. "$BST_LIB"
# BST-TIMEOUT: 1200  # THREE python3 starts, and they are the whole cost. Every
#                    # assertion below either reads the filesystem/dpkg with
#                    # shell builtins or greps the single probe run's output.
#                    # The three starts are `import', the probe, and the one
#                    # UNICODE_VERSION case, which has to be a separate process
#                    # because the `auto' lookup is lru_cached and cannot be
#                    # re-read inside a running interpreter (see `env-version').
#                    # This package ships no script, so there is no fourth.
#                    #
#                    # The probe's own work is a table of ~50 width calls and one
#                    # walk of ~6000 table ranges -- microseconds beside an
#                    # interpreter start on this arm.
#                    #
#                    # 1200 is 99-python3-jstyleson's and 99-python3-jsonpointer's
#                    # figure, for the same reason: the start dominates.
#                    #
#                    # NOT MEASURED ON TARGET YET: this port's first on-target
#                    # run is the batch verify, and the measured elapsed replaces
#                    # this note in the test plan rather than in this comment,
#                    # which would cost an image build to keep the shipped suite
#                    # byte-identical.
#
# python3-wcwidth 0.2.13 (poky/meta recipes-devtools/python/python3-wcwidth_0.2.13.bb,
# `inherit pypi setuptools3 ptest', PyPI sdist) on the wasm32-hwjs glibc image.
#
# ============================================================================
# WHAT THE PACKAGE MEASURES, AND WHY IT IS NOT len().
# ============================================================================
#
# `wcwidth(c)' returns how many terminal COLUMNS one character occupies, and
# `wcswidth(s)' the same for a string. That number is neither the byte count nor
# the code-point count, and the three ways it differs are the three classes this
# suite is built around:
#
#   * a CJK ideograph occupies TWO cells -- `w-cjk', `w-hiragana', `w-hangul';
#   * a combining mark occupies ZERO -- `w-comb-acute', and `s-combining', where
#     "e" + COMBINING ACUTE is two code points and ONE column;
#   * a C0 or C1 control character has an INDETERMINATE effect, for which the
#     POSIX signature returns -1 -- `w-soh' and the three `s-c0-*' cases.
#
# The -1 is the case naive width code gets wrong, and `wcswidth' propagates it:
# ONE control anywhere in the string makes the whole answer -1 rather than the
# width of the rest. It is asserted in three positions -- first, middle and last
# -- because an implementation that checked only the leading character would pass
# one of them.
#
# ============================================================================
# THE PACKAGE IS ITS DATA, WHICH IS WHY THIS IS NOT AN IMPORT CHECK.
# ============================================================================
#
# wcwidth.py is 14 KB of binary search over 467 KB of GENERATED Unicode tables --
# table_zero.py alone is 359 KB -- and each table carries a separate range list
# for each of NINETEEN Unicode version levels. So the deliverable is not "does
# the module import", it is "did the tables land, whole, for every level".
#
# Three groups of cases say that, and no one of them would do it alone:
#
#   * `zero-ranges-*' / `wide-ranges-*' pin the table SIZES at the latest level
#     and at the earliest. Equal sizes at both would be the signature of a build
#     that shipped one table nineteen times.
#   * `table-monotonic' walks EVERY range of EVERY level checking the invariant
#     `_bisearch' relies on -- sorted, non-overlapping, start <= end. A binary
#     search over an unsorted table returns a WRONG ANSWER rather than an error,
#     which is the one failure a behavioural case sampling thirty code points
#     could not see. This is the part of upstream's own
#     tests/test_table_integrity.py that runs without pytest.
#   * `v410-emoji' / `v900-emoji' ask for U+1F600's width at two levels and get
#     1 and 2, because GRINNING FACE became wide in Unicode 9.0.0. That is what
#     says the older tables are real data rather than aliases of the latest.
#
# ============================================================================
# THE CONTROL IS THE STDLIB, NOT A TRANSCRIBED LITERAL.
# ============================================================================
#
# Every literal below was written down by a person, so `ctl-eaw' and `ctl-comb'
# are the two cases that would still mean something if all of them had been
# transcribed wrong: the probe cross-checks wcwidth against the INTERPRETER'S OWN
# Unicode database -- `unicodedata.east_asian_width' for the two-cell claim,
# `unicodedata.combining' for the zero-cell claim -- computed on target over 22
# code points, and emits the disagreements. The expected value is the empty
# string.
#
# `unicodedata' is a BUILTIN here rather than a lib-dynload .so -- this image's
# python3 is one static binary with an empty lib-dynload (BRINTOS-1476), and
# 99-python3's `builtin-modules' case asserts unicodedata is among the modules
# linked in -- so the control costs nothing and cannot fail for want of an
# extension module.
#
# The samples are long-stable code points on purpose: the control interpreter
# carries a different UCD level from the image's, and a character whose class had
# moved between them would red on version skew rather than on a defect. Nothing
# sampled has moved since Unicode 4.1. `unidata-version' records the image's
# level so a future skew is legible rather than mysterious.
#
# ============================================================================
# WHY IT PORTS AT ALL: PURE PYTHON, NO C, NOTHING BUT THE STDLIB.
# ============================================================================
#
# This image's python3 is ONE STATIC BINARY with an EMPTY lib-dynload and no
# dynamic loader (BRINTOS-1476; 99-python3 asserts both directly). A module
# carrying a compiled extension could not be imported here -- not "would be
# slow", could not be imported.
#
# wcwidth ships NO C: no .c, .pyx or .so in the sdist, and setup.py declares
# `packages' with no ext_modules at all. Its imports are `os', `sys',
# `warnings' and `functools' -- all sub-packages of the ported python3
# (BRINTOS-1476), which is why the built package's control stanza reads
# `Depends: python3-core' and nothing else and this port waits on no other
# package. `pkg-no-so' below is a REGRESSION TRIPWIRE on a version bump rather
# than a statement about today.
#
# It IS A PACKAGE, not a single module: setup.py declares `packages=['wcwidth']',
# so what lands in site-packages is a DIRECTORY of six modules -- the opposite
# shape from 99-python3-jstyleson's one file, and what lets the tables be
# separate modules. `is-package' asserts it, because a release that collapsed the
# shape would still import and still pass every behavioural case below.
#
# ============================================================================
# THE UPSTREAM ptest SUITE IS NOT RUN HERE, AND THAT IS A MEASURED DECISION.
# ============================================================================
#
# The recipe does `inherit ptest' and ships a run-ptest that calls
# `pytest --automake', so the obvious reading of this port is "run upstream's
# tests". It cannot be done on this image, for two reasons that compound:
#
#   1. This distro sets DISTRO_FEATURES without "ptest"
#   (conf/distro/brintos.conf), so
#      ptest.bbclass never adds ${PN}-ptest to PACKAGES at all. MEASURED: the
#      recipe-scope build of this port produced python3-wcwidth,
#      python3-wcwidth-dev and python3-wcwidth-dbg, and NO -ptest deb.
#   2. Producing one would put its RDEPENDS on the image -- python3-pytest and
#      python3-unittest-automake-output. pytest's own RDEPENDS names sixteen
#      packages, of which SEVEN are not on this image and are not this ticket's:
#      python3-atomicwrites, python3-attrs, python3-importlib-metadata,
#      python3-pathlib2, python3-pluggy, python3-py and python3-setuptools.
#
# Landing seven unrelated packages as a side effect of a pure-Python port is the
# shape this image recipe already refuses for acl-dev's RRECOMMENDS on attr. So
# the cases below are the only statement about this package on this image -- and
# `table-monotonic' deliberately reproduces the one part of upstream's suite that
# needs no test framework, so the decision costs coverage rather than assurance.
#
# ============================================================================
# ONE UPSTREAM LIMITATION IS ASSERTED AS ONE.
# ============================================================================
#
# POSIX wcswidth() bounds its `n' argument by the string. 0.2.13 indexes the
# string directly, so `wcswidth("ab", 10)' raises IndexError out of a width
# calculation rather than returning 2. `n-beyond' records that, with `n-0'
# through `n-4' as the controls that show `n' is otherwise honoured. It is
# 0.2.13's behaviour on x86-64 and it is UPSTREAM'S, not this platform's; it is
# asserted so the port ships a statement about what it does rather than a
# silence, and so a release that fixed it announces itself here instead of
# somewhere a caller would find it.
#
# ============================================================================
# WHERE THE ASSERTED VALUES COME FROM.
# ============================================================================
#
# Every literal below was measured on x86-64 against the sdist the recipe's
# SRC_URI sha256 pins (72ea0c06...2b5, verified by digest before measuring), and
# again against the bytes unpacked from the built
# python3-wcwidth_0.2.13-r0_wasm32.deb -- the six shipped .py files are
# byte-identical between the two, and the probe's output over them differs only
# in `module_file'. The control interpreter is CPython 3.10 and the image ships
# 3.12; that gap is stated here and in the test plan rather than papered over,
# and it is the reason `ctl-eaw'/`ctl-comb' compare against the target's own UCD
# instead of against a number from the control. The on-target result is not yet
# measured -- the batch verify is this port's first run on wasm32-hwjs.

P=python3-wcwidth

# THE CANARY, which asserts nothing about python3-wcwidth.
#
# A suite's FIRST BST result line is lost between the guest's printf and the
# host's read -- per-suite, deterministic, and NOT a defect in the host tool
# (BRINTOS-1149 established that and filed it as linux-in-the-browser#177). The
# harness computes its pass count as total-minus-failures, so THE DROPPED LINE
# IS COUNTED AS A PASS. A nothing-case therefore stands in that position and the
# eaten line costs nothing. It sits ahead of the `require' guards deliberately,
# because a `require' that skips would emit its skip as the first line and be
# eaten too.
#
# Remove this case when linux-in-the-browser#177 is fixed.
bst_ok $P/suite-canary

require python3 $P/python3-present

SITE=/usr/lib/python3.12/site-packages
PKGDIR=$SITE/wcwidth
DI=$SITE/wcwidth-0.2.13.dist-info
PKGLIST=/var/lib/dpkg/info/python3-wcwidth.list
PROBE=$BST_DATA/python3-wcwidth-probe.py
tmp=${TMPDIR:-/tmp}
W=$tmp/bst-wcwidth.$$
mkdir -p "$W" || { bst_notok $P/workdir "cannot create $W"; exit 0; }

# ------------------------------------------------------------- the payload ---
# Where the package has to land. `import wcwidth' further down is the functional
# statement; these are the ones that say WHICH copy answered, so a package that
# imported out of some other path on sys.path is distinguishable from one that
# imported out of what this image installed.
check $P/site-packages-dir test -d "$SITE"
check $P/package-dir       test -d "$PKGDIR"
check $P/module-init       test -f "$PKGDIR/__init__.py"
check $P/module-core       test -f "$PKGDIR/wcwidth.py"

# THE THREE GENERATED TABLES, AS FILES. `zero-ranges-latest' and its neighbours
# below assert what is INSIDE them; these assert that the files shipped at all,
# so a build that dropped one is distinguishable from one that shipped an empty
# table. table_zero.py is the large one -- 359 KB -- and it is the package's
# actual deliverable.
check $P/table-zero test -s "$PKGDIR/table_zero.py"
check $P/table-wide test -s "$PKGDIR/table_wide.py"
check $P/table-vs16 test -s "$PKGDIR/table_vs16.py"
check $P/table-versions test -s "$PKGDIR/unicode_versions.py"

# NO COMPILED EXTENSION ANYWHERE THE PACKAGE PUT ANYTHING. See the header: on
# this image a .so could not be loaded at all, so this is the case that fails a
# version bump which grew one -- loudly, here, instead of at `import' later.
#
# The list is copied to tmpfs first and never redirected from the rootfs: a
# top-level stdin redirect on the virtiofs root makes dup2(2) return 513 and ash
# EXITS, ending the suite silently (brintos/linux#73).
cat "$PKGLIST" > "$W/pkgf" 2>/dev/null || : > "$W/pkgf"
check $P/pkg-list-present test -s "$W/pkgf"
check_eq $P/pkg-no-so "" \
  "$(grep -E '\.so($|\.)' "$W/pkgf" 2>/dev/null)"

# EVERY REGULAR FILE THE PACKAGE SHIPS IS PYTHON OR ITS METADATA. The case above
# names the failure that is expected; this one catches the shape nobody predicted
# -- a binary, a .pyd, a stray data file -- without having to enumerate what it
# might be. This package ships no script at all, so unlike 99-python3-jsonpointer's
# equivalent there is no executable to exempt: anything outside site-packages is
# unexpected here by construction.
#
# A dpkg .list names DIRECTORIES too, and without a trailing slash, so a
# name-only filter would report every directory in the package as an unexpected
# shape; each entry is therefore tested for being a regular file first.
_bad=""
while IFS= read -r _f; do
  [ -f "$_f" ] || continue
  case "$_f" in
    *.py|*.pyc)         continue ;;
    */__pycache__/*)    continue ;;
    */*.dist-info/*)    continue ;;
  esac
  _bad="$_bad $_f"
done < "$W/pkgf"
check_eq $P/pkg-only-python "" "$_bad"

# ----------------------------------------------------------- the dist-info ---
# The installed metadata. Unlike python3-jstyleson this module DOES carry its own
# `__version__', so the dist-info is the SECOND of two independent statements of
# the version rather than the only one, and `version-agrees' below is what
# compares them.
check $P/dist-info-dir      test -d "$DI"
check $P/dist-info-metadata test -f "$DI/METADATA"
check_out $P/dist-info-version "Version: 0.2.13" cat "$DI/METADATA"
# top_level.txt names the ONE importable this package installs. It is what an
# `import wcwidth' has to match, and a wheel that renamed the package would
# change it here first.
check_out $P/dist-info-top-level "wcwidth" cat "$DI/top_level.txt"

# ------------------------------------------------------------- the package ---
# dpkg KNOWS the package. Every case above would look identical if some other
# package had installed the same files, so this is the one that says
# python3-wcwidth is a real package this image installed rather than an
# IMAGE_INSTALL line that resolved to nothing.
#
# The stanza is read from a tmpfs copy for the same reason as the file list, and
# the anchor is exact -- an unanchored match would also take the -dev or -dbg
# stanza of the same family.
sed -n '/^Package: python3-wcwidth$/,/^$/p' /var/lib/dpkg/status > "$W/st" 2>/dev/null
check $P/pkg-known test -s "$W/st"
check_out $P/pkg-installed "Status: install ok installed" cat "$W/st"

# THE RUNTIME DEPENDENCY THE RECIPE DECLARES, as dpkg recorded it -- and here it
# is the WHOLE of it. wcwidth imports only os, sys, warnings and functools, so
# the built package depends on python3-core and nothing else; that is why this
# port waits on no other package, and this case is what would say so if a release
# grew a third-party requirement.
grep '^Depends:' "$W/st" > "$W/dep" 2>/dev/null || : > "$W/dep"
check $P/pkg-depends-line test -s "$W/dep"
check_out $P/pkg-depends-core "python3-core" cat "$W/dep"

# NO -ptest PACKAGE IS INSTALLED, and that is asserted rather than left silent.
# See the header: this distro's DISTRO_FEATURES carries no "ptest" so no -ptest
# package is produced, and producing one would drag pytest's seven unported
# dependencies onto the image. A future image that DID install one should have
# this suite run it instead of quietly shipping an untested duplicate, and this
# case is what says so.
check_eq $P/no-ptest-package "absent" \
  "$(if sed -n '/^Package: python3-wcwidth-ptest$/,/^$/p' /var/lib/dpkg/status 2>/dev/null | grep -q 'install ok installed'; then echo present; else echo absent; fi)"

# ---------------------------------------------------------- the interpreter ---
# THE IMPORT, ON ITS OWN, BEFORE THE PROBE. If wcwidth cannot be imported then
# every probe case below reds at once and each names a different symptom; this
# case is what says the cause is the import. It is also the cheapest possible
# statement of the deliverable.
check $P/import python3 -c 'import wcwidth'

# ---------------------------------------------------------------- the probe ---
if [ ! -f "$PROBE" ]; then
  bst_notok $P/probe-present "$PROBE did not ship"
  rm -rf "$W"
  exit 0
fi
bst_ok $P/probe-present

# ONE interpreter start for everything that follows. bst_probe bounds it and
# keeps BOTH streams: a probe that dies early otherwise leaves every case below
# reporting got:'' and naming a symptom each, with the actual traceback nowhere.
# The bound is generous because a python3 start on this arm is tens of seconds
# before any work begins.
#
# UNICODE_VERSION is cleared for this run. It is the environment variable the
# `auto' default reads, so a value inherited from the guest's environment would
# silently move every width answer below to a different table -- and the probe
# reports what it saw in `env_unicode_version', which `probe-env-clean' asserts.
bst_probe $P/probe-run "$W/probe" 900 env -u UNICODE_VERSION python3 "$PROBE"

OUT="$W/probe"
# Read a single key. The probe emits each key at most once, so head -n 1 is
# defensive rather than load-bearing; the anchors keep `w_cjk' distinguishable
# from `s_cjk', and `versions_last' from `versions_joined'.
pv() { grep "^$1=" "$OUT" 2>/dev/null | head -n 1 | cut -d= -f2-; }

# THE EVIDENCE PLANE. If the probe emitted nothing, every case below is reading
# an absent file: each would red with got:'' and the set would read exactly like
# a broken library, when what actually happened is that no measurement was taken.
# The word `unscorable' is PARSED -- brintos_local_verify re-measures a suite that
# prints it on a fresh guest rather than trusting a shared boot's numbers
# (BRINTOS-1644) -- so it goes in this structural check, not in a comment.
#
# The cases below still print their reds: withdraw, do not hide. This case is
# what says what those reds measure.
if [ -s "$OUT" ]; then
  bst_ok $P/probe-output
else
  bst_notok $P/probe-output "the probe emitted nothing -- python3 did not start, or died before its first line; every CASE below is unscorable"
fi

# ...and it ran to the END. Every key is emitted in file order, so a probe that
# died two thirds of the way through would leave the earlier cases green and the
# later ones reading got:'' -- which looks like a partial library rather than a
# partial run. This is the case that tells those apart.
check_eq $P/probe-complete "yes" "$(pv probe_done)"
check_eq $P/probe-env-clean "<unset>" "$(pv env_unicode_version)"

# The module answered out of the directory this image installed, not from a stray
# copy on sys.path. This is what makes the payload cases above and the behaviour
# cases below statements about the SAME wcwidth.
check_eq $P/module-answered "$PKGDIR/__init__.py" "$(pv module_file)"
# ...and it is a PACKAGE. See the header: a release that collapsed the six
# modules into one file would still import and still pass every behavioural case.
check_eq $P/is-package "yes" "$(pv is_package)"

# THE VERSION, FROM BOTH PLACES, AND THEN AGAINST EACH OTHER. The module's own
# `__version__' and the installed dist-info read through importlib.metadata are
# independent statements; `version-agrees' is the one that catches a module that
# landed beside a dist-info describing some other release, which either case
# alone would pass.
check_eq $P/version-attr "0.2.13" "$(pv version_attr)"
check_eq $P/dist-version "0.2.13" "$(pv dist_version)"
check_eq $P/version-agrees "$(pv version_attr)" "$(pv dist_version)"

# THE PUBLIC API, as `__all__' declares it. This module DOES define `__all__', so
# it is asserted directly rather than reconstructed from dir() -- it is the
# package's own statement of what it exports, and a release that widened or
# narrowed it says so here.
check_eq $P/api-all "wcwidth,wcswidth,list_versions" "$(pv all_names)"
for _n in wcwidth wcswidth list_versions; do
  check_eq $P/api-$(echo "$_n" | tr _ -) "yes" "$(pv has_$_n)"
done
# The lru_cache on `wcwidth' is load-bearing on this arm rather than incidental:
# wcswidth calls it once per character, so a release that dropped it would be a
# measurable regression in every table formatter downstream.
check_eq $P/api-cache "yes" "$(pv has_cache_info)"

# ------------------------------------------------------- the version levels ---
# The list of Unicode levels is part of the PAYLOAD, not metadata about it: each
# one names a separate shipped table. The count and the endpoints are asserted
# separately from the joined list so a red says whether a level went missing or
# the list drifted wholesale.
check_eq $P/versions-count  "19"     "$(pv versions_count)"
check_eq $P/versions-first  "4.1.0"  "$(pv versions_first)"
check_eq $P/versions-last   "15.1.0" "$(pv versions_last)"
check_eq $P/versions-joined \
  "4.1.0,5.0.0,5.1.0,5.2.0,6.0.0,6.1.0,6.2.0,6.3.0,7.0.0,8.0.0,9.0.0,10.0.0,11.0.0,12.0.0,12.1.0,13.0.0,14.0.0,15.0.0,15.1.0" \
  "$(pv versions_joined)"

# THE NEAREST-MATCH RULE, one case per branch of it. A caller may pass any
# version string, so what happens to an unsupported one is part of the contract:
# a value BETWEEN two levels rounds DOWN (4.9.9 -> 4.1.0), a short dotted form
# matches on its leading pieces (8.0 -> 8.0.0), a value below every level clamps
# to the earliest (1 -> 4.1.0), and `latest' resolves to the newest shipped.
# These are upstream's own documented examples.
check_eq $P/match-4990   "4.1.0"  "$(pv match_499)"
check_eq $P/match-80     "8.0.0"  "$(pv match_80)"
check_eq $P/match-1      "4.1.0"  "$(pv match_1)"
check_eq $P/match-latest "15.1.0" "$(pv match_latest)"
check_eq $P/version-value "15,1,0" "$(pv version_value_1510)"
# An UNPARSEABLE level warns and still answers with the latest. It does not
# raise, and that is the promise worth pinning: the alternative failure mode is a
# traceback out of a width calculation inside somebody else's terminal renderer.
check_eq $P/match-bogus     "15.1.0" "$(pv match_bogus)"
check_eq $P/match-bogus-warns "yes"  "$(pv match_bogus_warned)"

# ------------------------------------------------------------- the tables ---
# One range list per level, for each of the three tables. Nineteen keys in the
# two large ones is the count the level list above demands.
check_eq $P/zero-keys "19"    "$(pv zero_keys)"
check_eq $P/wide-keys "19"    "$(pv wide_keys)"
# The VS16 table is deliberately the odd one out: it exists only from 9.0.0, so
# ONE key is correct and nineteen would be wrong.
check_eq $P/vs16-keys "9.0.0" "$(pv vs16_keys)"

# THE SIZES, at the latest level and at the earliest. Asserting only the latest
# would pass on a build that shipped one table nineteen times; the pair is what
# says the levels differ. The numbers are large and specific because they are
# generated data -- 327 zero-width ranges and 121 wide ranges at 15.1.0, against
# 145 and 34 at 4.1.0 -- and a truncated table_zero.py shows up here first.
check_eq $P/zero-ranges-latest "327" "$(pv zero_ranges_latest)"
check_eq $P/wide-ranges-latest "121" "$(pv wide_ranges_latest)"
check_eq $P/zero-ranges-410    "145" "$(pv zero_ranges_410)"
check_eq $P/wide-ranges-410     "34" "$(pv wide_ranges_410)"
check_eq $P/vs16-ranges        "113" "$(pv vs16_ranges)"

# THE STRUCTURAL INVARIANT, over every range of every level. See the header:
# `_bisearch' is a binary search, so an unsorted or overlapping table returns a
# WRONG ANSWER rather than an error -- the one failure a table of thirty sampled
# code points cannot see. This walks all of them and names the level that broke.
check_eq $P/table-monotonic "ok" "$(pv table_monotonic)"

# ------------------------------------------------- wcwidth(): ONE CHARACTER ---
# ONE CELL: ordinary printable characters, including the ones that look like they
# might not be. NBSP is a space and occupies a cell; a half-width katakana is
# east-asian and occupies ONE, which is the case that separates "east-asian" from
# "wide"; U+2588 FULL BLOCK and U+263A are drawing and symbol characters that
# render narrow. `greek-alpha' is the EAST ASIAN AMBIGUOUS class, which upstream
# resolves to 1 by policy -- a deliberate choice documented in wcwidth.py's
# header, not an accident, and a release that changed it would be a visible
# regression for every CJK terminal.
for _c in space letter_a tilde nbsp eacute box greek_alpha snowman halfwidth_ka; do
  check_eq $P/w-$(echo "$_c" | tr _ -) "1" "$(pv w_$_c)"
done

# TWO CELLS: the east-asian table answered. A CJK ideograph, a hiragana, a
# fullwidth latin letter, a hangul syllable, an emoji, and a plane-2 ideograph --
# the last of which is a NON-BMP code point, so a build whose tables had been
# truncated at U+FFFF would red here and nowhere else.
for _c in cjk hiragana fullwidth_a hangul emoji_grin cjk_extb; do
  check_eq $P/w-$(echo "$_c" | tr _ -) "2" "$(pv w_$_c)"
done

# ZERO CELLS: the zero-width table answered. A combining acute, a zero-width
# space, a zero-width joiner, a variation selector, a soft hyphen and a
# devanagari virama -- six different reasons for the same answer, so a table that
# lost one class is distinguishable from one that lost the lot.
for _c in comb_acute zwsp zwj vs16 soft_hyphen virama; do
  check_eq $P/w-$(echo "$_c" | tr _ -) "0" "$(pv w_$_c)"
done

# NOT A WIDTH AT ALL: -1, the POSIX "not printable" signal. This is the answer
# naive width code gets wrong, and it is the reason the return type is signed.
# Both control ranges are covered -- C0 below 0x20, DEL at 0x7F, and C1 from 0x80
# to 0x9F -- because they are three separate tests inside the function. TAB and
# LF are here rather than among the printables on purpose: they are C0 characters
# whose effect on the cursor is real but is not a WIDTH, and a caller who assumed
# TAB was 1 or 8 is the audience for this case.
for _c in soh bel tab lf esc del c1_80 c1_9f; do
  check_eq $P/w-$(echo "$_c" | tr _ -) "-1" "$(pv w_$_c)"
done

# THE TWO THAT ARE NEITHER. NUL is a control character and yet answers 0, not -1
# -- it has no printable effect, which is a different statement from having an
# indeterminate one -- and the empty string is not a character at all. Both are
# arguments a caller reaches by accident, both answer 0, and they reach it by
# different paths inside the function.
check_eq $P/w-nul   "0" "$(pv w_nul)"
check_eq $P/w-empty "0" "$(pv w_empty)"

# --------------------------------------------------------- wcswidth(): STRINGS ---
# The plain cases, where the answer is the sum: five ASCII characters are five
# columns, three ideographs are six, and a mixture is four.
check_eq $P/s-ascii "5" "$(pv s_ascii)"
check_eq $P/s-empty "0" "$(pv s_empty)"
check_eq $P/s-cjk   "6" "$(pv s_cjk)"
check_eq $P/s-mixed "4" "$(pv s_mixed)"

# THE CASE THE PACKAGE EXISTS FOR. "e" + COMBINING ACUTE is TWO code points and
# ONE column, so a caller measuring len() is already wrong -- by one column, on a
# string that looks like one character. `s-wide-comb' is the same statement over
# an ideograph, where the answer is 2 rather than 3.
check_eq $P/s-combining "1" "$(pv s_combining)"
check_eq $P/s-wide-comb "2" "$(pv s_wide_comb)"
# A zero-width space between two letters adds no column either.
check_eq $P/s-zwsp-mid  "2" "$(pv s_zwsp_mid)"

# THE POSIX -1 PROPAGATION. One control character ANYWHERE makes the whole answer
# -1 rather than the width of the rest, and the three positions are asserted
# separately because an implementation that checked only the leading character
# would pass `s-c0-first' and fail the other two. NUL is the control: it is a
# control character that does NOT trigger the rule, because its width is 0 rather
# than -1, so "a<NUL>b" is 2.
check_eq $P/s-c0-first "-1" "$(pv s_c0_first)"
check_eq $P/s-c0-mid   "-1" "$(pv s_c0_mid)"
check_eq $P/s-c0-last  "-1" "$(pv s_c0_last)"
check_eq $P/s-nul-mid   "2" "$(pv s_nul_mid)"

# THE TWO SEQUENCE RULES, which are what a string can do that a character cannot.
#
# ZWJ: the joiner AND THE CHARACTER AFTER IT are both skipped, so man+ZWJ+woman
# measures as the man alone -- 2, not 4. That is upstream's rule for the emoji
# sequences a terminal renders as one glyph.
check_eq $P/s-zwj-emoji "2" "$(pv s_zwj_emoji)"
# VS16 after a character the VS16 table names PROMOTES it from one cell to two,
# so U+26A0 WARNING SIGN + VS16 is 2 while the sign alone is 1. VS16 with nothing
# before it adds nothing, which is the case that says the promotion is
# conditional on the preceding character rather than unconditional.
check_eq $P/s-vs16-narrow "2" "$(pv s_vs16_narrow)"
check_eq $P/s-vs16-alone  "0" "$(pv s_vs16_alone)"

# THE `n' ARGUMENT, which exists only for compatibility with the C signature.
# Over "a<CJK>b<CJK>" the running total steps 0,1,3,4,6: the odd steps are the
# ASCII characters and the even ones the ideographs. The sequence states BOTH
# that `n' is honoured and that it counts CHARACTERS while the answer counts
# COLUMNS -- confusing those two is the bug this argument invites, and no single
# value of `n' would separate them.
check_eq $P/n-0 "0" "$(pv n_0)"
check_eq $P/n-1 "1" "$(pv n_1)"
check_eq $P/n-2 "3" "$(pv n_2)"
check_eq $P/n-3 "4" "$(pv n_3)"
check_eq $P/n-4 "6" "$(pv n_4)"
check_eq $P/n-negative "0" "$(pv n_neg)"

# THE UPSTREAM LIMITATION. See the header: POSIX bounds `n' by the string and
# 0.2.13 does not, so an `n' past the end raises IndexError out of a width
# calculation instead of returning the whole width. Asserted as IndexError rather
# than worked around, because the value a caller would want -- 2 -- is what a
# FIXED release returns, and this case is where that release announces itself.
check_eq $P/n-beyond "IndexError" "$(pv n_beyond)"

# ------------------------------------ the version-scoped widths: THE PROOF ---
# U+1F600 GRINNING FACE became WIDE in Unicode 9.0.0. Asking for its width at
# 4.1.0 and 8.0.0 gives 1, and at 9.0.0 and later 2. That transition is what says
# the nineteen shipped tables are nineteen DIFFERENT tables rather than nineteen
# references to the latest -- which no single-level case, and no count of ranges,
# could distinguish from a correct build.
check_eq $P/v410-emoji  "1" "$(pv v410_emoji)"
check_eq $P/v800-emoji  "1" "$(pv v800_emoji)"
check_eq $P/v900-emoji  "2" "$(pv v900_emoji)"
check_eq $P/v1300-emoji "2" "$(pv v1300_emoji)"
check_eq $P/v1510-emoji "2" "$(pv v1510_emoji)"
# THE CONTROL for that transition: the CJK ideograph has been wide since before
# 4.1.0, so it must answer 2 at EVERY level. Without it, a build that had made
# every old table empty would pass the pair above by answering 1 for want of
# data rather than by consulting a real 4.1.0 table.
for _v in 410 800 900 1300 1510; do
  check_eq $P/v$_v-cjk "2" "$(pv v${_v}_cjk)"
done
# The VS16 promotion is level-scoped too -- the table exists only from 9.0.0, so
# the same string measures 1 at 8.0.0 and 2 at 9.0.0. This is the string-level
# reading of the same transition.
check_eq $P/v410-vs16  "1" "$(pv v410_vs16_str)"
check_eq $P/v800-vs16  "1" "$(pv v800_vs16_str)"
check_eq $P/v900-vs16  "2" "$(pv v900_vs16_str)"
check_eq $P/v1510-vs16 "2" "$(pv v1510_vs16_str)"

# ------------------------------------------------ THE STDLIB AS THE CONTROL ---
# Every case above compares wcwidth against a value a person wrote down. These
# two compare it against the INTERPRETER'S OWN Unicode database, read on target:
# `east_asian_width' for the two-cell claim over 16 code points, and `combining'
# for the zero-cell claim over 6. They are the cases that would still be
# meaningful if every literal above had been transcribed wrong, and their
# expected value is the empty string -- the list of disagreements.
check_eq $P/ctl-eaw-count "16" "$(pv ctl_eaw_count)"
check_eq $P/ctl-eaw       ""   "$(pv ctl_eaw_bad)"
check_eq $P/ctl-comb-count "6" "$(pv ctl_comb_count)"
check_eq $P/ctl-comb      ""   "$(pv ctl_comb_bad)"
# The image's own UCD level, recorded rather than asserted against a literal.
# CPython ships a UCD generation of its own and wcwidth ships another, so the two
# can differ legitimately; what a future reader needs when `ctl-eaw' reds is to
# know WHICH levels were being compared. Asserting only that the interpreter
# answered at all keeps this from going red on an ordinary python3 bump.
check $P/unidata-version test -n "$(pv unidata_version)"

# ---------------------------------------------- the environment-driven level ---
# `auto' -- the default -- reads the UNICODE_VERSION environment variable, so a
# deployment can pin the whole image's width policy without touching a caller.
# This is a SEPARATE python3 start, and it has to be: `_wcmatch_version' is
# lru_cached, so an interpreter that has already resolved `auto' will not see a
# changed environment, and doing it inside the probe would assert the cache
# rather than the feature.
#
# 4.1.0 is chosen because it answers 1 for U+1F600 where the default answers 2,
# so the case fails on an image that ignored the variable rather than passing by
# coincidence.
check_eq $P/env-version "1" \
  "$(UNICODE_VERSION=4.1.0 python3 -c 'from wcwidth import wcwidth; print(wcwidth("\U0001f600"))' 2>/dev/null)"

rm -rf "$W"