[gentoo-dev] estack_{push,pop}: cool new helpers or over engineering?

[gentoo-dev] estack_{push,pop}: cool new helpers or over engineering?

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 4371 bytes --]

i needed to temporarily modify the umask in some vcs eclasses.  rather than
open coding the umask saving/restoring, i decided to re-use the
eshopts_{push,pop} logic so the umask can be pushed/popped easily.

the resulting code was mostly copy & paste the same, and the stack maintenance
ends up drowning out the meat of the stuff i care about -- screwing with the
umask.  so to that end, i added a set of generic stack helpers:
estack_{push,pop}.  then i rewrote eshopts_{push,pop} and based
eumask_{push,pop} on top of that.

what do people think ?  good stuff, or am i trying too hard ?
-mike

--- eutils.eclass	14 Dec 2011 17:36:18 -0000	1.372
+++ eutils.eclass	14 Dec 2011 22:23:02 -0000
@@ -100,6 +100,51 @@ esvn_clean() {
 	find "$@" -type d -name '.svn' -prune -print0 | xargs -0 rm -rf
 }
 
+# @FUNCTION: estack_push
+# @USAGE: <stack> [items to push]
+# @DESCRIPTION:
+# Push any number of items onto the specified stack.  Pick a name that
+# is a valid variable (i.e. stick to alphanumerics), and push as many
+# items as you like onto the stack at once.
+#
+# The following code snippet will echo 5, then 4, then 3, then ...
+# @CODE
+#		estask_push mystack 1 2 3 4 5
+#		while i=$(estack_pop mystack) ; do
+#			echo ${i}
+#		done
+# @CODE
+estack_push() {
+	[[ $# -eq 0 ]] && die "estack_push: incorrect # of arguments"
+	local stack_name="__ESTACK_$1__" ; shift
+	eval ${stack_name}+=\( \"\$@\" \)
+}
+
+# @FUNCTION: estack_pop
+# @USAGE: <stack>
+# @DESCRIPTION:
+# Pop a single item off the specified stack and return 0.  If no more
+# items are available, return 1.  See estack_push for more info.
+estack_pop() {
+	if [[ $# -ne 1 ]] ; then
+		# Would like to call `die` here, but people will usually
+		# be calling this in a subshell; e.g.
+		# val=$(estack_pop foo)
+		eerror "estack_pop: incorrect # of arguments"
+		return 1
+	fi
+
+	local stack_name="__ESTACK_$1__" ; shift
+	eval local i=\${#${stack_name}[@]}
+	# Don't warn -- let the caller interpret this as a failure
+	# or as normal behavior (akin to `shift`)
+	[[ $(( --i )) -eq -1 ]] && return 1
+
+	eval local s=\"\${${stack_name}[${i}]}\"
+	eval unset ${stack_name}[${i}]
+	echo "${s}"
+}
+
 # @FUNCTION: eshopts_push
 # @USAGE: [options to `set` or `shopt`]
 # @DESCRIPTION:
@@ -126,15 +171,14 @@ esvn_clean() {
 eshopts_push() {
 	# have to assume __ESHOPTS_SAVE__ isn't screwed with
 	# as a `declare -a` here will reset its value
-	local i=${#__ESHOPTS_SAVE__[@]}
 	if [[ $1 == -[su] ]] ; then
-		__ESHOPTS_SAVE__[$i]=$(shopt -p)
+		estack_push eshopts "$(shopt -p)"
 		[[ $# -eq 0 ]] && return 0
 		shopt "$@" || die "eshopts_push: bad options to shopt: $*"
 	else
-		__ESHOPTS_SAVE__[$i]=$-
+		estack_push eshopts $-
 		[[ $# -eq 0 ]] && return 0
 		set "$@" || die "eshopts_push: bad options to set: $*"
 	fi
@@ -144,19 +188,36 @@ eshopts_push() {
 # Restore the shell options to the state saved with the corresponding
 # eshopts_push call.  See that function for more details.
 eshopts_pop() {
-	[[ $# -ne 0 ]] && die "eshopts_pop takes no arguments"
-	local i=$(( ${#__ESHOPTS_SAVE__[@]} - 1 ))
-	[[ ${i} -eq -1 ]] && die "eshopts_{push,pop}: unbalanced pair"
-	local s=${__ESHOPTS_SAVE__[$i]}
-	unset __ESHOPTS_SAVE__[$i]
+	local s
+	s=$(estack_pop eshopts) || die # do not merge with `local` above
 	if [[ ${s} == "shopt -"* ]] ; then
 		eval "${s}" || die "eshopts_pop: sanity: invalid shopt options: ${s}"
 	else
 		set +$-     || die "eshopts_pop: sanity: invalid shell settings: $-"
 		set -${s}   || die "eshopts_pop: sanity: unable to restore saved shell settings: ${s}"
 	fi
 }
 
+# @FUNCTION: eumask_push
+# @USAGE: <new umask>
+# @DESCRIPTION:
+# Set the umask to the new value specified while saving the previous
+# value onto a stack.  Useful for temporarily changing the umask.
+eumask_push() {
+	estack_push eumask "$(umask)"
+	umask "$@" || die "${FUNCNAME}: bad options to umask: $*"
+}
+
+# @FUNCTION: eumask_pop
+# @USAGE:
+# @DESCRIPTION:
+# Restore the previous umask state.
+eumask_pop() {
+	local s
+	s=$(estack_pop eumask) || die # do not merge with `local` above
+	umask ${s} || die "${FUNCNAME}: sanity: could not restore umask: ${s}"
+}
+
 # @VARIABLE: EPATCH_SOURCE
 # @DESCRIPTION:
 # Default directory to search for patches.

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [gentoo-dev] estack_{push,pop}: cool new helpers or over engineering?

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 4236 bytes --]

some people pointed out typos/bugs, so here's v2
-mike

--- eutils.eclass	14 Dec 2011 17:36:18 -0000	1.372
+++ eutils.eclass	14 Dec 2011 23:46:37 -0000
@@ -100,6 +100,54 @@ esvn_clean() {
 	find "$@" -type d -name '.svn' -prune -print0 | xargs -0 rm -rf
 }
 
+# @FUNCTION: estack_push
+# @USAGE: <stack> [items to push]
+# @DESCRIPTION:
+# Push any number of items onto the specified stack.  Pick a name that
+# is a valid variable (i.e. stick to alphanumerics), and push as many
+# items as you like onto the stack at once.
+#
+# The following code snippet will echo 5, then 4, then 3, then ...
+# @CODE
+#		estack_push mystack 1 2 3 4 5
+#		while estack_pop mystack i ; do
+#			echo ${i}
+#		done
+# @CODE
+estack_push() {
+	[[ $# -eq 0 ]] && die "estack_push: incorrect # of arguments"
+	local stack_name="__ESTACK_$1__" ; shift
+	eval ${stack_name}+=\( \"\$@\" \)
+}
+
+# @FUNCTION: estack_pop
+# @USAGE: <stack> [variable]
+# @DESCRIPTION:
+# Pop a single item off the specified stack.  If a variable is specified,
+# the popped item is stored there.  If no more items are available, return
+# 1, else return 0.  See estack_push for more info.
+estack_pop() {
+	( [[ $# -eq 0 ]] || [[ $# -gt 2 ]] ) && die "estack_pop: incorrect # of arguments"
+
+	# We use the fugly __estack_xxx var names to avoid collision with
+	# passing back the return value.  If we used "local i" and the
+	# caller ran `estack_pop ... i`, we'd end up setting the local
+	# copy of "i" rather than the caller's copy.  The __estack_xxx
+	# garbage is preferable to using $1/$2 everywhere as that is a
+	# bit harder to read.
+	local __estack_name="__ESTACK_$1__" ; shift
+	local __estack_retvar=$1 ; shift
+	eval local __estack_i=\${#${__estack_name}[@]}
+	# Don't warn -- let the caller interpret this as a failure
+	# or as normal behavior (akin to `shift`)
+	[[ $(( --__estack_i )) -eq -1 ]] && return 1
+
+	if [[ -n ${__estack_retvar} ]] ; then
+		eval ${__estack_retvar}=\"\${${__estack_name}[${__estack_i}]}\"
+	fi
+	eval unset ${__estack_name}[${__estack_i}]
+}
+
 # @FUNCTION: eshopts_push
 # @USAGE: [options to `set` or `shopt`]
 # @DESCRIPTION:
@@ -126,15 +174,14 @@ esvn_clean() {
 eshopts_push() {
 	# have to assume __ESHOPTS_SAVE__ isn't screwed with
 	# as a `declare -a` here will reset its value
-	local i=${#__ESHOPTS_SAVE__[@]}
 	if [[ $1 == -[su] ]] ; then
-		__ESHOPTS_SAVE__[$i]=$(shopt -p)
+		estack_push eshopts "$(shopt -p)"
 		[[ $# -eq 0 ]] && return 0
 		shopt "$@" || die "eshopts_push: bad options to shopt: $*"
 	else
-		__ESHOPTS_SAVE__[$i]=$-
+		estack_push eshopts $-
 		[[ $# -eq 0 ]] && return 0
 		set "$@" || die "eshopts_push: bad options to set: $*"
 	fi
 }
 
@@ -144,19 +191,36 @@ eshopts_push() {
 # Restore the shell options to the state saved with the corresponding
 # eshopts_push call.  See that function for more details.
 eshopts_pop() {
-	[[ $# -ne 0 ]] && die "eshopts_pop takes no arguments"
-	local i=$(( ${#__ESHOPTS_SAVE__[@]} - 1 ))
-	[[ ${i} -eq -1 ]] && die "eshopts_{push,pop}: unbalanced pair"
-	local s=${__ESHOPTS_SAVE__[$i]}
-	unset __ESHOPTS_SAVE__[$i]
+	local s
+	estack_pop eshopts s || die "${FUNCNAME}: unbalanced push"
 	if [[ ${s} == "shopt -"* ]] ; then
 		eval "${s}" || die "eshopts_pop: sanity: invalid shopt options: ${s}"
 	else
 		set +$-     || die "eshopts_pop: sanity: invalid shell settings: $-"
 		set -${s}   || die "eshopts_pop: sanity: unable to restore saved shell settings: ${s}"
 	fi
 }
 
+# @FUNCTION: eumask_push
+# @USAGE: <new umask>
+# @DESCRIPTION:
+# Set the umask to the new value specified while saving the previous
+# value onto a stack.  Useful for temporarily changing the umask.
+eumask_push() {
+	estack_push eumask "$(umask)"
+	umask "$@" || die "${FUNCNAME}: bad options to umask: $*"
+}
+
+# @FUNCTION: eumask_pop
+# @USAGE:
+# @DESCRIPTION:
+# Restore the previous umask state.
+eumask_pop() {
+	local s
+	estack_pop eumask s || die "${FUNCNAME}: unbalanced push"
+	umask ${s} || die "${FUNCNAME}: sanity: could not restore umask: ${s}"
+}
+
 # @VARIABLE: EPATCH_SOURCE
 # @DESCRIPTION:
 # Default directory to search for patches.

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[gentoo-dev] Re: estack_{push,pop}: cool new helpers or over engineering?

[Steven J Long]

Just to point out that arithmetic context can be more efficient; no bugs, 
except for a /minor/ possibility (second last comment.)

Mike Frysinger wrote:
> --- eutils.eclass	14 Dec 2011 17:36:18 -0000	1.372
> +++ eutils.eclass	14 Dec 2011 23:46:37 -0000
> @@ -100,6 +100,54 @@ esvn_clean() {
>  find "$@" -type d -name '.svn' -prune -print0 | xargs -0 rm -rf
>  }
>  
> +# @FUNCTION: estack_push
> +# @USAGE: <stack> [items to push]
> +# @DESCRIPTION:
> +# Push any number of items onto the specified stack.  Pick a name that
> +# is a valid variable (i.e. stick to alphanumerics), and push as many
> +# items as you like onto the stack at once.
> +#
> +# The following code snippet will echo 5, then 4, then 3, then ...
> +# @CODE
> +#		estack_push mystack 1 2 3 4 5
> +#		while estack_pop mystack i ; do
> +#			echo ${i}
A minor #bash point in passing: although these values of i are safe, for 
tutorial code, I really would recommend quoting: echo "$i" (or "${i}"). It's 
better to get people used to quoting by default, and only not quoting iff 
they need field-splitting on parameter expansions (eg for a variable used 
for command options.)

> +#		done
> +# @CODE
> +estack_push() {
> +	[[ $# -eq 0 ]] && die "estack_push: incorrect # of arguments"
> +	local stack_name="__ESTACK_$1__" ; shift
> +	eval ${stack_name}+=\( \"\$@\" \)
> +}
((..)) is quicker than [[ .. ]] for arithmetic stuff, and usually easier to 
grok swiftly.
(($#)) || die .. is how this would normally be done.

> +
> +# @FUNCTION: estack_pop
> +# @USAGE: <stack> [variable]
> +# @DESCRIPTION:
> +# Pop a single item off the specified stack.  If a variable is specified,
> +# the popped item is stored there.  If no more items are available,
> return
> +# 1, else return 0.  See estack_push for more info.
> +estack_pop() {
> +	( [[ $# -eq 0 ]] || [[ $# -gt 2 ]] ) && die "estack_pop: incorrect 
# of arguments"

(($# == 0 || $# > 2)) && die.. # does it in one command, with no subshell.
[[ $# -eq 0 || $# -gt 2 ]] && die .. would work too, but more slowly.
In general if you want to do complex chains without a subshell, you would 
use: { } && .. instead of: ( ) && ..

TBH I would type (($#==0||$#>2)) in bash, though I space in C, where it 
doesn't affect execution time. But it's not as clear, especially if you're 
not in a highlighting editor.

> +	# We use the fugly __estack_xxx var names to avoid collision with
> +	# passing back the return value.  If we used "local i" and the
> +	# caller ran `estack_pop ... i`, we'd end up setting the local
> +	# copy of "i" rather than the caller's copy.  The __estack_xxx
> +	# garbage is preferable to using $1/$2 everywhere as that is a
> +	# bit harder to read.
> +	local __estack_name="__ESTACK_$1__" ; shift
> +	local __estack_retvar=$1 ; shift
> +	eval local __estack_i=\${#${__estack_name}[@]}
> +	# Don't warn -- let the caller interpret this as a failure
> +	# or as normal behavior (akin to `shift`)
> +	[[ $(( --__estack_i )) -eq -1 ]] && return 1
((--__estack_i == -1)) && ..

> +
> +	if [[ -n ${__estack_retvar} ]] ; then
> +		eval ${__estack_retvar}=\"\${${__estack_name}
[${__estack_i}]}\"
> +	fi
> +	eval unset ${__estack_name}[${__estack_i}]
> +}
> +
>  # @FUNCTION: eshopts_push
>  # @USAGE: [options to `set` or `shopt`]
>  # @DESCRIPTION:
> @@ -126,15 +174,14 @@ esvn_clean() {
>  eshopts_push() {
>  # have to assume __ESHOPTS_SAVE__ isn't screwed with
>  # as a `declare -a` here will reset its value
> -	local i=${#__ESHOPTS_SAVE__[@]}
>  if [[ $1 == -[su] ]] ; then
> -		__ESHOPTS_SAVE__[$i]=$(shopt -p)
> +		estack_push eshopts "$(shopt -p)"
>  [[ $# -eq 0 ]] && return 0
I'm not sure how this will ever match, given that $1 has been checked above?
(($#==1)) && return 0 # if that applies (might be a 'bug'.)

>  shopt "$@" || die "eshopts_push: bad options to shopt: $*"
>  else
> -		__ESHOPTS_SAVE__[$i]=$-
> +		estack_push eshopts $-
>  [[ $# -eq 0 ]] && return 0
(($#)) || return 0

>  set "$@" || die "eshopts_push: bad options to set: $*"
>  fi
>  }
>  

HTH,
Steve.
-- 
#friendly-coders -- We're friendly, but we're not /that/ friendly ;-)

Re: [gentoo-dev] Re: estack_{push,pop}: cool new helpers or over engineering?

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 1933 bytes --]

On Friday 16 December 2011 02:29:25 Steven J Long wrote:
> Mike Frysinger wrote:
> > +	[[ $# -eq 0 ]] && die "estack_push: incorrect # of arguments"
> 
> ((..)) is quicker than [[ .. ]] for arithmetic stuff, and usually easier to
> grok swiftly.

i'm not used to using this style, so for now i think i'll keep the existing 
(and it's more common in Gentoo atm, or at least in the code base i sample, 
ignoring the obvious selection bias).  i'll noodle on this though and see if i 
can't convince myself to start using this and live migrating code in the tree.

> (($#)) || die .. is how this would normally be done.

i think this is a little less clear, but considering some of the 
[advanced/complicated] bash code i've written elsewhere in Gentoo, maybe 
that's a specious argument ...

> >  eshopts_push() {
> >  	if [[ $1 == -[su] ]] ; then
> > -		__ESHOPTS_SAVE__[$i]=$(shopt -p)
> > +		estack_push eshopts "$(shopt -p)"
> >  		[[ $# -eq 0 ]] && return 0
> 
> I'm not sure how this will ever match, given that $1 has been checked
> above? (($#==1)) && return 0 # if that applies (might be a 'bug'.)

the larger idea when i first wrote eshopts_{push,pop} was to not do any arg 
parsing at all, but then i hit the issue that `shopt` and `set` options are 
like a venn diagram -- a lot of common stuff, but each also has unique options.  
so i had to introduce a little arg parsing to make it actually work.  and the 
duplicated/forked code in the sub-branches were kept as similar as possible to 
(hopefully) make things simpler to read at a glance.  but that means sharper 
eyes notice a code branch which can never be hit as you've highlighted here.

as for the $#==1 check at the top, i don't want these helpers to require 
arguments as it would prevent more creative uses.  such as:
	eshopts_push
	. ./some-script-from-a-package
	eshopts_pop
but maybe that's not a big deal ...
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]