. "$BST_LIB"
# enca 1.19 (meta-oe recipes-support) -- charset analyser and converter.
#
# Two halves, deliberately kept apart because they exercise different things:
#
#   DETECTION is pure computation inside libenca -- it reads bytes, scores them
#   against per-language charset tables compiled into the binary, and prints a
#   name. Nothing outside the process is involved, so every detection case below
#   must pass on any target the binary runs on at all.
#
#   CONVERSION goes out through iconv(3), and this build HAS NO ICONV INTERFACE:
#   enca's own m4/libiconv.m4 settles the question by compiling and RUNNING
#   tools/iconvcap.c to generate the charset-alias table it is then built
#   against, and under cross-compilation that binary cannot execute on the build
#   host. configure falls back to an empty table, `enca --version` reports
#   -iconv-interface, and every -x fails. That is BRINTOS-603, not this port.
#
# SIX CASES BELOW ARE THEREFORE RED ON PURPOSE, and going green is the signal
# BRINTOS-603 landed: version-iconv, koi8r-iconv, the three convert-* over 8-bit
# charsets, and enconv-convert. They are NOT downgraded to skip -- the tool is
# present and its converter half genuinely does not work, which is a `not ok` by
# the packager guide's own rule. The two conversions that pass are the ones
# needing no converter at all, which is what separates "enca cannot convert"
# from "enca is broken".
#
# Every expectation is a value measured against a native x86_64 build of the
# same 1.19 tarball, driven over the same fixtures.

require enca enca/present

D="$BST_DATA"
tmp=${TMPDIR:-/tmp}/bst-enca.$$
mkdir -p "$tmp"

# ---------------------------------------------------------------- identity ---
check_out enca/version        "enca 1.19"           enca --version
check_out enca/version-iconv  "+iconv-interface"    enca --version
check_out enca/version-langs  "+language-detection" enca --version
check_out enca/help           "Usage:  enca"        enca --help

# enca refuses to guess without a language it understands: there is no locale on
# this image, so a bare `enca FILE` is the documented rc-2 refusal, not a crash.
check_rc  enca/no-language 2 sh -c 'enca "$1" >/dev/null 2>&1' _ "$D/enca-ascii.txt"
# Every error-path case below pairs a check_rc (the status) with a check_out (the
# message). check_out also requires exit 0, so the message half runs the command
# under `|| true` -- the status is already asserted by its own case, and folding
# the two would only make a wrong rc fail twice under one name.
check_out enca/no-language-msg "Cannot determine (or understand) your language preferences" \
          sh -c 'enca "$1" 2>&1 || true' _ "$D/enca-ascii.txt"

# ------------------------------------------------------------------- lists ---
check_out enca/list-langs-ru   "russian: KOI8-R CP1251"    enca --list languages
check_out enca/list-langs-cs   "czech: ISO-8859-2 CP1250"  enca --list languages
check_eq  enca/list-langs-n    15 "$(enca --list languages | wc -l)"
check_eq  enca/list-charsets-n 32 "$(enca --list charsets  | wc -l)"
check_eq  enca/list-builtin-n  23 "$(enca --list built-in-charsets | wc -l)"
check_out enca/list-surfaces   "CRLF line terminators"     enca --list surfaces
check_out enca/list-names      "rfc1345"                   enca --list names

# --------------------------------------------------------------- detection ---
# ASCII and UTF-8 are the two a language-less run can reach; the rest need -L.
check_out enca/ascii-human "7bit ASCII characters" enca -L ru "$D/enca-ascii.txt"
check_out enca/ascii-enca  "ASCII"                 enca -L ru -e "$D/enca-ascii.txt"
check_out enca/ascii-mime  "US-ASCII"              enca -L ru -m "$D/enca-ascii.txt"
check_out enca/ascii-rfc   "ANSI_X3.4-1968"        enca -L ru -r "$D/enca-ascii.txt"

check_out enca/utf8-human  "Universal transformation format 8 bits; UTF-8" \
          enca -L ru "$D/enca-ru-utf8.txt"
check_out enca/utf8-enca   "UTF-8"  enca -L ru -e   "$D/enca-ru-utf8.txt"
check_out enca/utf8-nolang "UTF-8"  enca -L none -e "$D/enca-ru-utf8.txt"

# Russian: one fixture per charset in enca's own russian list, so a table that
# goes wrong for one encoding is not masked by the others.
check_out enca/koi8r-human "KOI8-R Cyrillic"      enca -L ru    "$D/enca-ru-koi8r.txt"
check_out enca/koi8r-enca  "KOI8-R/LF"            enca -L ru -e "$D/enca-ru-koi8r.txt"
check_out enca/koi8r-iconv "KOI8-R"               enca -L ru -i "$D/enca-ru-koi8r.txt"
check_out enca/koi8r-mime  "KOI8-R"               enca -L ru -m "$D/enca-ru-koi8r.txt"

check_out enca/cp1251-human "MS-Windows code page 1251" enca -L ru    "$D/enca-ru-cp1251.txt"
check_out enca/cp1251-enca  "CP1251/LF"                 enca -L ru -e "$D/enca-ru-cp1251.txt"
check_out enca/cp1251-mime  "windows-1251"              enca -L ru -m "$D/enca-ru-cp1251.txt"

check_out enca/iso88595-human "ISO 8859-5 standard; ISO Cyrillic" enca -L ru    "$D/enca-ru-iso88595.txt"
check_out enca/iso88595-enca  "ISO-8859-5"                        enca -L ru -e "$D/enca-ru-iso88595.txt"
check_out enca/iso88595-rfc   "ISO_8859-5:1988"                   enca -L ru -r "$D/enca-ru-iso88595.txt"

check_out enca/ibm866-human "IBM/MS code page 866" enca -L ru    "$D/enca-ru-ibm866.txt"
check_out enca/ibm866-enca  "IBM866/LF"            enca -L ru -e "$D/enca-ru-ibm866.txt"

# Czech, on a second language's tables and a second name form (cstocs).
check_out enca/cs-il2-human "ISO 8859-2 standard; ISO Latin 2" enca -L cs    "$D/enca-cs-iso88592.txt"
check_out enca/cs-il2-enca  "ISO-8859-2"                       enca -L cs -e "$D/enca-cs-iso88592.txt"
check_out enca/cs-il2-cstocs "il2"                             enca -L cs -s "$D/enca-cs-iso88592.txt"
check_out enca/cs-1250-human "MS-Windows code page 1250"       enca -L cs    "$D/enca-cs-cp1250.txt"
check_out enca/cs-1250-enca  "CP1250/LF"                       enca -L cs -e "$D/enca-cs-cp1250.txt"
check_out enca/cs-1250-mime  "windows-1250"                    enca -L cs -m "$D/enca-cs-cp1250.txt"

# Surfaces. enca's reference line terminator is CRLF, so a CRLF file reports NO
# surface and an LF file reports one -- backwards from the Unix intuition, and
# pinned here so a change to that convention is caught rather than shrugged at.
check_out enca/surface-lf   "LF line terminators" enca -L ru "$D/enca-ru-koi8r.txt"
check_out enca/surface-crlf "KOI8-R Cyrillic"     enca -L ru "$D/enca-ru-koi8r-crlf.txt"
# stdout only: this case is an EXACT match on the name, and enca writes an
# unrelated diagnostic to stderr on any image with no /usr/share/locale/
# locale.alias -- folding that in would make a surface assertion fail for a
# reason that has nothing to do with surfaces.
crlf=$(enca -L ru -e "$D/enca-ru-koi8r-crlf.txt" 2>/dev/null)
check_eq  enca/surface-crlf-bare "KOI8-R" "$crlf"

# -L none reaches only the multibyte/ASCII cases: an 8-bit Cyrillic file is
# genuinely unrecognizable without a language, and that is a result, not a bug.
check_rc  enca/nolang-8bit-rc 1 sh -c 'enca -L none "$1" >/dev/null 2>&1' _ "$D/enca-ru-koi8r.txt"
check_out enca/nolang-8bit    "Unrecognized encoding" sh -c 'enca -L none "$1" 2>&1 || true' _ "$D/enca-ru-koi8r.txt"

# stdin, and the multi-file form that prefixes each result with its filename.
check_out enca/stdin      "KOI8-R Cyrillic"  sh -c 'enca -L ru < "$1" 2>&1' _ "$D/enca-ru-koi8r.txt"
check_out enca/multi-file "enca-ru-koi8r.txt: KOI8-R Cyrillic" \
          sh -c 'cd "$1" && enca -L ru enca-ascii.txt enca-ru-koi8r.txt 2>&1' _ "$D"

# ------------------------------------------------------------- error paths ---
: > "$tmp/empty.txt"
check_rc  enca/empty-rc 1 sh -c 'enca -L ru "$1" >/dev/null 2>&1' _ "$tmp/empty.txt"
check_out enca/empty     "Unrecognized encoding"        sh -c 'enca -L ru "$1" 2>&1 || true'    _ "$tmp/empty.txt"
check_out enca/empty-why "Failure reason: Sample is empty." sh -c 'enca -L ru -d "$1" 2>&1 || true' _ "$tmp/empty.txt"

check_rc  enca/nofile-rc 2 sh -c 'enca -L ru /nonexistent-enca-file >/dev/null 2>&1'
check_out enca/nofile     "No such file or directory" sh -c 'enca -L ru /nonexistent-enca-file 2>&1 || true'
check_rc  enca/badlang-rc 2 sh -c 'enca -L zzz "$1" >/dev/null 2>&1' _ "$D/enca-ascii.txt"
check_rc  enca/badopt-rc  2 sh -c 'enca -L ru --no-such-option "$1" >/dev/null 2>&1' _ "$D/enca-ascii.txt"
check_out enca/badopt      "Unknown option" sh -c 'enca -L ru --no-such-option "$1" 2>&1 || true' _ "$D/enca-ascii.txt"
check_rc  enca/badtarget-rc 1 sh -c 'enca -L ru -x NOSUCHCS "$1" >/dev/null 2>&1' _ "$D/enca-ascii.txt"
check_out enca/badtarget "no converter is able/allowed to perform conversion" \
          sh -c 'enca -L ru -x NOSUCHCS "$1" 2>&1 || true' _ "$D/enca-ascii.txt"

# -------------------------------------------------------------- conversion ---
# The strong form: enca's conversion is deterministic, so a converted file must
# be byte-identical to the fixture already in the target charset. cmp, not a
# content grep -- the native control confirmed each of these identical.
#
# The three 8-bit cases are the BRINTOS-603 reds; the two that follow them need
# no converter at all and must stay green. The detail string carries enca's own
# message so a NEW failure is told apart from the known one.
conv() {                       # conv <case> <src fixture> <target cs> <expected fixture>
  cp "$D/$2" "$tmp/c.txt" || { bst_notok "enca/$1" "fixture $2 unreadable"; return; }
  out=$(enca -L ru -x "$3" "$tmp/c.txt" 2>&1); rc=$?
  if [ "$rc" -ne 0 ]; then
    bst_notok "enca/$1" "rc=$rc $(_bst_1line "$out")"
  elif cmp -s "$tmp/c.txt" "$D/$4"; then
    bst_ok "enca/$1"
  else
    bst_notok "enca/$1" "converted output differs from $4"
  fi
}
conv convert-koi8r-utf8  enca-ru-koi8r.txt  UTF-8  enca-ru-utf8.txt
conv convert-cp1251-utf8 enca-ru-cp1251.txt UTF-8  enca-ru-utf8.txt
conv convert-utf8-koi8r  enca-ru-utf8.txt   KOI8-R enca-ru-koi8r.txt

# Both endpoints are converters glibc compiles into libc, so this one is
# reachable with no gconv module at all -- it is what separates "iconv is
# limited here" from "iconv is broken here".
conv convert-ascii-utf8  enca-ascii.txt     UTF-8  enca-ascii.txt
conv convert-utf8-utf8   enca-ru-utf8.txt   UTF-8  enca-ru-utf8.txt

# ------------------------------------------------------------------ enconv ---
# enconv is the same binary under a second name (a symlink), and argv[0] is what
# flips it from guess mode to convert mode.
check enca/enconv-present sh -c 'command -v enconv >/dev/null'
check_out enca/enconv-help "enconv [-L LANGUAGE]" sh -c 'enconv --help 2>&1'
cp "$D/enca-ru-koi8r.txt" "$tmp/e.txt"
out=$(enconv -L ru -x UTF-8 "$tmp/e.txt" 2>&1); rc=$?
if [ "$rc" -ne 0 ]; then bst_notok enca/enconv-convert "rc=$rc $(_bst_1line "$out")"
elif cmp -s "$tmp/e.txt" "$D/enca-ru-utf8.txt"; then bst_ok enca/enconv-convert
else bst_notok enca/enconv-convert "converted output differs from enca-ru-utf8.txt"; fi

# --------------------------------------------------------------- packaging ---
check enca/installed-enca   test -x /usr/bin/enca
check enca/installed-enconv test -L /usr/bin/enconv
# The distro builds no shared objects, so a stray libenca.so here would mean the
# blanket --disable-shared was lost for this recipe.
check enca/no-shared-lib sh -c '! ls /usr/lib/libenca.so* >/dev/null 2>&1'

rm -rf "$tmp"