[libvirt] [PATCH] [TCK] Convert scripts to also run with dash

[Stefan Berger]

On 11/15/2010 12:35 PM, Eric Blake wrote:
> On 11/15/2010 08:33 AM, Stefan Berger wrote:
>> I am converting the shell-scripted tap tests to also run using the dash
>> shell, fixing some other problems on the way as well.
>>
>> Two of the tests' data files contained '\\1' in an sed command. This
>> created char(1) with the dash, '\1' with the bash. I found it easier to
>> replace the sed command with a different one than figuring out how to
>> solve this escape problem so it works for both shells.
>>
>>
>>
>> -function tap_fail() {
>> -  echo "not ok $1 - ${2:0:66}"
>> -  TAP_FAIL_LIST+="$1 "
>> -  ((TAP_FAIL_CTR++))
>> -  ((TAP_TOT_CTR++))
>> +tap_fail() {
>> +  txt=$(echo "$2" | gawk '{print substr($0,1,66)}')
>> +  echo "not ok $1 - ${txt}"
>> +  TAP_FAIL_LIST="$TAP_FAIL_LIST $1 "
>> +  TAP_FAIL_CTR=$((TAP_FAIL_CTR+1))
> Older versions of dash have a bug where you have to pre-expand a
> variable to get this to work:
>
> TAP_FAIL_CTR=$(($TAP_FAIL_CTR + 1))
>
> (throughout the patch)
>
Fixed.
>> @@ -112,8 +111,8 @@ function checkExpectedOutput() {
>>
>>     exec 4<"${datafile}"
>>
>> -  read<&4
>> -  line="${REPLY}"
>> +  IFS=""
> Where do you restore IFS back to its original space-tab-newline setting?
>
Fixed.
>>       while :; do
>> -      read<&4
>> -      line="${REPLY}"
>> +      read line<&4
>> +
>> +      letter=$(echo ${line} | gawk '{print substr($1,1,1)}')
>>
>> -      if [ "${line:0:1}" == "#" ] || [ "x${line}x" == "xx"  ]; then
>> +      if [ "x${letter}x" = "x#x" ] || [ "x${line}x" = "xx"  ]; then
> Rather than wasting a $() and gawk subprocess to compute $letter, you
> can use the much more efficient:
>
> case $line in
>    '#'*) # line started with #
>      ;;
>    *)    # line did not start with #
>      ;;
> esac
>
Ok, since it's faster.
>> @@ -327,13 +320,13 @@ function main() {
>>     fi
>>
>>     if [ $((flags&  FLAG_LIBVIRT_TEST)) -ne 0 ]; then
>> -    pushd "${PWD}">  /dev/null
>> +    curdir="${PWD}"
>>       . ./test-lib.sh
>>       if [ $? -ne 0 ]; then
>>           exit 1
>>       fi
>>       test_intro $this_test
>> -    popd>  /dev/null
>> +    cd "${curdir}"
> You really should check whether cd succeeds; as failure to return to the
> parent directory deserves aborting the script to avoid corrupting
> unintended files in a different directory.
>
Fixed.
>> Index: libvirt-tck/scripts/networks/100-apply-verify-host.t
>> ===================================================================
>> --- libvirt-tck.orig/scripts/networks/100-apply-verify-host.t
>> +++ libvirt-tck/scripts/networks/100-apply-verify-host.t
>> @@ -1,10 +1,5 @@
>> -#!/bin/bash
>> +#!/bin/sh
>>
>>   pwd=$(dirname $0)
>>
>> -pushd ${PWD}>  /dev/null
>> -
>> -cd ${pwd}
>> -bash ./networkApplyTest.sh --tap-test
>> -
>> -popd>  /dev/null
>> +(cd ${pwd}; sh ./networkApplyTest.sh --tap-test)
> To be robust to starting the script via a relative pathname that starts
> with -, I'd write this as:
>
> pwd=$(dirname -- "$0")
> (cd -- "${pwd}"&&  sh ./networkApplyTest.sh --tap-test
>
Fixed.
>> Index: libvirt-tck/scripts/networks/networkxml2hostout/tck-testnet-1.dat
>> ===================================================================
>> --- libvirt-tck.orig/scripts/networks/networkxml2hostout/tck-testnet-1.dat
>> +++ libvirt-tck/scripts/networks/networkxml2hostout/tck-testnet-1.dat
>> @@ -5,7 +5,7 @@ MASQUERADE  all  --  10.1.2.0/24
>>   #iptables -n -L FORWARD | grep ' 10\\.1\\.2\\.'
>>   ACCEPT     all  --  0.0.0.0/0            10.1.2.0/24         state RELATED,ESTABLISHED
>>   ACCEPT     all  --  10.1.2.0/24          0.0.0.0/0
>> -#ps aux | sed -n '/dnsmasq .*10\\.1\\.2\\./ s|.*\\(dnsmasq[[:print:]*]\\)|\\1|p'
>> +#ps aux | sed -n '/dnsmasq .*10\\.1\\.2\\./ s|.*dnsmasq|dnsmasq|p'
> Hmm; you used 'read' to read in these lines, but unless you use 'read
> -r', that means that backslash interpolation is taking place in the
> shell.  'read -r' is portable to POSIX (and therefore dash), even if it
> is not portable to generic /bin/sh (but we already have lots of other
> things that are not portable to generic /bin/sh that aren't worth
> fretting about, since we can assume that the nwfilter tests only make
> sense on Linux where /bin/sh will mostly conform to POSIX).  Maybe it's
> better to fix the driver scripts to use 'read -r', at which point these
> lines could use single \, while at the same time working around the
> problem you were seeing where dash interpolated \1 inside "" into char(1).
>
I tried this now and converted all \\ to \. I still get the same problem 
here with dash using 'read -r line' now. The same problem exists with

echo "\1"

or

echo '\1'

run via bash or dash. Bash prints out '\1' and dash (0.5.6-2.fc14) 
prints char(1) in both cases.

    Stefan