[gentoo-dev] Re: New eclass osgi.eclass
Alistair Bush wrote:
> Im sure Elvanor can't wait for you constructive feedback on his eclass
> and depending on your feedback the eclass will enter the tree this
> weekend.
>
A couple of very minor performance points, which I think are more
significant in eclasses. Firstly the basename thing Donnie pointed out
before:
> > local feedfile=$(basename "${src}")
> You could do this in pure bash, although it doesn't really matter:
> local feedfile=${src##*/}
[[ ${#} -lt 4 ]] && die "At least four arguments needed"
Arithmetic context is quicker for this:
(($#<4)) && die "At least four arguments needed"
although in this case, _java-pkg_osgijar(), it looks it requires exactly 4:
(($#==4)) || die 'Four arguments needed'
You have a couple of functions that take, say, 4 or 5 arguments. It would be
more robust to use a case, eg:
case $# in
5)..;;
4)..;;
*) die "Incorrect use of $FUNCNAME";;
esac
..than if (($#>4)); then ..; else .. ;fi
_java-pkg_osgi-plugin ${4} # this should be quoted (_java-pkg_osgijar)
With regard to:
debug-print-function ${FUNCNAME} $*
if you want debug-print-function to get the arguments correctly, use "$@"
not $* (cf 'Special Parameters' in man bash for a proper explanation: this
applies to all arrays. http://wooledge.org/mywiki/BashFAQ/073 shows how to
manipulate all members of an array, eg to add a prefix.)
This use of counter in _java-pkg_osgijar-fromfile is odd:
while [[ -n "${1}" ]]; do
# while [[ $1 ]] or while (($#)) also work
if [[ "${1}" == "--noversion" ]]; then
noversion=1
else
arguments[${counter}]="${1}"
((++counter))
fi
shift 1 # just shift?
done
(([EMAIL PROTECTED] < 3)) && die "At least three arguments (not
counting --noversion) are needed for java-pkg_osgijar-fromfile()"
You can either just add to the array with: arguments+=("$1")
or add using the counter: arguments[counter++]="$1"
and then check: ((counter < 3)) && die ..
Arithmetic context[1] applies in array indexes, so you don't need to use a $
for variables and you can post/pre-incr etc there.
Yuu can also use it for C-style flags, with 0 as false and non-zero true:
if [[ "${noversion}" == 1 ]];
can be
if ((noversion));
This is handy since unset variables evaluate as zero, which is false inside
((..)), so a simple flag=1 is all that's needed, ie you don't have to set a
default, although ofc it's more robust to do so, especially for functions.
declare -i flag=0 # makes a local var which can only have integer values
assigned (setting it to string typically evaluates to zero; arithmetic
context is used for all assignments, so a string which happens to be a
variable with an integer will return that value.)
[1] http://wooledge.org/mywiki/ArithmeticExpression
--
[EMAIL PROTECTED] mailing list
Re: [gentoo-dev] Re: New eclass osgi.eclass
Steve Long kirjoitti:
> Alistair Bush wrote:
>> Im sure Elvanor can't wait for you constructive feedback on his eclass
>> and depending on your feedback the eclass will enter the tree this
>> weekend.
>>
> A couple of very minor performance points, which I think are more
> significant in eclasses. Firstly the basename thing Donnie pointed out
> before:
Well usually you don't noticy any differences. In global scope things
could have some effect of course.
>
> You have a couple of functions that take, say, 4 or 5 arguments. It would be
> more robust to use a case, eg:
> case $# in
> 5)..;;
> 4)..;;
> *) die "Incorrect use of $FUNCNAME";;
> esac
> ..than if (($#>4)); then ..; else .. ;fi
>
Well imho they are equivalent and eclass maintainers should go with what
they prefer.
>
> With regard to:
> debug-print-function ${FUNCNAME} $*
> if you want debug-print-function to get the arguments correctly, use "$@"
> not $* (cf 'Special Parameters' in man bash for a proper explanation: this
> applies to all arrays. http://wooledge.org/mywiki/BashFAQ/073 shows how to
> manipulate all members of an array, eg to add a prefix.)
Doesn't matter as arguments are just written to a file and for some
reason this form is used all over the base in eclasses but shouldn't
hurt to use "[EMAIL PROTECTED]" with new code.
>
> This use of counter in _java-pkg_osgijar-fromfile is odd:
> while [[ -n "${1}" ]]; do
> # while [[ $1 ]] or while (($#)) also work
>if [[ "${1}" == "--noversion" ]]; then
> noversion=1
>else
> arguments[${counter}]="${1}"
> ((++counter))
>fi
>shift 1 # just shift?
> done
>
> (([EMAIL PROTECTED] < 3)) && die "At least three arguments (not
> counting --noversion) are needed for java-pkg_osgijar-fromfile()"
>
> You can either just add to the array with: arguments+=("$1")
> or add using the counter: arguments[counter++]="$1"
> and then check: ((counter < 3)) && die ..
>
> Arithmetic context[1] applies in array indexes, so you don't need to use a $
> for variables and you can post/pre-incr etc there.
> Yuu can also use it for C-style flags, with 0 as false and non-zero true:
> if [[ "${noversion}" == 1 ]];
> can be
> if ((noversion));
>
> This is handy since unset variables evaluate as zero, which is false inside
> ((..)), so a simple flag=1 is all that's needed, ie you don't have to set a
> default, although ofc it's more robust to do so, especially for functions.
> declare -i flag=0 # makes a local var which can only have integer values
> assigned (setting it to string typically evaluates to zero; arithmetic
> context is used for all assignments, so a string which happens to be a
> variable with an integer will return that value.)
>
> [1] http://wooledge.org/mywiki/ArithmeticExpression
>
We really don't need to create the arguments array at all. We could just
check if ${1} is --noversion and then use shift. After that we can use
the positional arguments as usual so instead of ${arguments[0]} we can
use just ${1}.
Regards,
Petteri
signature.asc
Description: OpenPGP digital signature
Re: [gentoo-dev] Re: New eclass osgi.eclass
Thanks all for this feedback. I think this is sufficient to warrant
some modifications as suggested, and I propose to postpone the
inclusion in tree a little bit.
Regarding performance, there is an important point to note as IMHO it
is much more penalizing than using basename or such.
The fact that I use the jar utility to unzip a file, combined with the
fact that jar does not have an option to specify a destination
directory, is troublesome. I have to copy the .jar file to the
destination directory and then remove it. This .jar file can be
several MBs of course.
Initially I was using zip but I replaced it with jar in order to avoid
a dependency on zip. But maybe it's not worth the performance penalty.
Regards,
Elvanör
On Dec 6, 2007 1:11 PM, Petteri Räty <[EMAIL PROTECTED]> wrote:
> Steve Long kirjoitti:
> > Alistair Bush wrote:
> >> Im sure Elvanor can't wait for you constructive feedback on his eclass
> >> and depending on your feedback the eclass will enter the tree this
> >> weekend.
> >>
> > A couple of very minor performance points, which I think are more
> > significant in eclasses. Firstly the basename thing Donnie pointed out
> > before:
>
> Well usually you don't noticy any differences. In global scope things
> could have some effect of course.
>
> >
> > You have a couple of functions that take, say, 4 or 5 arguments. It would be
> > more robust to use a case, eg:
> > case $# in
> > 5)..;;
> > 4)..;;
> > *) die "Incorrect use of $FUNCNAME";;
> > esac
> > ..than if (($#>4)); then ..; else .. ;fi
> >
>
> Well imho they are equivalent and eclass maintainers should go with what
> they prefer.
>
> >
> > With regard to:
> > debug-print-function ${FUNCNAME} $*
> > if you want debug-print-function to get the arguments correctly, use "$@"
> > not $* (cf 'Special Parameters' in man bash for a proper explanation: this
> > applies to all arrays. http://wooledge.org/mywiki/BashFAQ/073 shows how to
> > manipulate all members of an array, eg to add a prefix.)
>
> Doesn't matter as arguments are just written to a file and for some
> reason this form is used all over the base in eclasses but shouldn't
> hurt to use "[EMAIL PROTECTED]" with new code.
>
>
> >
> > This use of counter in _java-pkg_osgijar-fromfile is odd:
> > while [[ -n "${1}" ]]; do
> > # while [[ $1 ]] or while (($#)) also work
> >if [[ "${1}" == "--noversion" ]]; then
> > noversion=1
> >else
> > arguments[${counter}]="${1}"
> > ((++counter))
> >fi
> >shift 1 # just shift?
> > done
> >
> > (([EMAIL PROTECTED] < 3)) && die "At least three arguments (not
> > counting --noversion) are needed for java-pkg_osgijar-fromfile()"
> >
> > You can either just add to the array with: arguments+=("$1")
> > or add using the counter: arguments[counter++]="$1"
> > and then check: ((counter < 3)) && die ..
> >
> > Arithmetic context[1] applies in array indexes, so you don't need to use a $
> > for variables and you can post/pre-incr etc there.
> > Yuu can also use it for C-style flags, with 0 as false and non-zero true:
> > if [[ "${noversion}" == 1 ]];
> > can be
> > if ((noversion));
> >
> > This is handy since unset variables evaluate as zero, which is false inside
> > ((..)), so a simple flag=1 is all that's needed, ie you don't have to set a
> > default, although ofc it's more robust to do so, especially for functions.
> > declare -i flag=0 # makes a local var which can only have integer values
> > assigned (setting it to string typically evaluates to zero; arithmetic
> > context is used for all assignments, so a string which happens to be a
> > variable with an integer will return that value.)
> >
> > [1] http://wooledge.org/mywiki/ArithmeticExpression
> >
>
> We really don't need to create the arguments array at all. We could just
> check if ${1} is --noversion and then use shift. After that we can use
> the positional arguments as usual so instead of ${arguments[0]} we can
> use just ${1}.
>
> Regards,
> Petteri
>
>
>
��í¢^����(� ��X��X�
Re: [gentoo-dev] Re: New eclass osgi.eclass
Jean-Noël Rivasseau kirjoitti: > > The fact that I use the jar utility to unzip a file, combined with the > fact that jar does not have an option to specify a destination > directory, is troublesome. I have to copy the .jar file to the > destination directory and then remove it. This .jar file can be > several MBs of course. > The jar utility supports writing to standard output and you can just use normal output redirection. See man jar. Regards, Petteri signature.asc Description: OpenPGP digital signature
Re: [gentoo-dev] Re: New eclass osgi.eclass
Hello all, I have a new version of the eclass ready, with much of the remarks addressed. It now goes by the name java-osgi and in the new form, should be ready to enter the tree. I fixed the performance problem I mentionned earlier, cleaned up the eclass API, and simplified the code almost everywhere. Waiting for your feedback. Cheers Elvanör java-osgi.eclass Description: Binary data
