#!/usr/bin/env bash


# Script to do automated testing and data collection for various test
# files, so that we don't have to do this by hand on every test file.
# It attempts to collect some diagnostic info about size and speed that
# should be useful in the future as the library gets tuned for size
# and speed.  In addition, it tests static and shared linkage, iff each
# has been enabled.

# Invocation 
# mkcheck [012] (path to build) (path to src) (path to install)

### XXX There are a lot of tests in here for OS-specific stuff.  If we
###     move to a 'configure.target' method of determining those extra
###     flags and whatnot, we can take out all those things and source
###     that file from here.  (Write that file with this in mind...)

### XXX Note that breaking out of this with ^C will not work.  Dunno why.


# This has been true all along.  Found out about it the hard way...
case $BASH_VERSION in
    1*)  echo 'You need bash 2.x to run mkcheck.  Exiting.'; exit 1 ;;
    *)   ;;   # ??
esac


#
# 1: variables
#
# WHICH determines if you are
# (0) testing the build binary and headers, or
# (1) testing the installed binary and headers, or
# (2) testing under dejagnu (just print the standard flags needed).
WHICH=$1
if [ "$WHICH"x = 0x ] && [ $# -eq 3 ]; then
  echo "running mkcheck"
  echo "$0: testing the build directory"
elif [ "$WHICH"x = 1x ] && [ $# -eq 4 ]; then
  echo "running mkcheck"
  echo "$0: testing the install directory $4"
elif [ "$WHICH"x = 2x ] && [ $# -eq 3 ]; then
  true
else
  echo 'Usage: mkcheck 0 (path to build) (path to src)'
  echo '       mkcheck 1 (path to build) (path to src) (path to install)'
  echo '       mkcheck 2 (path to build) (path to src)'
  exit 1;
fi
  
BUILD_DIR=$2
if [ ! -d "$BUILD_DIR" ]; then
  echo "build directory $BUILD_DIR not found, exiting."
  exit 1
fi

SRC_DIR=$3
if [ ! -d "$SRC_DIR" ]; then
  echo "source directory $SRC_DIR not found, exiting."
  exit 1
fi

if [ $WHICH -eq 1 ]; then 
    PREFIX_DIR=$4
    if [ ! -d "$PREFIX_DIR" ]; then
    echo "install directory $PREFIX_DIR not found, exiting."
    exit 1
    fi
fi

# This is LIBTOOL=@LIBTOOL@ piped through a bit of sanity that we can
# assume for this script (by the time we run this).
LIBTOOL="$BUILD_DIR/libtool"
chmod u+x $LIBTOOL

# INC_PATH == include path to new headers for use on gcc command-line
top_srcdir=@top_srcdir@
C_DIR="`basename @C_INCLUDE_DIR@`"
if [ $WHICH != "1" ]; then
  INC_PATH="-nostdinc++ @CSHADOW_FLAGS@ -I$BUILD_DIR/include \
    -I$SRC_DIR/include/std  -I$SRC_DIR/include/$C_DIR \
    -I$SRC_DIR/include -I$SRC_DIR/libsupc++ -I$SRC_DIR/libio \
    -I$SRC_DIR/testsuite"
elif [ $WHICH -eq 1 ]; then
  INC_PATH="-I$SRC_DIR/testsuite"
fi

if [ $WHICH -eq 2 ]; then
  echo $INC_PATH -I$SRC_DIR/include/backward -I$SRC_DIR/include/ext
  exit 0;
fi

# It's not dejagnu; we need to do things ourselves.  Pick up any extra
# settings for this target.
. ${top_srcdir}/configure.target

# LIB_PATH == where to find the build libraries for libtool's use
# CXX == how to call the compiler
if [ $WHICH -eq 0 ]; then
  LIB_PATH="$BUILD_DIR/src"
  CXX="$BUILD_DIR/../../gcc/g++ -B$BUILD_DIR/../../gcc/"
elif [ $WHICH -eq 1 ]; then
  LIB_PATH="$PREFIX_DIR/lib"
  CXX="$PREFIX_DIR/bin/g++"
fi

# gcc compiler flags (maybe use glibcpp_cxxflags from configure.target,
# but thst's really meant for building the library itself, not using it)
CXX_FLAG="-ggdb3 -DDEBUG_ASSERT @SECTION_FLAGS@ @SECTION_LDFLAGS@"

# specific libtool flag(s) to force the use of shared libraries, if any
SH_FLAG=""

# specific libtool flag(s) to force the use of static libraries, if any
#ST_FLAG="-static"
ST_FLAG="-all-static"

# LTCXX == how to call libtool when creating an executable
# LTEXE == how to call libtool when running an executable
# LIBS == any extra needed -l switches, etc (may need more libs, lose lose)
if [ $WHICH -eq 0 ]; then
  LTCXX="$LIBTOOL --tag=CXX --mode=link \
          $CXX $CXX_FLAG $INC_PATH   \
          $LIB_PATH/../libsupc++/libsupc++.la  $LIB_PATH/libstdc++.la   \
          -no-install"
  LTEXE="$LIBTOOL --mode=execute"
  LIBS="-nodefaultlibs -lc -lgcc -lc"
elif [ $WHICH -eq 1 ]; then
  # For the installed version, we really only need to use libtool and
  # the .la file to get correct rpaths.
  LTCXX="$LIBTOOL --tag=CXX --mode=link \
          $CXX $CXX_FLAG $INC_PATH -L$LIB_PATH   \
          $LIB_PATH/libstdc++.la -no-install -rpath $LIB_PATH"
  LTEXE="$LIBTOOL --mode=execute"
  LIBS=
fi

# Set up the testing directory, which should be in a directory called
# "testsuite" in the root level of the build directory.
TEST_DIR="`pwd`/testsuite"
if [ ! -d "$TEST_DIR" ]; then
    echo "making directory $TEST_DIR"
    mkdir $TEST_DIR
    mkdir $TEST_DIR/.libs    # help libtool keep quiet
    chmod u+w $TEST_DIR
fi

# the name of the file that will collect and hold all this useful data:
RESULTS_FILE="$TEST_DIR/$(date +%Y%m%d)-mkcheck.txt"

# the name of the log file that will append compiler diagnostics:
LOG_FILE="$TEST_DIR/$(date +%Y%m%d)-mkchecklog.txt"

# the names of the specific test files to be run
TESTS_FILE="$TEST_DIR/$(date +%Y%m%d)-mkcheckfiles.txt"

# the heap size limit for testsuite binaries; start with a 2MB limit as per
# http://sources.redhat.com/ml/libstdc++/2000-10/msg00029.html
MAX_MEM_USAGE=3072

#
# 2: clean, make files, append general test info
#
if [ -f $RESULTS_FILE ]; then
    rm $RESULTS_FILE
fi
if [ -f $LOG_FILE ]; then
    rm $LOG_FILE
fi

# Make a list of the files we're going to run, or use an old one if it exists.
if [ ! -f "$TESTS_FILE" ]; then
    echo "making file $TESTS_FILE"
    for LONG_NAME in $SRC_DIR/testsuite/*/*.cc
    do
        DIR_NAME=$(dirname $LONG_NAME)
        SHORT_NAME="`basename $DIR_NAME`/`basename $LONG_NAME`"
        echo "$SHORT_NAME" >> $TESTS_FILE
    done
fi

# Nasty solution to replace GNU date(1)'s %s time_t output function.
if [ ! -x "$TEST_DIR/printnow" ]; then
    echo "making utility $TEST_DIR/printnow"
    gcc -o "$TEST_DIR/printnow" "$SRC_DIR/testsuite/printnow.c"
    strip "$TEST_DIR/printnow"
fi

# Remove old executables.
rm -rf "$TEST_DIR"/*exe

# Remove old core files (which now get left in cwd, not $TEST_DIR).
rm -rf ./*core*

# Copy over the data files for filebufs
cp $SRC_DIR/testsuite/27_io/*.txt $TEST_DIR
cp $SRC_DIR/testsuite/27_io/*.tst $TEST_DIR
chmod u+w $TEST_DIR/*.txt
chmod u+w $TEST_DIR/*.tst

# Emit useful info about compiler and platform
echo "host: $(uname -mrsv)" >> $RESULTS_FILE
echo "compiler: $($CXX -v 2>&1 | tail -1)" >> $RESULTS_FILE
echo "$($CXX -v 2>&1 | grep ^Configured)" >> $RESULTS_FILE
echo "compiler flags: $CXX_FLAG" >> $RESULTS_FILE
echo "date: $(date +%Y%m%d)" >> $RESULTS_FILE
echo "" >> $RESULTS_FILE

explanation='+: pass, -b: build failure, -r: run failure, x: disabled'
printf "%s\n     %s\n"  'p == pass/fail execution test'  "$explanation"  \
       >> $RESULTS_FILE
echo "ctime == time to compile and link" >> $RESULTS_FILE
echo "etime == time for executable to run" >> $RESULTS_FILE
echo "text == size of the executable text section" >> $RESULTS_FILE
echo "data == size of the executable data section" >> $RESULTS_FILE
echo "total == size of the executable" >> $RESULTS_FILE
echo "" >> $RESULTS_FILE
echo "(First static, then shared.)" >> $RESULTS_FILE

echo "p" | awk '{printf("%s ", $1)}' >> $RESULTS_FILE
echo "ctime" "etime" | awk '{printf("%s\t%s\t", $1, $2)}' >> $RESULTS_FILE
echo "text" "data" | awk '{printf("%s\t%s\t", $1, $2)}' >> $RESULTS_FILE
echo "total" "name" | awk '{printf("%s\t%s\t", $1, $2)}' >> $RESULTS_FILE
echo "" >> $RESULTS_FILE


#
# 2.5:  support functions
#

# Figure out how to extract size information from binaries.  We take
# the text of the value we want as an argument, and leave the size in
# the appropriate variable.
#
# We discover what kind of size(1) we are using *once* and build a shell
# function named 'size_command' to wrap it.  (The "function" keyword is
# redundant here, but helps me read it, so there.)  Previously we were
# re-discovering the size(1) arguments three times for each test; sloooow.
#
# It is VERY IMPORTANT not to compare these numbers across platforms.
# Different size(1)'s extract section information differently.  For
# example, using the native Sun size(1) and GNU size(1) built for Suns
# on the exact same binary will give very different numbers, due to all
# the variance in command-line options and arbitrary names of ELF sections.
#
# and suddenly we go to 2-space indentations...
setup_size_command()
{
  if size --version 2> /dev/null | grep -c GNU > /dev/null;
  then    # Then we're using a GNU size(1) built for this platform.
    # We lose .rodata and .data1 and who knows what else... kludge.
    function size_command()
    {
      case $1 in
        TEXT)  TEXT=$(size -A $EXENAME | grep ^.text | awk '{print $2}')  ;;
        DATA)  DATA=$(size -A $EXENAME | grep -w ^.data | awk '{print $2}')  ;;
        SIZE)  SIZE=$(size -A $EXENAME | grep otal | awk '{print $2}')  ;;
      esac
    }
  else
    # Not using GNU size; check for platform.  These numbers seem to match
    # up to text/data/total, although their meanings seem to be different.
    # THIS TABLE IS SORTED.  KEEP IT THAT WAY.
    case @host_os@ in
      *aix*)
        function size_command()
        {
          case $1 in
            TEXT)  TEXT=$(size -X32_64 $EXENAME | awk '{print $2}')  ;;
            DATA)  DATA=$(size -X32_64 $EXENAME | awk '{print $4}')  ;;
            SIZE)  SIZE=$(size -X32_64 $EXENAME | awk '{print $12}')  ;;
          esac
        }
        ;;
      *irix*)
        function size_command()
        {
          case $1 in
            TEXT)  TEXT=$(size -4 $EXENAME | awk '{print $1}')  ;;
            DATA)  DATA=$(size -4 $EXENAME | awk '{print $3}')  ;;
            SIZE)  SIZE=$(size -4 $EXENAME | awk '{print $7}')  ;;
          esac
        }
        ;;
      *solaris*)
        function size_command()
        {
          case $1 in
            TEXT)  TEXT=$(size $EXENAME | awk '{print $1}')  ;;
            DATA)  DATA=$(size $EXENAME | awk '{print $3}')  ;;
            SIZE)  SIZE=$(size $EXENAME | awk '{print $7}')  ;;
          esac
        }
        ;;
      *)
        echo ' * Warning!  Skipping section sizes!' 1>&2
        function size_command()
        {
        case $1 in
          TEXT)  TEXT=0 ;;
          DATA)  DATA=0 ;;
          SIZE)  SIZE=0 ;;
        esac
        }
        ;;
    esac
  fi
}

# Test for file output
test_for_output()
{
    # This checks for emitted output files, which is useful when
    # testing file-related output.  The rules for this working are as
    # follows: the emitted file must have the ".txt" extension, and be
    # based on the actual *.cc file's name.  For example, 27/filbuf.cc
    # currently outputs files named 27/filebuf-2.txt and 27/filebuf-3.txt.
    # Also, the first emitted file must be in the form $NAME-1.txt.
    # The control file must follow the same constraints, but have a
    # ".tst" extension.  Thus, you have 27/filebuf-2.tst, etc.

    # NAME contains the source name, like 27/filebuf.cc
    # From that NAME, we want to generate some possible names, using
    # ls on MATCH, a pattern description generated with sed.

    # this is the name of the resulting diff file, if any
    DIFF_FILE="`echo $TEST_NAME | sed 's/cc$/diff/'`"
    # construct wildcard names, ie for $NAME=filebuf.cc, makes "filebuf*.tst"
    DATA_FILES="`echo $NAME | sed 's/\.cc/\*\.tst/g'`"
    # make sure there is at least one, then go
    ST_E="`echo $NAME | sed 's/\.cc/\-1\.tst/g'`"
    if [ -f $ST_E ]; then
        # list of actual files that match the wildcard above, ie
        # "filebuf-1.tst"
        ST_MATCH_LIST="`ls $DATA_FILES`"
        for i in $ST_MATCH_LIST; do
            # ST_OUT_FILE is generated in the build directory.
            PRE_NAME2="$TEST_DIR/`basename $i`"
            ST_OUT_FILE="`echo $PRE_NAME2 | sed 's/tst$/txt/'`"
            diff $ST_OUT_FILE $i > $DIFF_FILE
            if [ -s $DIFF_FILE ]; then
                RESULT="-r"
                echo "$ST_OUT_FILE has some problems, dude"
            else
                RESULT="+"
            fi
            rm $DIFF_FILE
        done
    else
        # the file does no output, and didn't abnormally
        # terminate, so assume passed.
        RESULT="+"
    fi
}
    

#
# 3: compile, link, execute, time
#
# Abstract out the common code for compiling, linking, executing and printing.
test_file()
{
    # NB: S_FLAG has to be last argument because it may be null, and
    # error checking hasn't been invented yet.
    NAME=$1
    EXENAME=$2
    S_FLAG=$3

    SRC_NAME="$SRC_DIR/testsuite/$1"
    TEST_NAME="$TEST_DIR/`basename $NAME`"

    # This would be deliciously easy if GNU date's %s were always around.
    # There are three ways to do this:  1) use the builtin 'time' like we
    # do later; then getting compiler errors into LOG_FILE is a nightmare.
    # 2) Grab the output of a formatted date(1) and do the math; harder
    # and harder as we try compiling at, say, top of the hour; we would
    # eventually have to calculate time_t anyhow.  Or 3) just grab two
    # time_t's (no more overhead than grabbing two date(1)'s).
    compiler_invocation="$LTCXX $S_FLAG $SRC_NAME -o $EXENAME $LIBS"
    # Why the leading weird character, like a colon?  That way you can
    # 'grep -v ^: *mkchecklog.txt' and just see the errors.
    #printf "\n: " >> $LOG_FILE
    printf "\n" >> $LOG_FILE
    COMP_TIME_START=$($TEST_DIR/printnow)
    $compiler_invocation >> $LOG_FILE 2>&1
    COMP_TIME_END=$($TEST_DIR/printnow)

    if [ $COMP_TIME_START -lt $COMP_TIME_END ]; then
        C_TIME=$[ $COMP_TIME_END - $COMP_TIME_START ]
    else
        C_TIME="0"
    fi

    if [ -f $EXENAME ]; then
        size_command TEXT
        size_command DATA
        size_command SIZE

        # Actually run the executable and time it.  Note that output
        # printed by the executable will be lost and cannot be redirected,
        # because we need to capture the output of 'time'.  Bummer.
        TIMEFORMAT='timemark %R'
        E_TIME_TEXT="$(exec 2>&1; ulimit -d $MAX_MEM_USAGE; \
                     time $LTEXE $EXENAME)"
        E_ABNORMAL_TERMINATION=$?
        E_TIME="$(echo $E_TIME_TEXT | awk '{print $2}')"
        # joining those two commands does not work due to quoting problems:
        #E_TIME="$(exec 2>&1; time $EXENAME | awk '{print $2}')"
        # this will work as a fallback on certain systems...?
        #E_TIME=$(exec 2>&1; time $EXENAME | cut -d ' ' -f 2)
 
        if [ "$E_ABNORMAL_TERMINATION" -ne 0 ]; then
            RESULT='-r'
            rm -f ./*core
            # sometimes you want to save all core files for review:
            #mv ./core $EXENAME.core
            # sometimes the OS allows you to name core files yourself:
            #mv ./*core $EXENAME.core
            #mv ./core* $EXENAME.core
        else
            test_for_output
        fi

        # sometimes you want to save all failing exe files for review:
        if [ "$RESULT" = "+" ]; then
            rm "$EXENAME"
        fi
    else
        # the file did not compile/link.
        RESULT="-b"
        TEXT="0"
        DATA="0"
        SIZE="0"
    fi

    printf "%s\t" "$RESULT"
    printf "%-2s %d\t%.3f\t%s\t%s\t%s\t%s\n"   \
        "$RESULT" $C_TIME $E_TIME $TEXT $DATA $SIZE $NAME >> $RESULTS_FILE
}

setup_size_command
echo ""
echo "Detailed test results in .${RESULTS_FILE/$BUILD_DIR}"
echo $explanation
echo "------------------------------------------------------------------------"
printf "static\tshared\ttest\n"
echo "------------------------------------------------------------------------"

TEST_TIME_START=$($TEST_DIR/printnow)
for NAME in `cat $TESTS_FILE`
do
    PRE_NAME="$TEST_DIR/`basename $NAME`"
    ST_NAME="`echo $PRE_NAME | sed 's/cc$/st-exe/'`"
    SH_NAME="`echo $PRE_NAME | sed 's/cc$/sh-exe/'`"

    if test @enable_static@ = yes; then
        test_file $NAME $ST_NAME $ST_FLAG 
    else
        printf "x\t"
        printf "static skipped\n" >> $RESULTS_FILE
    fi
    if test @enable_shared@ = yes; then
        test_file $NAME $SH_NAME $SH_FLAG 
    else
        printf "x\t"
        printf "shared skipped\n" >> $RESULTS_FILE
    fi
    printf "%s\n" "$NAME"

    echo "" >> $RESULTS_FILE
done
TEST_TIME_END=$($TEST_DIR/printnow)


#
# 4: summary
#
# grep can count faster than we can...
total_failures=$(egrep -c "^\-" $RESULTS_FILE)
total_successes=$(egrep -c "^\+" $RESULTS_FILE)
resultstext="pass/fail results:  ${total_successes}/${total_failures}"
if [ $total_failures -eq 0 ]; then
    resultstext="${resultstext}, WIN WIN"
fi
sed -e "/^date:/a\\
$resultstext" $RESULTS_FILE > ${RESULTS_FILE}.tmp
mv ${RESULTS_FILE}.tmp $RESULTS_FILE

if [ $TEST_TIME_START -lt $TEST_TIME_END ]; then
    TEST_TIME=$[ $TEST_TIME_END - $TEST_TIME_START ]
    echo "testrun == $TEST_TIME seconds"
    echo "testrun == $TEST_TIME seconds" >> $RESULTS_FILE
fi

exit 0