. "$BST_LIB"

# BST-TIMEOUT: 300   # one probe run, then 155 extractions that fork nothing at
                     # all (see xp_line). Measured at 45 s end to end; the bound
                     # is headroom for a loaded host, not a budget being spent.

# libXpresent 1.0.1 -- the Present extension client library (meta-oe
# recipes-graphics/xorg-lib). See libxpresent_%.bbappend for what the port adds
# and why.
#
# ---------------------------------------------------------------------------
# ALMOST NONE OF THIS LIBRARY CAN REACH A SERVER HERE, AND THAT IS THE POINT
#
# libXpresent is the client half of Present -- the modern X path for handing a
# finished pixmap to the server for display at a chosen frame, replacing the
# copy-and-hope of the older extensions. Every operation it exposes is a round
# trip: XPresentFindDisplay(dpy), then a request on a wire. THIS IMAGE HAS NO X
# SERVER, so no display opens (section 12) and not one of those requests can
# complete -- not untested today, unreachable by construction.
#
# A shell suite is therefore left saying that an archive and a header arrived,
# which is a statement about tar rather than about a cross compiler. So the port
# builds the consumer libXpresent does not ship, and this suite drives what
# genuinely stands alone.
#
# PRESENT IS THE EXTENSION WHERE 64-BIT COUNTERS MEET A 32-BIT TARGET HEAD ON,
# and that is the half this suite can assert absolutely. Serials, target MSCs
# and UST timestamps are 64-bit quantities fixed by the protocol --
# XPresentPixmap takes three of them, PresentCompleteNotify carries two back --
# and on wasm32 those are `long long' rather than `long', aligned to 8 inside an
# otherwise ILP32 struct. A layout that came out wrong here would be complete,
# installed and silently truncating, and no amount of "the package built" would
# see it. Section 9 is that assertion, member by member.
#
# WHY THE HONEST-FAILURE ARM IS SAFE, AND WHY IT DISCRIMINATES.
#
# 57-libxext.test's probe refuses to pass an invented Display into an entry
# point, and is right to: libXext's find_display misses, falls into
# XextAddDisplay and dereferences. libXpresent is the libXfixes shape instead.
# It keeps its own list -- `XPresentExtInfo XPresentExtensionInfo;', external
# linkage, src/Xpresent.c -- and XPresentExtFindDisplay's fast path COMPARES the
# pointer without loading through it. Seeded with entries whose codes are NULL,
# XPresentFindDisplay returns on the cache hit or on the list walk and never
# reaches XInitExtension.
#
# Every entry point then runs its real prologue out of libXpresent.a and takes
# the arm XPresentCheckExtension defines for a server without Present -- which
# is the state this image is permanently in. The arm differs per family:
#
#   0      XPresentQueryVersion (Status), XPresentSelectInput (XID),
#          XPresentQueryCapabilities (uint32_t)
#   void   XPresentPixmap, XPresentNotifyMSC, XPresentFreeInput -- measured by
#          RETURNING AT ALL, which is not weak: a build that stopped
#          short-circuiting would run on into LockDisplay(token) and fault
#   False  XPresentQueryExtension, which asks XPresentHasExtension itself and
#          leaves ALL THREE out-parameters untouched
#
# The second seeded entry is keyed on NULL, so the ticket's NULL-Display error
# case is measured as the documented refusal rather than as the undefined
# behaviour a raw NULL into an unseeded library would be. Section 8 says so in
# its own comment; it is a real limit of the fixture and is stated, not hidden.
#
# TWO THINGS THE TICKET ASKED FOR CANNOT BE ASSERTED AS WRITTEN, and this suite
# asserts the true thing instead rather than inventing an answer:
#
#   - THERE IS NO SONAME. This distro is static-only: libtool produces
#     libXpresent.a and no libXpresent.so.1, and ${PN} -- whose FILES is
#     ${libdir}/libXpresent.so.* -- is never written at all. Section 1 asserts
#     the archive AND the absence of the shared object, which is the fact.
#   - THE HEADER'S VERSION MACROS ARE NOT 1.0.1. PRESENT_MAJOR and PRESENT_MINOR
#     come from xorgproto's presenttokens.h and are the PROTOCOL version, 1.3;
#     PRESENT_REVISION is Xpresent.h's own 0; PRESENT_VERSION is therefore
#     10300. 1.0.1 is the LIBRARY version and is asserted where it actually
#     lives -- pkg-config --modversion, and the probe's own -DLIBXPRESENT_PV.
#     Keeping the two apart is what makes section 6 evidence: the archive's
#     answer against the sysroot's headers says whether libXpresent.a and the
#     xorgproto on this image came from the same place.
#
# AND ONE EVENT IS NOT HERE AT ALL. XPresentRedirectNotifyEvent, the
# PresentRedirectNotify event code and PresentRedirectNotifyMask all sit behind
# `#if PRESENT_FUTURE_VERSION', and nothing in xorgproto, in libXpresent or in
# this distro ever defines that macro. Asserting its offsets is impossible;
# section 9 asserts that it is compiled out and that the compatibility
# PRESENT_REDIRECT_NOTIFY_MASK is therefore 0, which is what a consumer needs to
# know.
#
# THE SECTIONS
#   1  the payload                 headers, archive, no .so, probe
#   2  the .pc                     how a consumer finds any of it
#   3  the packaging               dpkg's own file lists
#   4  the probe runs, to its end
#   5  the trap did not fire
#   6  the version                 protocol out of the archive, library out of PV
#   7  the extension list          the seeded cache and its walk
#   8  the honest-failure arm      including NULL display and a zero notify list
#   9  the layouts                 sizeof, offsetof, every 64-bit field
#  10  the protocol constants
#  11  the entry-point census
#  12  the display that is not there
#
# A build where 10 passes and 8 does not has a fault in libXpresent.a rather
# than in the headers -- 10 is compiled on this image, 8 came out of the
# archive. One where 9 goes red on a single offset has a struct that changed
# shape; one where it goes red on a width has a 64-bit counter that stopped
# being 64 bits, which is the specific regression this port exists to notice.
# ---------------------------------------------------------------------------

# A root prefix, empty on the target and set only when this suite is validated
# NATIVELY under dash before it ships. The runner never sets it, so on the image
# every path below is exactly the absolute one.
R="${BST_LIBXPRESENT_ROOT:-}"

INC=$R/usr/include/X11/extensions
LIB=$R/usr/lib
PCD=$R/usr/lib/pkgconfig
DPKG=$R/var/lib/dpkg/info
XT=$R/usr/libexec/libxpresent-tests

work=${TMPDIR:-/tmp}/bst-libxpresent.$$
rm -rf "$work"; mkdir -p "$work"

hasfile() {
  if [ -s "$2" ]; then bst_ok "$1"; else bst_notok "$1" "missing or empty: $2"; fi
}

# xp_line <id> <unique-prefix> <expected-remainder>
#
# A red reports WHAT THE PROBE ACTUALLY SAID for that case instead of merely
# "absent". Every prefix is unique by construction: each is a section tag plus a
# case name, and the case name is followed by a space that ends it.
#
# THE EXTRACTION FORKS NOTHING, AND THAT IS WHY THIS SUITE RUNS AT ALL.
# This file calls xp_line 155 times. Written the obvious way -- `$(grep -F ... |
# head -1)' -- each call costs three spawns, and guest memory is not returned on
# process exit (brintos/linux#21, brintos/hardwarejs#20), so the ~465 spawns grew
# the HOST guest process at ~24 GiB/min to a peak of 18.5 GiB. That is past
# brintos_local_verify.py's runaway bound, so the guard ended the guest and
# quarantined the suite with ZERO cases -- which reads exactly like a suite that
# hung at case 0, and is how this port was dropped from batch-2471 twice while
# passing every one of its 203 cases. Measured, not inferred: with the greps the
# same boot peaks at 18500 MiB, and the probe alone peaks at 347 MiB.
#
# So the probe's output is read ONCE into a shell variable by the builtin `read'
# below, and every lookup here is parameter expansion. The match is anchored at
# a line start -- XP_OUT carries a leading newline so that the first line is
# reachable by the same pattern as the rest -- which is also stricter than the
# grep -F it replaces, since that would have matched a prefix mid-line.
NL='
'
xp_line() {
  case "$XP_OUT" in
    *"$NL$2"*) _r=${XP_OUT#*"$NL$2"}; _l="$2${_r%%"$NL"*}" ;;
    *)         _l= ;;
  esac
  check_eq "$1" "$2$3" "$_l"
}

# ---------------------------------------------------------------------------
# 0. THE CANARY, which is not an assertion about libXpresent at all.
#
# The guest console drops a suite's FIRST BST result line before the host parser
# sees it -- brintos/linux#96. The harness computes its pass count as
# total-minus-failures, so THE DROPPED LINE IS COUNTED AS A PASS, and a suite
# whose first case is a genuine `not ok' loses the one line that mattered. So
# the first line out of this file asserts something already true.
# Remove this case when #96 is fixed.
bst_ok libxpresent/suite-canary

# ---------------------------------------------------------------------------
# 1. THE PAYLOAD.
#
# libXpresent installs exactly one public header, and it is unusable alone:
# Xpresent.h includes presenttokens.h for the protocol constants and the
# library's own PRESENT_VERSION formula is built out of them. presentproto.h is
# what the wire structures live in. Both are xorgproto's rather than this
# recipe's, and an image carrying one and not the other fails a consumer's
# do_compile rather than anything on this image -- which is exactly why they are
# asserted here.

check_rc libxpresent/dir-extensions 0 test -d "$INC"

hasfile libxpresent/hdr-Xpresent      "$INC/Xpresent.h"
hasfile libxpresent/hdr-presenttokens "$INC/presenttokens.h"
hasfile libxpresent/hdr-presentproto  "$INC/presentproto.h"

# Xpresent.h also takes XserverRegion from Xfixes.h, RRCrtc from Xrandr.h and
# XSyncFence from sync.h -- the three DEPENDS this port has. A consumer that
# includes Xpresent.h needs all three present, so their absence is this port's
# problem to report even though none is this recipe's file.
hasfile libxpresent/hdr-dep-Xfixes "$INC/Xfixes.h"
hasfile libxpresent/hdr-dep-Xrandr "$INC/Xrandr.h"
hasfile libxpresent/hdr-dep-sync   "$INC/sync.h"

# THE ARCHIVE IS THE WHOLE DELIVERY, AND THE ABSENCE OF THE .so IS THE OTHER
# HALF OF THAT STATEMENT. This distro appends --disable-shared to every
# class-target configure, so a libXpresent.so.1 appearing here would mean the
# recipe had started producing something nothing on this image can load -- and
# that ${PN}, which is empty and never written, had started having content.
# This is also the case that stands in for the ticket's "assert the shipped
# soname": there is no soname on this arm, and saying so is the true assertion.
hasfile  libxpresent/archive   "$LIB/libXpresent.a"
check_rc libxpresent/no-shared 1 test -e "$LIB/libXpresent.so.1"
check_rc libxpresent/no-shared-link 1 test -e "$LIB/libXpresent.so"

hasfile libxpresent/probe-shipped "$XT/libxpresent-probe"

# Asserted on contents rather than size: a truncated install would still be
# non-empty. The formula is what section 6 measures the archive against, and
# XPresentQueryExtension is the entry point every consumer starts at.
check libxpresent/hdr-has-version-formula \
  grep -q 'PRESENT_VERSION' "$INC/Xpresent.h"
check libxpresent/hdr-has-queryextension \
  grep -q 'XPresentQueryExtension' "$INC/Xpresent.h"

# The header's own statement that the redirect event is not compiled in. Read
# here as text, and measured by the probe as a preprocessor fact in section 9;
# the two together say the header ships the guard AND that the guard is closed.
check libxpresent/hdr-redirect-is-guarded \
  grep -q 'PRESENT_FUTURE_VERSION' "$INC/Xpresent.h"

# ---------------------------------------------------------------------------
# 2. THE .pc -- how a consumer finds any of the above.

hasfile libxpresent/pc-present "$PCD/xpresent.pc"

# THE .pc IS ASSERTED ON ITS CONTENTS FIRST, AND THROUGH pkg-config SECOND.
# Read directly, these lines are true whatever else is installed; asked through
# pkg-config they are true only if the packages xpresent.pc NAMES are also on
# the image, which is a fact about the image rather than about this port. Both
# are worth having, and separating them is what makes a red say which.
#
# `Requires: xproto presentproto' -- both from xorgproto-dev, so this is the
# dependency edge Yocto turns into an RDEPENDS on libxpresent-dev (poky's
# oe.package.process_pkgconfig reads `Requires' and NOT `Requires.private').
check libxpresent/pc-requires-presentproto \
  grep -q '^Requires:.*presentproto' "$PCD/xpresent.pc"
check libxpresent/pc-requires-xproto \
  grep -q '^Requires:.*xproto' "$PCD/xpresent.pc"

# x11, xext, xfixes and xrandr are PRIVATE requires, and correctly so: a
# consumer of the shared library would not need them, but this distro is
# static-only, so every consumer here does. x11 is the one libXpresent.a
# actually calls into -- XInitExtension, _XReply, LockDisplay; the other three
# supply the TYPES Xpresent.h takes (XserverRegion, RRCrtc, XSyncFence), which
# is why the probe's own link line names libX11 and not them.
check libxpresent/pc-requires-private-x11 \
  grep -q '^Requires\.private:.*x11' "$PCD/xpresent.pc"
check libxpresent/pc-requires-private-xfixes \
  grep -q '^Requires\.private:.*xfixes' "$PCD/xpresent.pc"
check libxpresent/pc-requires-private-xrandr \
  grep -q '^Requires\.private:.*xrandr' "$PCD/xpresent.pc"

check libxpresent/pc-libs-name grep -q '\-lXpresent' "$PCD/xpresent.pc"

# NOT `require pkg-config'. require ENDS THE FILE when the tool is absent, and
# everything this suite exists for comes after this section. Gating the probe on
# a query tool would mean an image without pkg-config reported a clean skip for
# the library it never exercised. So these cases skip individually and the file
# runs on.
if command -v pkg-config >/dev/null 2>&1; then
  bst_ok libxpresent/pkg-config-present

  # The graph queries below resolve xpresent.pc's requires, so they answer about
  # xorgproto-dev and libx11-dev as much as about this port. Gated on both being
  # present, and gated by NAME, so that an image missing one reports which --
  # rather than reporting `xpresent.pc is broken', which it would not be.
  if [ -f "$PCD/presentproto.pc" ] && [ -f "$PCD/x11.pc" ]; then
    bst_ok libxpresent/pc-deps-present

    check libxpresent/pc-exists   pkg-config --exists xpresent
    check libxpresent/pc-validate pkg-config --validate xpresent

    # THE LIBRARY VERSION, and the only place on this image where 1.0.1 is a
    # correct expectation. The header's macros are the PROTOCOL version; see
    # this file's opening comment and section 6.
    check_eq libxpresent/pc-modversion 1.0.1 \
      "$(pkg-config --modversion xpresent 2>/dev/null)"

    # PKG_CONFIG_ALLOW_SYSTEM_LIBS is required for the same reason
    # 57-xtrans.test records for its --cflags case: pkg-config SUPPRESSES a -L
    # naming a directory already on the linker's default path, so the bare query
    # answers usefully only for the -l part. Asked for the flags it COMPUTED,
    # the library name has to be there.
    check_out libxpresent/pc-libs -lXpresent \
      env PKG_CONFIG_ALLOW_SYSTEM_LIBS=1 pkg-config --libs xpresent
  else
    bst_skip libxpresent/pc-deps-present \
      "xpresent.pc requires presentproto and (privately) x11; one is missing on this image"
  fi
else
  bst_skip libxpresent/pkg-config-present "pkg-config not installed on this image"
fi

# ---------------------------------------------------------------------------
# 3. THE PACKAGING -- dpkg's own file lists.
#
# Section 1 asserted that the files are THERE; this one asserts they arrived as
# the packages the recipe declares, which is the ticket's "the .deb installs on
# target with dpkg and the files land where the packaging says". A file that
# reached the image by some other route satisfies section 1 and does not satisfy
# this.
#
# Three packages, and the split is the port's own decision: -dev has the header
# and the .pc, -staticdev the archive (this distro's whole delivery), -tests the
# probe in its own package so an image can take the library without it.

nl='
'
haspkgline() {
  # $1 case id, $2 path, $3 the list contents
  case "$nl$3$nl" in
    *"$nl$2$nl"*) bst_ok "$1" ;;
    *)            bst_notok "$1" "not listed: $2" ;;
  esac
}

for pkg in libxpresent-dev libxpresent-staticdev libxpresent-tests; do
  if [ -s "$DPKG/$pkg.list" ]; then
    bst_ok "libxpresent/pkg-list-$pkg"
  else
    bst_notok "libxpresent/pkg-list-$pkg" \
      "no $DPKG/$pkg.list -- the package did not install through dpkg"
  fi
done

devlist=$(cat "$DPKG/libxpresent-dev.list" 2>/dev/null)
haspkgline libxpresent/pkg-dev-header /usr/include/X11/extensions/Xpresent.h "$devlist"
haspkgline libxpresent/pkg-dev-pc     /usr/lib/pkgconfig/xpresent.pc         "$devlist"

stalist=$(cat "$DPKG/libxpresent-staticdev.list" 2>/dev/null)
haspkgline libxpresent/pkg-static-archive /usr/lib/libXpresent.a "$stalist"

tstlist=$(cat "$DPKG/libxpresent-tests.list" 2>/dev/null)
haspkgline libxpresent/pkg-tests-probe \
  /usr/libexec/libxpresent-tests/libxpresent-probe "$tstlist"

# ${PN} IS NEVER WRITTEN ON THIS ARM, so there is no libxpresent.list to find.
# Asserting its absence is the packaging half of section 1's no-shared case: a
# list here would mean a shared object had been produced and packaged.
check_rc libxpresent/pkg-no-runtime 1 test -e "$DPKG/libxpresent.list"

# ---------------------------------------------------------------------------
# 4. THE PROBE RUNS, AND RUNS TO ITS END.
#
# do_install ships it 0755, but the rootfs is read-only at browser runtime and
# the exec bit is not reliably readable there either, so it is copied to /tmp --
# a mode=1777 tmpfs with no noexec -- and given the bit where the copy lands.

probe=$work/libxpresent-probe
if cp "$XT/libxpresent-probe" "$probe" 2>/dev/null && chmod +x "$probe" 2>/dev/null; then
  "$probe" >"$work/out" 2>"$work/err"
  rc=$?

  # Read once, with builtins only -- see xp_line's header for what the forked
  # spelling of this cost. The `|| [ -n "$_l" ]' keeps a last line that lacks a
  # trailing newline, and the leading newline is what anchors xp_line's match.
  XP_OUT=
  while IFS= read -r _l || [ -n "$_l" ]; do
    XP_OUT="$XP_OUT$NL$_l"
  done < "$work/out"

  # The probe exits 0 whatever it measures -- every expectation lives in this
  # file, not in the binary -- so a non-zero status here means it did not RUN: a
  # trap, a missing import, an exec refusal. On this arm that is a real and
  # common failure and is what this case is for.
  #
  # It is also the case that would catch the seeded-list section going wrong. If
  # XPresentCheckExtension had stopped short-circuiting, section 8 would run on
  # into LockDisplay(token) and fault, and the probe would die HERE rather than
  # reporting a wrong sentinel.
  if [ "$rc" -eq 0 ]; then bst_ok libxpresent/probe-runs
  else bst_notok libxpresent/probe-runs "rc=$rc $(_bst_1line "$(cat "$work/err")")"; fi

  total=$(sed -n 's/^XPRESENT-TOTAL //p' "$work/out")
  if [ -n "$total" ]; then bst_ok libxpresent/probe-complete
  else bst_notok libxpresent/probe-complete \
    "no XPRESENT-TOTAL line: $(_bst_1line "$(cat "$work/out")")"; fi

  # The measurement count, bounded from below. Adding a fixture raises it; a
  # probe that got THINNER while still terminating cleanly is what this catches,
  # and nothing else here would -- every case below greps for what it expects,
  # so a line that stopped being printed is indistinguishable from one whose
  # grep simply found nothing.
  if [ -n "$total" ] && [ "$total" -ge 160 ]; then bst_ok libxpresent/probe-coverage
  else bst_notok libxpresent/probe-coverage \
    "reported ${total:-0} measurements, expected not less than 160"; fi

  # The version the archive was built from, reported by the binary rather than
  # read off a path, so a stale probe left behind by an earlier build cannot
  # pass for this one.
  xp_line libxpresent/probe-version "XPRESENT-VERSION " "1.0.1"

  # -------------------------------------------------------------------------
  # 5. THE TRAP DID NOT FIRE.
  #
  # The probe DEFINES XQueryExtension, so that a seeding failure lands there
  # instead of in XInitExtension's dereference of the token. Reaching it while
  # armed prints XPRESENT-ABORT and _exit(0)s -- which section 4's
  # probe-complete would already have caught, but only as "short", saying
  # nothing about why. These two cases say why, and they are the reason this
  # probe is safe to run on an image other tickets share.
  #
  # stub-calls is asserted at 0 rather than merely "no ABORT line": a call that
  # arrived after the trap was disarmed (section 12 disarms it before
  # XOpenDisplay) would be Xlib's own, would be harmless, and would still mean
  # something about this image's Xlib had changed.
  check_rc libxpresent/trap-not-fired 1 grep -q '^XPRESENT-ABORT ' "$work/out"
  xp_line libxpresent/trap-stub-unused "XPRESENT-FAIL stub-calls " "0"

  # -------------------------------------------------------------------------
  # 6. THE VERSION -- two independent quantities, deliberately kept apart.
  #
  # PRESENT_MAJOR and PRESENT_MINOR are xorgproto's and are the PROTOCOL
  # version; PRESENT_REVISION is Xpresent.h's own; PRESENT_VERSION is the
  # formula over the three. XPresentVersion() returns that formula as compiled
  # INTO libXpresent.a, so `agree' compares the archive with the sysroot: a
  # mismatch means the .a and the xorgproto here came from different places, and
  # nothing else in this suite would notice.
  #
  # 1.0.1 IS NOT ANY OF THESE. It is the library's own version, and it is
  # asserted in section 2 through pkg-config and here through the -DLIBXPRESENT_PV
  # the recipe hands the probe. A suite that expected 1.0.1 out of the header
  # macros -- which is the obvious reading, and the one the ticket carried --
  # would be red against a correct image.

  xp_line libxpresent/ver-pv        "XPRESENT-VER pv "             "1.0.1"
  xp_line libxpresent/ver-major     "XPRESENT-VER proto-major "    "1"
  xp_line libxpresent/ver-minor     "XPRESENT-VER proto-minor "    "3"
  xp_line libxpresent/ver-revision  "XPRESENT-VER proto-revision " "0"

  # (PRESENT_MAJOR * 10000) + (PRESENT_MINOR * 100) + PRESENT_REVISION.
  xp_line libxpresent/ver-header "XPRESENT-VER header-version " "10300"
  xp_line libxpresent/ver-lib    "XPRESENT-VER lib-version "    "10300"
  xp_line libxpresent/ver-agree  "XPRESENT-VER agree "          "yes"

  # The extension name the library puts on the wire, out of the archive's own
  # `char XPresentExtensionName[] = PRESENT_NAME;'.
  xp_line libxpresent/ver-ext-name "XPRESENT-VER ext-name " "Present"

  # -------------------------------------------------------------------------
  # 7. THE EXTENSION LIST (src/Xpresent.c).
  #
  # libXpresent does NOT use extutil's list -- it keeps its own, because it has
  # to cache the negotiated version alongside the codes. The cache is one entry
  # (extinfo->cur) and a fault in it would surface only with two displays open,
  # which on this image is never, so it would ship silently and break the first
  # real X client.
  #
  # The Display* is an opaque token the probe never loads through; see the
  # probe's header for why that is legitimate for these paths and for nothing
  # else in libXpresent.

  xp_line libxpresent/list-ndisplays "XPRESENT-LIST seeded-ndisplays " "2"

  # The fast path: extinfo->cur matches by pointer, so XPresentFindDisplay
  # returns without reaching XPresentExtAddDisplay and therefore without
  # XInitExtension.
  xp_line libxpresent/list-find-cached "XPRESENT-LIST find-cached " "seeded"

  # The walk, with the cache cleared. It must find the same entry AND re-promote
  # it -- the promotion is what makes the next call take the fast path, and a
  # walk that found the entry without promoting it would be a cache that never
  # warmed.
  xp_line libxpresent/list-find-walk   "XPRESENT-LIST find-walk "       "seeded"
  xp_line libxpresent/list-repromotes  "XPRESENT-LIST walk-repromotes " "seeded"

  # The second entry, keyed on NULL, reached only by the walk. It is the fixture
  # section 8's NULL-display cases run through.
  xp_line libxpresent/list-find-null   "XPRESENT-LIST find-null " "seeded-null"

  # The guard every entry point branches on, `((i) && ((i)->codes))'. It is
  # false here because codes is NULL, which is exactly the state a server
  # without Present leaves the entry in -- and the state section 8 measures the
  # consequences of. It is also false for a NULL info.
  xp_line libxpresent/list-has-ext-seeded "XPRESENT-LIST has-ext-seeded "    "0"
  xp_line libxpresent/list-has-ext-null   "XPRESENT-LIST has-ext-null-info " "0"

  # A display that is not in the list, walked rather than looked up: looking it
  # up would miss and fall into XPresentExtAddDisplay, which is the path the
  # trap exists for.
  xp_line libxpresent/list-absent "XPRESENT-LIST absent-in-list " "0"

  # -1/-1 is what XPresentExtAddDisplay leaves when the server has no Present,
  # and it is what XPresentQueryVersion would hand a client that ignored the
  # return status.
  xp_line libxpresent/list-unknown-version \
    "XPRESENT-LIST unknown-version " "major=-1 minor=-1"

  # -------------------------------------------------------------------------
  # 8. THE HONEST-FAILURE ARM.
  #
  # One case per documented arm. This is the arm this image is permanently in,
  # so these are not error paths in the abstract -- they are what every Present
  # call on this image does.

  # XPresentQueryExtension asks XPresentHasExtension itself and returns False,
  # leaving all three out-parameters UNTOUCHED at the probe's sentinels. That
  # second case is the part a client actually depends on: a build that wrote
  # zeroes into them before refusing would look identical here without it.
  xp_line libxpresent/fail-query-extension "XPRESENT-FAIL query-extension " "0"
  xp_line libxpresent/fail-query-extension-untouched \
    "XPRESENT-FAIL query-extension-untouched " "op=-1 ev=-1 err=-1"

  # Status 0, via XPresentCheckExtension(dpy, info, 0). The out-parameters are
  # written only AFTER the guard, so they too stay at the sentinels.
  xp_line libxpresent/fail-query-version "XPRESENT-FAIL query-version " "0"
  xp_line libxpresent/fail-query-version-untouched \
    "XPRESENT-FAIL query-version-untouched " "major=-1 minor=-1"

  # XID 0 rather than an allocated id: a refused SelectInput must not burn one,
  # because the client has no way to free an id it was never told about.
  xp_line libxpresent/fail-select-input "XPRESENT-FAIL select-input " "0"

  # uint32_t 0 -- no capabilities, which is the truthful answer for a server
  # that does not have the extension.
  xp_line libxpresent/fail-query-capabilities \
    "XPRESENT-FAIL query-capabilities " "0"

  # THE VOID ARM, MEASURED BY REACHING THE NEXT LINE. Not weak: a build in which
  # XPresentSimpleCheckExtension had stopped short-circuiting would carry on
  # into LockDisplay(token) and fault, and section 4's probe-runs would go red
  # instead.
  xp_line libxpresent/fail-notify-msc "XPRESENT-FAIL notify-msc " "reached"
  xp_line libxpresent/fail-free-input "XPRESENT-FAIL free-input " "reached"
  xp_line libxpresent/fail-pixmap     "XPRESENT-FAIL pixmap "     "reached"

  # The ticket's zero-count notify list. The guard fires before `notifies' is
  # read at all, so a NULL pointer with count 0 is refused rather than walked --
  # which is what makes the case a statement about ORDERING inside XPresentPixmap
  # rather than about malloc.
  xp_line libxpresent/fail-pixmap-zero-notifies \
    "XPRESENT-FAIL pixmap-zero-notifies " "reached"

  # --- the NULL Display ----------------------------------------------------
  #
  # THE LIMIT OF THIS FIXTURE IS STATED RATHER THAN HIDDEN. A raw NULL into an
  # unseeded libXpresent is undefined behaviour, not a documented failure: the
  # lookup misses, falls into XPresentExtAddDisplay and XInitExtension
  # dereferences. The probe seeds a second entry keyed on NULL, so what these
  # cases measure is the DOCUMENTED refusal with a NULL Display -- the same arm
  # the token takes, reached by the same comparison. That is the useful half:
  # it says the refusal does not depend on the Display's identity.

  xp_line libxpresent/null-query-extension "XPRESENT-NULL query-extension " "0"
  xp_line libxpresent/null-query-extension-untouched \
    "XPRESENT-NULL query-extension-untouched " "op=-1 ev=-1 err=-1"
  xp_line libxpresent/null-query-capabilities \
    "XPRESENT-NULL query-capabilities " "0"
  xp_line libxpresent/null-select-input "XPRESENT-NULL select-input " "0"
  xp_line libxpresent/null-notify-msc   "XPRESENT-NULL notify-msc "   "reached"

  # -------------------------------------------------------------------------
  # 9. THE LAYOUTS -- what this port exists for.
  #
  # THE NUMBERS BELOW ARE THIS ARM'S, AND THAT IS THE POINT. wasm32 is ILP32
  # with a 64-bit `long long' that aligns to 8: int, Bool, a pointer and the
  # XID/Window/Pixmap/XSyncFence family are all 4 bytes, while uint64_t is 8 and
  # drags the struct's own alignment to 8 with it. On the LP64 build host every
  # one of these would be different. They are stated as the target's answers
  # because that is the only machine whose answer matters to a client running
  # here, and they were derived from the compiler rather than from arithmetic:
  # the same headers, compiled by the same cross compiler this image was built
  # with (BRINTOS-2495).

  # --- the arm, first: every offset below follows from these ----------------
  xp_line libxpresent/arch-sizeof-long     "XPRESENT-ARCH sizeof-long "     "4"
  xp_line libxpresent/arch-sizeof-longlong "XPRESENT-ARCH sizeof-longlong " "8"
  xp_line libxpresent/arch-sizeof-ptr      "XPRESENT-ARCH sizeof-ptr "      "4"
  xp_line libxpresent/arch-sizeof-int      "XPRESENT-ARCH sizeof-int "      "4"
  xp_line libxpresent/arch-sizeof-uint64   "XPRESENT-ARCH sizeof-uint64 "   "8"

  # THE ONE THAT IS NOT A RESTATEMENT OF ILP32. uint64_t aligns to 8 here, not
  # to 4 -- which is what puts four bytes of padding between serial_number and
  # ust below, and it is the single fact that makes Present's layout on this arm
  # different from the obvious one. A toolchain that aligned it to 4 (as 32-bit
  # x86 does) would give a 56-byte complete event and every offset from ust on
  # would move.
  xp_line libxpresent/arch-align-uint64 "XPRESENT-ARCH align-uint64 " "8"
  xp_line libxpresent/arch-align-uint32 "XPRESENT-ARCH align-uint32 " "4"
  xp_line libxpresent/arch-align-ptr    "XPRESENT-ARCH align-ptr "    "4"

  # XID is `unsigned long', so it is 4 here and 8 on the host. Window, Pixmap
  # and XSyncFence are all XID.
  xp_line libxpresent/arch-sizeof-xid    "XPRESENT-ARCH sizeof-XID "    "4"
  xp_line libxpresent/arch-sizeof-window "XPRESENT-ARCH sizeof-Window " "4"
  xp_line libxpresent/arch-sizeof-bool   "XPRESENT-ARCH sizeof-Bool "   "4"
  xp_line libxpresent/arch-sizeof-xevent "XPRESENT-ARCH sizeof-XEvent " "96"

  # --- XPresentNotify: the element type of XPresentPixmap's list ------------
  #
  # Eight bytes of two 32-bit fields, and it is the one Present structure whose
  # size is the SAME here as on the wire (sz_xPresentNotify is 8 too, section
  # 10) -- because both its members are 32-bit by protocol. XPresentPixmap
  # passes the array straight to Data32, so a client's array and the wire have
  # to agree exactly, and this pair of cases is that agreement.
  xp_line libxpresent/size-notify     "XPRESENT-SIZE XPresentNotify "        "8"
  xp_line libxpresent/off-notify-window "XPRESENT-OFF XPresentNotify window " "0"
  xp_line libxpresent/off-notify-serial "XPRESENT-OFF XPresentNotify serial " "4"
  xp_line libxpresent/fsz-notify-window "XPRESENT-FSZ XPresentNotify window " "4"
  xp_line libxpresent/fsz-notify-serial "XPRESENT-FSZ XPresentNotify serial " "4"

  # --- the common event header ---------------------------------------------
  #
  # Six fields shared by every Present event, 24 bytes. A client switching on
  # ->evtype through the generic type reads them at these offsets whichever
  # event arrived, so the three HDR cases below assert they have not drifted
  # apart -- which the per-event offsets alone would not say.
  xp_line libxpresent/size-event      "XPRESENT-SIZE XPresentEvent "          "24"
  xp_line libxpresent/off-ev-type     "XPRESENT-OFF XPresentEvent type "      "0"
  xp_line libxpresent/off-ev-serial   "XPRESENT-OFF XPresentEvent serial "    "4"
  xp_line libxpresent/off-ev-send     "XPRESENT-OFF XPresentEvent send_event " "8"
  xp_line libxpresent/off-ev-display  "XPRESENT-OFF XPresentEvent display "   "12"
  xp_line libxpresent/off-ev-ext      "XPRESENT-OFF XPresentEvent extension " "16"
  xp_line libxpresent/off-ev-evtype   "XPRESENT-OFF XPresentEvent evtype "    "20"

  xp_line libxpresent/hdr-complete  "XPRESENT-HDR complete "  "same"
  xp_line libxpresent/hdr-idle      "XPRESENT-HDR idle "      "same"
  xp_line libxpresent/hdr-configure "XPRESENT-HDR configure " "same"

  # --- PresentCompleteNotify: the event that carries UST and MSC ------------
  #
  # THE SHARPEST ASSERTION IN THIS SUITE. ust and msc are the 64-bit counters
  # the whole extension is organised around -- the timestamp a frame was
  # presented at and the media stream counter it was presented on -- and a
  # client that read either through a 32-bit `unsigned long' would get a value
  # that is right for about 71 minutes (ust, microseconds) and then silently
  # wrong. The offsets say they are where they should be; the FSZ cases say each
  # is eight bytes wide, which is the assertion the offsets cannot make on their
  # own: a field that shrank AND a field after it that grew would leave every
  # offset unchanged.
  #
  # NOTE THE GAP AT 36. serial_number ends at 36 and ust begins at 40, because
  # uint64_t aligns to 8 here. The four bytes between are padding and are NOT a
  # field this suite has failed to name.
  xp_line libxpresent/size-complete "XPRESENT-SIZE XPresentCompleteNotifyEvent " "64"
  xp_line libxpresent/off-c-eid \
    "XPRESENT-OFF XPresentCompleteNotifyEvent eid " "24"
  xp_line libxpresent/off-c-window \
    "XPRESENT-OFF XPresentCompleteNotifyEvent window " "28"
  xp_line libxpresent/off-c-serial \
    "XPRESENT-OFF XPresentCompleteNotifyEvent serial_number " "32"
  xp_line libxpresent/off-c-ust \
    "XPRESENT-OFF XPresentCompleteNotifyEvent ust " "40"
  xp_line libxpresent/off-c-msc \
    "XPRESENT-OFF XPresentCompleteNotifyEvent msc " "48"
  xp_line libxpresent/off-c-kind \
    "XPRESENT-OFF XPresentCompleteNotifyEvent kind " "56"
  xp_line libxpresent/off-c-mode \
    "XPRESENT-OFF XPresentCompleteNotifyEvent mode " "57"

  xp_line libxpresent/fsz-c-ust \
    "XPRESENT-FSZ XPresentCompleteNotifyEvent ust " "8"
  xp_line libxpresent/fsz-c-msc \
    "XPRESENT-FSZ XPresentCompleteNotifyEvent msc " "8"
  xp_line libxpresent/fsz-c-serial \
    "XPRESENT-FSZ XPresentCompleteNotifyEvent serial_number " "4"
  xp_line libxpresent/fsz-c-kind \
    "XPRESENT-FSZ XPresentCompleteNotifyEvent kind " "1"
  xp_line libxpresent/fsz-c-mode \
    "XPRESENT-FSZ XPresentCompleteNotifyEvent mode " "1"

  # --- PresentIdleNotify ----------------------------------------------------
  #
  # No 64-bit field, so it is flat 4-byte members throughout and 44 bytes -- NOT
  # rounded to 8, because nothing in it demands 8-byte alignment. That
  # difference from the complete event above is itself worth asserting: it is
  # what says the 8-byte alignment is coming from the uint64_t members and not
  # from some blanket struct rule.
  xp_line libxpresent/size-idle "XPRESENT-SIZE XPresentIdleNotifyEvent " "44"
  xp_line libxpresent/off-i-eid \
    "XPRESENT-OFF XPresentIdleNotifyEvent eid " "24"
  xp_line libxpresent/off-i-window \
    "XPRESENT-OFF XPresentIdleNotifyEvent window " "28"
  xp_line libxpresent/off-i-serial \
    "XPRESENT-OFF XPresentIdleNotifyEvent serial_number " "32"
  xp_line libxpresent/off-i-pixmap \
    "XPRESENT-OFF XPresentIdleNotifyEvent pixmap " "36"
  xp_line libxpresent/off-i-fence \
    "XPRESENT-OFF XPresentIdleNotifyEvent idle_fence " "40"
  xp_line libxpresent/fsz-i-pixmap \
    "XPRESENT-FSZ XPresentIdleNotifyEvent pixmap " "4"
  xp_line libxpresent/fsz-i-fence \
    "XPRESENT-FSZ XPresentIdleNotifyEvent idle_fence " "4"

  # --- PresentConfigureNotify -----------------------------------------------
  #
  # The widest of the three at 68 bytes, and the one whose members are widened
  # by the CLIENT struct rather than by the protocol: the wire carries x, y,
  # width, height, off_x, off_y, pixmap_width and pixmap_height as 16-bit
  # quantities (section 10: the wire event is 40 bytes) and Xpresent.h declares
  # them int/unsigned. pixmap_flags is `long', so it is 4 here and would be 8 on
  # the host -- the one member of any Present event whose width is the arm's
  # rather than the protocol's, which is why it gets its own FSZ case.
  xp_line libxpresent/size-configure \
    "XPRESENT-SIZE XPresentConfigureNotifyEvent " "68"
  xp_line libxpresent/off-f-eid \
    "XPRESENT-OFF XPresentConfigureNotifyEvent eid " "24"
  xp_line libxpresent/off-f-window \
    "XPRESENT-OFF XPresentConfigureNotifyEvent window " "28"
  xp_line libxpresent/off-f-x \
    "XPRESENT-OFF XPresentConfigureNotifyEvent x " "32"
  xp_line libxpresent/off-f-y \
    "XPRESENT-OFF XPresentConfigureNotifyEvent y " "36"
  xp_line libxpresent/off-f-width \
    "XPRESENT-OFF XPresentConfigureNotifyEvent width " "40"
  xp_line libxpresent/off-f-height \
    "XPRESENT-OFF XPresentConfigureNotifyEvent height " "44"
  xp_line libxpresent/off-f-off-x \
    "XPRESENT-OFF XPresentConfigureNotifyEvent off_x " "48"
  xp_line libxpresent/off-f-off-y \
    "XPRESENT-OFF XPresentConfigureNotifyEvent off_y " "52"
  xp_line libxpresent/off-f-pixmap-width \
    "XPRESENT-OFF XPresentConfigureNotifyEvent pixmap_width " "56"
  xp_line libxpresent/off-f-pixmap-height \
    "XPRESENT-OFF XPresentConfigureNotifyEvent pixmap_height " "60"
  xp_line libxpresent/off-f-pixmap-flags \
    "XPRESENT-OFF XPresentConfigureNotifyEvent pixmap_flags " "64"
  xp_line libxpresent/fsz-f-pixmap-flags \
    "XPRESENT-FSZ XPresentConfigureNotifyEvent pixmap_flags " "4"

  # --- every event fits an XEvent -------------------------------------------
  #
  # Present events are generic events and arrive through XGenericEventCookie,
  # but a client that received one into an XEvent-shaped slot -- which is what
  # XNextEvent writes through -- would be corrupting its own stack if any of
  # them were larger. 96 against 68 is comfortable today; the case exists
  # because a future field would eat that margin silently.
  xp_line libxpresent/fits-complete  "XPRESENT-FIT complete "  "yes"
  xp_line libxpresent/fits-idle      "XPRESENT-FIT idle "      "yes"
  xp_line libxpresent/fits-configure "XPRESENT-FIT configure " "yes"
  xp_line libxpresent/fits-header    "XPRESENT-FIT header "    "yes"

  # --- the event that is not here -------------------------------------------
  #
  # XPresentRedirectNotifyEvent, PresentRedirectNotify and
  # PresentRedirectNotifyMask are all behind `#if PRESENT_FUTURE_VERSION', and
  # nothing defines that macro -- not xorgproto, not libXpresent, not this
  # distro. So the type does not exist on this image and the compatibility
  # PRESENT_REDIRECT_NOTIFY_MASK is 0, which means PresentAllEvents is 7 and not
  # 15 (section 10).
  #
  # ASSERTING THE ABSENCE IS THE HONEST EQUIVALENT OF ASSERTING THE OFFSETS, and
  # it is the case that would notice the day xorgproto turns the macro on: three
  # of these four flip together, PresentAllEvents changes, and a client
  # compiled against the new header would start selecting an event this library
  # cannot decode. A red here is a real change, not a formality.
  xp_line libxpresent/future-macro-undefined \
    "XPRESENT-FUTURE macro " "undefined"
  xp_line libxpresent/future-event-undefined \
    "XPRESENT-FUTURE redirect-event " "undefined"
  xp_line libxpresent/future-mask-undefined \
    "XPRESENT-FUTURE redirect-mask " "undefined"
  xp_line libxpresent/future-compat-mask-zero \
    "XPRESENT-FUTURE redirect-compat-mask " "0"

  # -------------------------------------------------------------------------
  # 10. THE PROTOCOL CONSTANTS.
  #
  # Wire values fixed by the Present specification, compiled on THIS image out
  # of the xorgproto it ships. They are stated here rather than diffed against a
  # reference build because they have an absolute answer -- which is what makes
  # a red readable without a second machine.

  xp_line libxpresent/const-name     "XPRESENT-CONST name "            "Present"
  xp_line libxpresent/const-nreq     "XPRESENT-CONST number-requests " "5"
  xp_line libxpresent/const-nevents  "XPRESENT-CONST number-events "   "0"
  xp_line libxpresent/const-nerrors  "XPRESENT-CONST number-errors "   "0"

  # PresentNumberEvents is 0 and that is not a mistake: Present defines no core
  # events at all. Its three notifications are GENERIC events, delivered through
  # the XGE cookie machinery Xpresent.c registers -- which is why the library
  # calls XESetWireToEventCookie rather than XESetWireToEvent, and why there is
  # no event-base arithmetic in this suite where 57-libxfixes.test has a whole
  # section of it.

  xp_line libxpresent/req-query-version "XPRESENT-REQ query-version " "0"
  xp_line libxpresent/req-pixmap        "XPRESENT-REQ pixmap "        "1"
  xp_line libxpresent/req-notify-msc    "XPRESENT-REQ notify-msc "    "2"
  xp_line libxpresent/req-select-input  "XPRESENT-REQ select-input "  "3"
  xp_line libxpresent/req-query-caps    "XPRESENT-REQ query-capabilities " "4"

  xp_line libxpresent/opt-none       "XPRESENT-OPT none "           "0"
  xp_line libxpresent/opt-async      "XPRESENT-OPT async "          "1"
  xp_line libxpresent/opt-copy       "XPRESENT-OPT copy "           "2"
  xp_line libxpresent/opt-ust        "XPRESENT-OPT ust "            "4"
  xp_line libxpresent/opt-suboptimal "XPRESENT-OPT suboptimal "     "8"
  xp_line libxpresent/opt-may-tear   "XPRESENT-OPT async-may-tear " "16"
  xp_line libxpresent/opt-all        "XPRESENT-OPT all "            "31"
  xp_line libxpresent/opt-all-async  "XPRESENT-OPT all-async "      "17"

  xp_line libxpresent/cap-none      "XPRESENT-CAP none "           "0"
  xp_line libxpresent/cap-async     "XPRESENT-CAP async "          "1"
  xp_line libxpresent/cap-fence     "XPRESENT-CAP fence "          "2"
  xp_line libxpresent/cap-ust       "XPRESENT-CAP ust "            "4"
  xp_line libxpresent/cap-may-tear  "XPRESENT-CAP async-may-tear " "8"
  xp_line libxpresent/cap-all       "XPRESENT-CAP all "            "15"
  xp_line libxpresent/cap-all-async "XPRESENT-CAP all-async "      "9"

  # THE CAPABILITY AND OPTION TABLES DO NOT LINE UP, and that is the one thing
  # about them worth remembering: PresentCapabilityFence is 2 while
  # PresentOptionCopy is 2, and PresentCapabilityAsyncMayTear is 8 while
  # PresentOptionSuboptimal is 8. A client that used one table's constant
  # against the other's field would be silently asking for something else, and
  # the pair of `all' cases above is what pins both tables at once.

  xp_line libxpresent/ev-configure "XPRESENT-EV configure " "0"
  xp_line libxpresent/ev-complete  "XPRESENT-EV complete "  "1"
  xp_line libxpresent/ev-idle      "XPRESENT-EV idle "      "2"

  xp_line libxpresent/mask-configure "XPRESENT-MASK configure " "1"
  xp_line libxpresent/mask-complete  "XPRESENT-MASK complete "  "2"
  xp_line libxpresent/mask-idle      "XPRESENT-MASK idle "      "4"

  # 7, not 15: the redirect bit is compiled out. See section 9's FUTURE cases.
  xp_line libxpresent/mask-all       "XPRESENT-MASK all "       "7"

  # Each mask is 1 << its own event subtype -- the relationship between the two
  # tables above, stated once as a fact rather than three times as arithmetic
  # the reader has to check. These are what notice a header edit that broke the
  # correspondence while leaving both numbers individually plausible.
  xp_line libxpresent/mask-shift-configure \
    "XPRESENT-MASK configure-is-shift " "yes"
  xp_line libxpresent/mask-shift-complete \
    "XPRESENT-MASK complete-is-shift " "yes"
  xp_line libxpresent/mask-shift-idle "XPRESENT-MASK idle-is-shift " "yes"

  xp_line libxpresent/kind-pixmap     "XPRESENT-KIND pixmap "     "0"
  xp_line libxpresent/kind-notify-msc "XPRESENT-KIND notify-msc " "1"

  xp_line libxpresent/mode-copy       "XPRESENT-MODE copy "             "0"
  xp_line libxpresent/mode-flip       "XPRESENT-MODE flip "             "1"
  xp_line libxpresent/mode-skip       "XPRESENT-MODE skip "             "2"
  xp_line libxpresent/mode-suboptimal "XPRESENT-MODE suboptimal-copy "  "3"

  # --- the wire structures --------------------------------------------------
  #
  # presentproto.h's, and the other half of section 9's question. The wire
  # event is 40 bytes where the client struct is 64: the wire packs the header
  # into 12 bytes and carries the geometry as 16-bit fields, and Xlib's
  # WireToCookie widens them. What must NOT differ is the counters -- 8 bytes on
  # the wire and 8 in the struct -- because a mismatch there is exactly the
  # truncation this port is about, and it would be invisible in the client
  # struct alone.
  xp_line libxpresent/wire-notify   "XPRESENT-WIRE sizeof-xPresentNotify " "8"
  xp_line libxpresent/wire-complete "XPRESENT-WIRE sizeof-complete "       "40"
  xp_line libxpresent/wire-idle     "XPRESENT-WIRE sizeof-idle "           "32"
  xp_line libxpresent/wire-ust      "XPRESENT-WIRE complete-ust "          "8"
  xp_line libxpresent/wire-msc      "XPRESENT-WIRE complete-msc "          "8"
  xp_line libxpresent/wire-req-tmsc \
    "XPRESENT-WIRE pixmap-req-target-msc " "8"

  # -------------------------------------------------------------------------
  # 11. THE ENTRY-POINT CENSUS.
  #
  # Section 12 establishes that no display opens here, which makes the
  # round-trip half of the API unreachable -- so the only thing left to say
  # about it is that it EXISTS: that the cross build compiled the module, that
  # the static linker resolved each name out of libXpresent.a, and that the
  # address is in the shipped image. Taking the address is what forces the
  # archive member in, so a symbol the build had silently dropped fails the LINK
  # in do_compile, naming what went missing.
  #
  # Nine names: the eight Xpresent.h declares, plus XPresentFindDisplay, which
  # is not in the header but has external linkage and is what section 7 drives.
  # Checked by COUNT and by every name -- the library is one translation unit,
  # so there is no module coverage to sample and naming all nine costs nothing.
  # The recipe's own LIBXPRESENT_MIN_ENTRY_POINTS floor is what notices the
  # probe's table shrinking to match a loss.

  nsym=$(sed -n 's/^XPRESENT-SYM-COUNT //p' "$work/out")
  if [ -n "$nsym" ] && [ "$nsym" -ge 9 ]; then bst_ok libxpresent/sym-count
  else bst_notok libxpresent/sym-count \
    "census names ${nsym:-0} entry points, expected not less than 9"; fi

  xp_line libxpresent/sym-none-null "XPRESENT-SYM-NULL " "0"

  for s in XPresentQueryExtension XPresentQueryVersion XPresentVersion \
           XPresentFindDisplay XPresentPixmap XPresentNotifyMSC \
           XPresentSelectInput XPresentFreeInput XPresentQueryCapabilities; do
    xp_line "libxpresent/sym-$s" "XPRESENT-SYM $s " "linked"
  done

  # -------------------------------------------------------------------------
  # 12. THE DISPLAY THAT IS NOT THERE.
  #
  # The documented no-server path, and the reason section 8 had to seed the list
  # to say anything at all: libXpresent's API begins at a Display, Xlib is the
  # only thing that makes one, and on this image it cannot.
  #
  # These are the ticket's two DISPLAY fixtures. `no-env' is the probe having
  # unset DISPLAY -- not a spec of "", because XOpenDisplay("") and
  # XOpenDisplay(NULL) both fall back to $DISPLAY and would make the case a test
  # of whatever the runner's environment carried. `unix-99' is DISPLAY pointing
  # at an address nothing is listening on; a unix-socket spelling deliberately,
  # because it fails at connect(2) with ENOENT and CANNOT block, where a TCP
  # spelling would put a resolver and a connect timeout between the call and its
  # answer.
  #
  # What must hold is that XOpenDisplay SAYS so -- NULL, promptly, for each
  # spelling -- rather than crashing, blocking, or handing back a half-built
  # Display that the first XPresentQueryVersion would then walk into. The last
  # case is what distinguishes a clean refusal from a hang: the probe reached
  # its end.

  xp_line libxpresent/dpy-no-env  "XPRESENT-DPY no-env "  "NULL"
  xp_line libxpresent/dpy-unix-99 "XPRESENT-DPY unix-99 " "NULL"
  xp_line libxpresent/dpy-end     "XPRESENT-DPY reached-end " "yes"
else
  bst_notok libxpresent/probe-stage "cannot stage $XT/libxpresent-probe"
fi

rm -rf "$work"