view tests/test-utils.sh @ 395:bc7a821004bb api-inversion

Invert audioDB::status / audiodb_status().

To do that without breaking abstractions, we actually need a new field
in the status structure, storing the size of the data region.
Previously, this was computed in the audioDB::status request from the
database header, but I'm assuming that "user" code doesn't have access
to such internals.  While we're at it, name some intermediate values in
audioDB::status() so that I don't get confused.

Here's the thing, though: we need to make sure that the adb_t * that we
have from audiodb_open() or audiodb_create() is propagated all the way
through into the C++ routines that implement library functions -- in
particular those which actually write to the database; otherwise we
won't have a consistent view in memory of the header on-disk (as the adb
header that will have been written to disk won't be the same as the one
in memory).

We can do that, by altering the "API" audioDB constructors to take the
adb_t * argument, and setting the adb field in the audioDB object that
we've already introduced to that.  But now we need to be careful a
couple of times: if we have one, then audioDB::initTables() mustn't
stomp on it; also, if we're only constructing an audioDB instance to
fulfil an API request, we mustn't audiodb_close() the one we have when
we destroy the audioDB object, because the adb_t * is the one we have
passed in and are going to reuse in later calls to the API.

The good news is that we can be careful in just these ways with minimal
code.  The really good news is that once the inversion is complete, all
of this horribleness will automatically go away (as there will be no
code which constructs audioDB objects to fulfil API functions).  Hooray!
It's almost like it was all planned this way.

# no shebang line: this file should be sourced by run-test.sh files

set -E

trap "exit 1" ERR

if [ -z "${AUDIODB}" ]; then
  AUDIODB=../../audioDB
fi

# FIXME: maybe generalize to multiple arguments?  Also, implement it
# properly, rather than just for a few floats that we know how to
# encode.  This might involve writing some C code, as Bash doesn't do
# Floating Point.  (scanf() is probably enough).

expect_clean_error_exit() {
  trap - ERR
  "$@"
  exit_code=$?
  trap "exit 1" ERR
  if [ $exit_code -eq 0 ]; then
    exit 1
  elif [ $exit_code -ge 126 ]; then
    exit 1
  fi
}

floatstring() {
  for arg in "$@"; do
    case ${arg} in
      0)
        printf "\x00\x00\x00\x00\x00\x00\x00\x00";;
      -0.5)
        printf "\x00\x00\x00\x00\x00\x00\xe0\xbf";;
      0.5)
        printf "\x00\x00\x00\x00\x00\x00\xe0\x3f";;
      -1)
        printf "\x00\x00\x00\x00\x00\x00\xf0\xbf";;
      1)
        printf "\x00\x00\x00\x00\x00\x00\xf0\x3f";;
      *)
        echo "bad arg to floatstring(): ${arg}"
        exit 1;;
    esac
  done
}

# FIXME: likewise.  And endianness issues (which are a reflection of
# the endianness of audioDB as of 2007-09-18, unfortunately).

intstring() {
  # works up to 9 for now
  if [ $1 -ge 10 ]; then echo "intstring() arg too large: ${1}"; exit 1; fi
  printf "%b\x00\x00\x00" "\\x${1}"
}

# Web services utilities
start_server() {
  $1 -s $2 &
  # HACK: deal with race on process creation
  sleep 1
  trap 'kill $!; exit 1' ERR
}

stop_server() {
  grep "${AUDIODB}" /proc/$1/cmdline > /dev/null
  kill $1
  # HACK: deal with race on process exit
  sleep 1
  expect_clean_error_exit grep ${AUDIODB} /proc/$1/cmdline
}

check_server() {
  grep "${AUDIODB}" /proc/$1/cmdline > /dev/null
}

expect_client_failure() {
  # FIXME: work out whether and how the client should report server
  # errors.  At present, the client exits with a zero exit code.
  "$@"
}