev/tag in --version.

# The convention to use for QSV_POLARS_REV is:

# - if we are using a release version of Rust Polars, leave QSV_POLARS_REV empty

# - if we are using a release version of Python Polars, use the Python tag

# (e.g. py-1.19.0)

# - if we are using a commit hash, separate the tag and commit hash with a colon, in the order of

# - Python Polars tag

# - short commit hash

# (e.g. py-1.19.0:52ea381 to indicate that we are Python Polars 1.19.0,

# and the commit hash 52ea381)

# ================================

# QSV_POLARS_REV=py-2.0.0:099c286

# polars = { git = "https://github.com/pola-rs/polars", tag = "py-2.0.0", optional = true }

polars = { git = "https://github.com/pola-rs/polars", rev = "099c28632d62ce41e941f88c4210a300650f7188", optional = true }

# Apache DataSketches (t-digest, HLL, Frequent Items) — upstream C++ does not

# support big-endian targets, so we exclude the crate on big-endian builds.

# Code paths that use it are guarded by #[cfg(not(target_endian = "big"))].

[target.'cfg(not(target_endian = "big"))'.dependencies]

datasketches = { version = "0.5", features = ["frequencies", "hll", "tdigest"] }

# AVX-512 is an x86_64-only ISA extension; Polars' avx512 kernels are inert on

# other targets, so gate the feature to x86_64. Cargo unions these features with

# the base `polars` dependency above, so x86_64 builds get avx512 while

# aarch64 (and other arches) don't carry it. Polars dispatches the AVX-512

# kernels at runtime, so enabling it for all x86_64 builds is safe.

[target.'cfg(target_arch = "x86_64")'.dependencies]

polars = { version = "0.55", features = ["avx512"], optional = true }

# geosuggest-core builds the tiny Geonames index at test time from the committed TSVs under

# resources/test/geonames-mini/ (see tests/test_viz_geocode_auto.rs). A committed .rkyv would

# silently break on every geosuggest bump (the layout is version-coupled); building it with the

# Cargo.lock-unified crate version removes that risk by construction.

#

# It pulls kiddo, whose build script calls yep_cache_line_size::get_cache_line_size()

# unconditionally — and that crate exports nothing outside x86/x86_64/arm/aarch64, so the build

# script fails to compile (E0425) when kiddo is built NATIVELY on any other host. Build scripts

# compile for the HOST, so cross-compiling is unaffected; it is the native runners that break.

# Dev-dependencies are built by `cargo test` regardless of --features, so an unconditional entry

# broke the s390x CI job even though it never enables `geocode`.

#

# The cfg below is an ALLOWLIST of the architectures yep-cache-line-size actually supports (it

# mirrors that crate's own cfg in its lib.rs). It is deliberately NOT `not(target_endian = "big")`:

# the constraint is architectural, not endianness. ppc64le is little-endian yet still outside

# x86/ARM, so an endian gate left rust-ppc64le.yml — a native `cargo test` runner using the same

# feature list as s390x — just as broken. Add an arch here only when yep-cache-line-size gains it.

#

# Nothing is lost on the excluded arches: the sole consumer, tests/test_viz_geocode_auto.rs, is

# gated on `viz` + `geocode`, and neither the s390x nor the ppc64le job enables either one.

[target.'cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "arm", target_arch = "aarch64"))'.dev-dependencies]

geosuggest-core = "0.8.5"

[features]

default = ["jemallocator"]

# NOTE: no publish workflow builds the prebuilt `qsv` with distrib_features. Each one passes an

# explicit --features list that INTENTIONALLY leaves out `get`, `get_cloud` and `geoconnex`

# (see #4737). So the prebuilt `qsv`/`qsvpy*` binaries have no `get` command, no `dc:` inputs,

# no s3:///gs:///az:// sources and no `profile --profile geoconnex`. Users who need them can

# use the prebuilt `qsvdp` (get + get_cloud) or build from source with

# -F distrib_features, or selectively with -F feature_capable,get[,get_cloud] /

# -F feature_capable,geoconnex (the `qsv` bin requires feature_capable).

distrib_features = [

"feature_capable",

"apply",

"fetch",

"foreach",

"geocode",

"geoconnex",

"get",

"get_cloud",

"luau",

"mcp",

"polars",

"profile",

"readstat",

"synthesize",

"to",

"viz_static",

]

all_features = ["distrib_features", "magika", "self_update", "ui"]

# NOTE: `get` AND `get_cloud` are both deliberately absent - removing only `get` would be a

# no-op, since `get_cloud = ["get", ...]` re-enables it. `get_cloud` turns on object_store's

# aws/azure/gcp backends, which is the largest feature delta between this build and a

# distrib_features `qsv` build (object_store is in BOTH dependency graphs, so it never shows up in a crate

# count - only in a resolved-feature comparison). Dropping it costs qsvmcp the `get` command,

# `dc:` cache-prefix inputs, and s3:///gs://az:// sources.

qsvmcp = [

"feature_capable",

"geocode",

"mcp",

"polars",

"profile",

"readstat",

"self_update",

"synthesize",

"to",

"viz",

]

apply = [

"base62",

"base64-simd",

"censor",

"cpc",

"crc32fast",

"eudex",

"gender_guesser",

"qsv_vader_sentiment_analysis",

"strsim",

"thousands",

"titlecase",

"whatlang",

]

clipboard = ["arboard"]

color = ["anstream", "crossterm", "terminal-colorsaurus", "unicode-width"]

fetch = ["console", "flate2", "governor", "serde_urlencoded"]

foreach = []

# get: fetch tabular data from local/URL/CKAN/cloud sources into a managed,

# zstd-compressed, BLAKE3-hashed, auto-indexed disk cache. Downloads use ETag/

# Last-Modified conditional revalidation with streaming, ranged (parallel)

# fetches for large objects. Backs the `dc:` input prefix usable by any qsv

# command. See `qsv get --help`.

get = ["zstd", "flate2", "glob"]

# get_cloud: adds cloud object-storage sources (s3://, gs://, az://) to `get`,

# via the object_store crate. Opt-in so a bare `-F get` build stays lean;

# included in distrib_features/datapusher_plus (where polars already pulls

# object_store). NOT in qsvmcp, and NOT in the prebuilt `qsv` (see the NOTE

# above distrib_features). See `qsv get --help`.

get_cloud = ["get", "dep:object_store"]

geocode = [

"bytemuck",

"dns-lookup",

"geosuggest-core",

"geosuggest-utils",

"geozero",

"governor",

"rmp-serde",

]

# luau lookup tables auto-decompress remote .gz/.zlib/.zst/.zip/.sz sources, so

# luau pulls the flate2 + zstd codecs (.zip/.sz need no extra deps). See #1417.

luau = ["mlua", "sanitize-filename", "flate2", "zstd"]

magika = ["dep:magika", "dep:ort"]

mcp = []

polars = ["dep:polars", "bytemuck"]

readstat = ["dep:polars-readstat-rs", "polars"]

prompt = ["rfd"]

python = ["pyo3"]

# profile: CKAN scheming-driven metadata extraction (DCAT-US v3 / Croissant).

# Native Rust implementation — minijinja replaces jinja2 for formula

# evaluation; polars (SQL context) backs the SQL-requiring helpers

# (temporal_resolution, guess_accrual_periodicity). yaml_serde parses the

# scheming spec YAML. See `qsv profile --help`.

profile = ["dep:sha2", "polars", "yaml_serde"]

# geoconnex: bundled SHACL-validated linked-data profile for hydrologic

# features (https://docs.geoconnex.us/). Embeds the Geoconnex SHACL

# shapes (~10KB Turtle bundle) so `qsv profile --profile geoconnex

# --validate` can spawn `pyshacl` without external file deps.

# Requires `profile`. Available in source builds of qsv (via distrib_features)

# and as an opt-in for qsvdp; NOT in qsvlite, qsvmcp, or any prebuilt binary

# (see the NOTE above distrib_features).

geoconnex = ["profile"]

synthesize = ["dep:fake", "dep:time"]

to = ["csvs_convert", "dep:spreadsheet-ods"]

# viz: generate charts (bar/line/scatter/histogram/box) and an auto-dashboard

# (`viz smart`) from CSV data via the plotly crate. Base feature = self-contained

# interactive HTML output (plotly_embed_js); NO polars dependency. `viz smart`

# reuses qsv's in-process stats + frequency caches. See `qsv viz --help` and #302.

# NOT SUPPORTED ON BIG-ENDIAN TARGETS (s390x; also powerpc64 BE) — this applies to BOTH

# `viz` and `viz_static`, and is enforced by a compile_error! in src/cmd/mod.rs. Keep both

# out of big-endian publish workflows.

# NOTE powerpc64LE is LITTLE-endian, so the guard deliberately does not fire there. `viz`

# is simply not enabled for the ppc64le prebuilt: it has no viz CI coverage, and polars

# does not compile on PowerPC (see docs/publishing_assets/qsv-powerpc64le-*.txt).

viz = [

"dep:plotly",

"dep:opener",

"dep:geojson",

"dep:rust-i18n",

"plotly/plotly_embed_js",

"base64-simd",

]

# viz_static: adds static PNG/SVG/PDF/JPEG/WebP export to `viz` via plotly_static,

# which drives a headless Chrome/Chromium through chromedriver. NOT Firefox: this

# resolves to plotly_static/chromedriver, and chromedriver/geckodriver are mutually

# exclusive upstream features, so a geckodriver can never serve this build (#4620).

# The driver is auto-downloaded AT BUILD TIME into $HOME/.local/bin and its absolute

# path is baked in via option_env!("WEBDRIVER_DOWNLOAD_PATH"), which is meaningless on

# an end user's machine - cmd::viz::ensure_webdriver_path() repairs that at runtime by

# probing for a real chromedriver. Requires a browser at runtime, so keep it out of

# headless-only publish targets. It also

# inherits `viz`'s big-endian restriction above (enforced at compile time).

# DO NOT enable on Windows publish targets: it pulls ~54 extra crates (plotly_static,

# fantoccini, webdriver-downloader, native-tls, icu_segmenter) and the resulting `qsv`

# release compilation unit does not finish inside GitHub's 6h job cap — observed under

# both fat LTO and thin LTO+PGO instrumentation (run 31243109948). This is also why

# `qsvmcp` uses plain `viz`: publish.yml builds qsvmcp on Windows with --features=qsvmcp.

viz_static = ["viz", "plotly/static_export_default", "webdriver-downloader"]

lens = ["csvlens"]

lite = []

datapusher_plus = [

"base64-simd",

"geocode",

"get",

"get_cloud",

"polars",

"profile",

"self_update",

]

ui = ["clipboard", "color", "prompt", "lens"]

jemallocator = ["tikv-jemallocator", "tikv-jemalloc-ctl"]

# feature_capable backs the always-compiled commands that consume lookup tables

# (validate/template/describegpt). Pull flate2 + zstd so their remote lookups

# auto-decompress .gz/.zlib/.zst (alongside the always-available .zip/.sz). #1417.

# base64-simd backs describegpt's multimodal --context-file (base64-encoding image/PDF

# attachments for the LLM); it's always available wherever describegpt is (non-lite).

feature_capable = ["base64-simd", "flate2", "zstd"]

nightly = [

"crc32fast/nightly",

"pyo3/nightly",

"rand/simd_support",

"simd-json/hints",

"foldhash/nightly",

"encoding_rs/simd-accel",

"encoding_rs/std",

]

nightly-polars = ["polars/nightly", "polars/simd"]

[package.metadata.deb]

maintainer = "Konstantin Sivakov <konstantin@datHere.com>"

copyright = "2015, Andrew Gallant; 2026, datHere Inc. <www.dathere.com>"

license-file = ["LICENSE-MIT", "0"]

extended-description = """A high performance CSV data-wrangling toolkit."""

depends = "$auto"

section = "utility"

priority = "optional"

# Default feature and asset. THIRD_PARTY_NOTICES.md ships alongside the binary

# so the vendored MIT/BSD notices travel with the package, as those licenses

# require; license-file above generates /usr/share/doc/qsv/copyright.

features = ["feature_capable"]

assets = [

[

"target/release/qsv",

"/usr/local/bin/",

"755",

],

[

"THIRD_PARTY_NOTICES.md",

"/usr/share/doc/qsv/",

"644",

],

]

# Conditional features and assets

# qsvlite (`lite = []`) pulls in none of the vendored third-party components — no viz

# (plotly/DataTables), no luau (LuaDate), no profile (DCAT/geoconnex schemas) — so it ships only

# qsv's own license, via the inherited `license-file` above. Shipping the notices file here would

# credit components this binary does not contain.

[package.metadata.deb.variants.lite]

features = ["lite"]

assets = [["target/release/qsvlite", "/usr/local/bin/", "755"]]

# qsvdp DOES vendor third-party content: `datapusher_plus` enables `profile`, which embeds the

# DCAT-US (CC0-1.0), DCAT-AP (CC-BY-4.0) and geoconnex (Apache-2.0) schemas. CC-BY-4.0 requires

# attribution, so the notices must ship with this package too. `assets` REPLACES the parent

# table's list rather than extending it, hence the explicit repeat.

[package.metadata.deb.variants.datapusher_plus]

features = ["datapusher_plus"]

assets = [

[

"target/release/qsvdp",

"/usr/local/bin/",

"755",

],

[

"THIRD_PARTY_NOTICES.md",

"/usr/share/doc/qsvdp/",

"644",

],

]