gcc/contrib/reghunt/bin/reg-hunt

#! /bin/bash

#set -x

########################################################################
#
# File:    reg-hunt
# Author:  Janis Johnson <janis187@us.ibm.com>
# Date:    2003/08/19
#
# Search for the patch identifier for which results for a test changed,
# using a binary search.  The functionality for getting sources,
# building the component to test, and running the test are in other
# scripts that are run from here.  Before the search begins, we verify
# that we get the expected behavior for the first and last patch
# identifiers.
#
# Define these in a file whose name is the argument to this script:
#   LOW_PATCH:  Patch identifier.
#   HIGH_PATCH: Patch identifier.
#   REG_UPDATE: Pathname of script to update your source tree; returns
#               zero for success, nonzero for failure.
#   REG_BUILD:  Pathname of script to build enough of the product to run
#               the test; returns zero for success, nonzero for failure.
#   REG_TEST:   Pathname of script to run the test; returns 1 if we
#               should search later patches, 0 if we should search
#               earlier patches, and something else if there was an
#               unexpected failure.
# Optional:
#   REG_REPORT  Pathname of script to call at the end with the id of the
#               patch that caused the change in behavior.
#   REG_FINISH  Pathname of script to call at the end with the two final
#               patch identifiers as arguments.
#   REG_NEWMID  Pathname of script to call when a build has failed, with
#               arguments of the failed id and the current low and high
#   SKIP_LOW    If 1, skip verifying the low patch identifier of the
#               range; define this only if you're restarting and have
#               already tested the low patch.
#   SKIP_HIGH   If 1, skip verifying the high patch identifier of the
#               range; define this only if you're restarting and have
#               already tested the high patch.
#   FIRST_MID   Use this as the first midpoint, to avoid a midpoint that
#               is known not to build.
#   VERBOSITY   Default is 0, to print only errors and final message.
#   DATE_IN_MSG If set to anything but 0, include the time and date in
#               messages.
#
#
#
# Copyright (c) 2002, 2003, 2004, 2005, 2007 Free Software Foundation, Inc.
#
# This file is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# For a copy of the GNU General Public License, write the the
# Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
# Boston, MA 02111-1301, USA.
# 
########################################################################

########################################################################
# Functions
########################################################################

# Issue a message if its verbosity level is high enough.

msg() {
  test ${1} -gt ${VERBOSITY}  && return

  if [ "x${DATE_IN_MSG}" = "x" ]; then
    echo "${2}"
  else
    echo "`date`  ${2}"
  fi
}

# Issue an error message and exit with a non-zero status.  If there
# is a valid current range whose end points have been tested, report
# it so the user can start again from there.

error() {
  msg 0 "error: ${1}"
  test ${VALID_RANGE} -eq 1 && \
    echo "current range:"
    echo "LOW_PATCH=${LATER_THAN}"
    echo "HIGH_PATCH=${EARLIER_THAN}"
  exit 1
}

# Build the components to test using sources as of a particular patch
# and run a test case.  Pass each of the scripts the patch identifier
# that we're testing; the first one needs it, the others can ignore it
# if they want.

process_patch () {
  TEST_ID=${1}

  # If we're keeping track of known failures, see if TEST_ID is one and
  # if so, don't bother updating sources and trying to build.

  FAILS=0
  SKIP=0
  if [ ${SKIP_FAILURES} -eq 1 ]; then
    ${REG_CHECKFAIL} ${TEST_ID}
    if [ $? -eq 0 ]; then
      msg 1 "skipping ${TEST_ID}; it is a known build failure"
      FAILS=1
      SKIP=1
    fi
  fi

  if [ ${FAILS} -eq 0 ]; then
    ${REG_UPDATE} ${TEST_ID} || error "source update failed for ${TEST_ID}"
    ${REG_BUILD} ${TEST_ID}
    if [ $? -ne 0 ]; then
      FAILS=1
      msg 1 "build failed for ${TEST_ID}"
      if [ ${SKIP_FAILURES} -eq 1 ]; then
        ${REG_RECORDFAIL} ${TEST_ID}
      fi
    fi
  fi

  if [ ${FAILS} -eq 0 ]; then
    ${REG_TEST} ${TEST_ID}
    LATER=$?
    if [ $LATER -ne 0 -a $LATER -ne 1 ]; then
      msg 0 "unexpected test failure for ${TEST_ID}"
      exit 1
    fi
  else

    # The build failed, or this patch is already known to fail to build.
    # If it's an endpoint, or if we don't have a way to recover from
    # build failures, quit now.

    if [ ${SKIP} -eq 0 ]; then
      if [ "x${REG_NEWMID}" == "x" \
           -o ${TEST_ID} -eq ${LATER_THAN} \
           -o ${TEST_ID} -eq ${EARLIER_THAN} ]; then
        error "build failed for ${TEST_ID}"
      fi
    fi

    # Try to find a new patch to try within the current range.

    FIRST_MID=`${REG_NEWMID} ${LATER_THAN} ${EARLIER_THAN}`
    if [ ${FIRST_MID} -eq 0 ]; then

      # The heuristics in the tool ran out of patches to try next;
      # let the user handle it from here.+
      error "build failed for ${TEST_ID}, could not find new candidate"
    fi
    msg 1 "using ${FIRST_MID}, between ${LATER_THAN} and ${EARLIER_THAN}"
  fi

  # Return with a valid LATER value or a new ID to try in FIRST_MID.
}

# Get the number of a patch within the range.  It's not actually the
# middle one, but the one that might minimize the number of checks.

get_mid_special() {
  LOW=$1
  HIGH=$2

  let DIFF=HIGH-LOW
  M=1
  POWER2=1
  while
      [ $POWER2 -lt $DIFF ]
  do
      let M=POWER2
      let POWER2=POWER2*2
  done
  let MID=LOW+M
}

# Get the number of the patch in the middle of the range.

get_mid () {
  LOW=$1
  HIGH=$2

  let DIFF=HIGH-LOW
  let M=DIFF/2
  let MID=LOW+M
}

# Perform a binary search on patch identifiers within the range
# specified by the arguments.

search_patches () {
  LOW=$1
  HIGH=$2

  # Get an identifier within the range.  The user can override the
  # initial mid patch if it is known to have problems, e.g., if a
  # build fails for that patch.

  if [ ${FIRST_MID} -ne 0 ]; then
    MID=${FIRST_MID}
    FIRST_MID=0
    let DIFF=HIGH-LOW
  else
    get_mid $LOW $HIGH
  fi

  while [ ${DIFF} -gt 1 ]; do
    TEST_ID="${MID}"

    # Test it.

    process_patch ${TEST_ID}

    # FIRST_MID being set is a signal that the build failed and we
    # should start over again.
    
    test ${FIRST_MID} -ne 0 && return

    # Narrow the search based on the outcome of testing TEST_ID.

    if [ ${LATER} -eq 1 ]; then
      msg 1 "search patches later than ${TEST_ID}"
      LATER_THAN=${TEST_ID}
      let LOW=MID
    else
      msg 1 "search patches earlier than ${TEST_ID}"
      EARLIER_THAN=${TEST_ID}
      let HIGH=MID
    fi

    get_mid $LOW $HIGH
  done
}

########################################################################
# Main program (so to speak)
########################################################################

# The error function uses this.

VALID_RANGE=0

# Process the configuration file.

if [ $# != 1 ]; then
  echo Usage: $0 config_file
  exit 1
fi

CONFIG=${1}
if [ ! -f ${CONFIG} ]; then
  error "configuration file ${CONFIG} does not exist"
fi

# OK, the config file exists.  Source it, make sure required parameters
# are defined and their files exist, and give default values to optional
# parameters.

. ${CONFIG}

test "x${REG_UPDATE}" = "x" && error "REG_UPDATE is not defined"
test "x${REG_BUILD}" = "x" && error "REG_BUILD is not defined"
test "x${REG_TEST}" = "x" && error "REG_TEST is not defined"
test -x ${REG_TEST} || error "REG_TEST is not an executable file"
test "x${SKIP_LOW}" = "x" && SKIP_LOW=0
test "x${SKIP_HIGH}" = "x" && SKIP_HIGH=0
test "x${VERBOSITY}" = "x" && VERBOSITY=0
test "x${REG_FINISH}" = "x" && REG_FINISH=true
test "x${REG_REPORT}" = "x" && REG_REPORT=true

msg 2 "LOW_PATCH  = ${LOW_PATCH}"
msg 2 "HIGH_PATCH = ${HIGH_PATCH}"
msg 2 "REG_UPDATE = ${REG_UPDATE}"
msg 2 "REG_BUILD  = ${REG_BUILD}"
msg 2 "REG_TEST   = ${REG_TEST}"
msg 2 "REG_NEWMID = ${REG_NEWMID}"
msg 2 "SKIP_LOW   = ${SKIP_LOW}"
msg 2 "SKIP_HIGH  = ${SKIP_HIGH}"
msg 2 "FIRST_MID  = ${FIRST_MID}"
msg 2 "VERBOSITY  = ${VERBOSITY}"

# If REG_NEWMID was defined, assume that we're skipping known failures
# and adding to the list for new failures.  If the list of failures
# doesn't exist, create it.  We use a different flag, SKIP_FAILURES,
# to make it easier to separate the flag from REG_NEWMID if we want
# to change the usage later.

if [ "x${REG_NEWMID}" != "x" ]; then
  touch ${REG_FAILLIST}
  SKIP_FAILURES=1
else
  SKIP_FAILURES=0
fi

# If FIRST_MID was defined, make sure it's in the range.

if [ "x${FIRST_MID}" != "x" ]; then
  test ${FIRST_MID} -le ${LOW_PATCH}  && \
    error "FIRST_MID id is lower than LOW_PATCH"
  test ${FIRST_MID} -ge ${HIGH_PATCH} && \
    error "FIRST_MID is higher than HIGH_PATCH"
else
  FIRST_MID=0
fi 

# Keep track of the bounds of the range where the test behavior changes.

LATER_THAN=${LOW_PATCH}
EARLIER_THAN=${HIGH_PATCH}
LATER=1

msg 1 "LATER_THAN   = ${LATER_THAN}"
msg 1 "EARLIER_THAN = ${EARLIER_THAN}"

# Verify that the range isn't backwards.

test ${LOW_PATCH} -lt ${HIGH_PATCH} || \
  error "patch identifier range is backwards"

# Verify that the first and last patches in the range get the results we
# expect.  If not, quit, because any of several things could be wrong.

if [ ${SKIP_HIGH} -eq 0 ]; then
  process_patch ${EARLIER_THAN}
  test ${LATER} -ne 0 && \
    error "unexpected result for high patch ${EARLIER_THAN}"
  msg 1 "result for high patch ${EARLIER_THAN} is as expected"
fi

if [ ${SKIP_LOW} -eq 0 ]; then
  process_patch ${LATER_THAN}
  test ${LATER} -ne 1 && \
    error "unexpected result for low patch ${LATER_THAN}"
  msg 1 "result for low patch ${LATER_THAN} is as expected"
fi

# Search within the range, now that we know that the end points are valid.
# If the build failed then FIRST_MID is set to a new patch to try.

VALID_RANGE=1
while true; do
  search_patches ${LATER_THAN} ${EARLIER_THAN}
  test ${FIRST_MID} -eq 0 && break
done

# Report where the test behavior changes.

echo "Test result changes with id ${EARLIER_THAN}"
${REG_REPORT} ${EARLIER_THAN}

# Invoke the optional script to verify the result and report additional
# information about changes between the two patches.

${REG_FINISH} ${LATER_THAN} ${EARLIER_THAN}