Re: [gentoo-dev] RFC: sed script redundancy

[Christopher Schwan]

Thank you for that script. I experimented a bit with it and have a number of 
corrections and suggestions:

- alias does not work because my_sed is not declared at this stage. I removed 
the whole alias line because I want to selectively enable my_sed
- oargs must be an array in order to make quoting work:

   local oargs=( "${@}" )

- In the ewarn line ${oargs} should be changed to ${nargs[@]} (!?)
- is it correct to treat -e and -f alike ? I am not sure about that, because 
the latter expects a file
- If no "-e" is given, the first non-option argument is treated as the sed-
script-expression, therefore I added hade=yes in the if-branch

The new function now reads:

my_sed() {
	local oargs=( "$@" )
	local arg
	local nargs=()
	local hadi=
	local hade=

	while [[ -n $1 ]] ; do
		case "$1" in
		-i|--in-place)
			# ignore this flag
			hadi=yes
			;;
		-e|--expression)
			shift
			nargs+=( "-e" "$1" )
			hade=yes
			;;
		-f|--file)
			shift
			nargs+=( "-f" "$1" )
			hade=yes
			;;
		-*)
			nargs+=( "$1" )
			;;
		*)
			if [[ -z ${hade} ]] ; then
				nargs+=( "-e" "$1" )
				hade=yes
			elif [[ -z ${hadi} ]] ; then
				# there is no inline replacing, not much we can do
				break
			else
				sed "${nargs[@]}" "$1" | diff -q "$1" - > /dev/null && \
					ewarn "sed ${nargs[@]} has no effect on $1" 
			fi
			;;
		esac
		shift
	done

	sed "${oargs[@]}"
}

As you can see, I added support for long-options. However, testing the 
individual sed commands remains to be done. This could be especially difficult 
if input is taken from stdin (e.g. in cat foo | sed "s:a:b:g").

I tested my_sed within our sage ebuild[1]. This ebuild contains 39 sed 
commands and I was able to spot one useless sed.



[1] https://github.com/cschwan/sage-on-gentoo/blob/master/sci-
mathematics/sage/sage-4.7.ebuild


On Sunday 22 May 2011 12:50:43 Fabian Groffen wrote:
> On 21-05-2011 19:34:34 +0200, Jeroen Roovers wrote:
> > On Fri, 20 May 2011 17:56:00 +0200
> >
> > Fabian Groffen <grobian@g.o> wrote:
> > >   sed -e "<pattern>" "${file}" | diff "${file}" -
> > >
> > > followed by the actual sed -i -e ...
> > >
> > > This way I didn't need to write an intermediate file.
> >
> > The problem there is that sed might be called just once on any one file,
> > but in the tree it is often invoked with multiple scripts, so this
> > simple implementation lacks a way to evaluate which sed scripts are
> > useful.
> >
> > Also, how do I ensure the sed replacement works only on invocations
> > inside the ebuild, and not, say, in portage's internals?
>
> (not tested, but as proof of concept)
>
> alias sed my_sed
> my_sed() {
> 	local oargs="${@}"
> 	local arg
> 	local nargs=()
> 	local hadi=
> 	local hade=
> 	while [[ -n $1 ]] ; do
> 		case "$1" in
> 			-i)
> 				# ignore this flag
> 				hadi=yes
> 				;;
> 			-e|-f)
> 				shift
> 				nargs+=( "-e$1" )
> 				hade=yes
> 				;;
> 			-*)
> 				nargs+=( "$1" )
> 				hade=yes
> 				;;
> 			*)
> 				if [[ -z ${hade} ]] ; then
> 					nargs+=( "$1" )
> 				elif [[ -z ${hadi} ]] ; then
> 					# there is no inline replacing, not much we can do
> 					break
> 				else
> 					sed "${nargs[@]}" "$1" | diff -q "$1" - > /dev/null \
> 						&& ewarn "sed ${oargs} has no effect on $1"
> 				fi
> 				;;
> 		esac
> 		shift
> 	done
>
> 	\sed "${oargs}"
> }