On 01 Feb 2018 22:35, Kyle Evans wrote: > On Thu, Feb 1, 2018 at 10:14 PM, Mike Frysinger <vapier@xxxxxxxxxx> wrote: > > On 10 Jan 2018 16:57, David Gibson wrote: > >> On Wed, Jan 10, 2018 at 12:47:56AM -0500, Mike Frysinger wrote: > >> > This changes from the bash-specific process substitution feature to > >> > reading with pipes. It relies on /dev/fd or /proc/self/fd existing. > >> > > >> > URL: https://crbug.com/756559 > >> > Signed-off-by: Mike Frysinger <vapier@xxxxxxxxxxxx> > >> > >> Yeah, sorry, replacing a dependency on a complex but widely available > >> shell with a much more cryptic dependency on a Linux specific feature > >> does not seem like a win to me. > > > > bash's <(...) only works when the host supports either /dev/fd or /proc/self/fd, > > so i don't think this is less portable. /dev/fd also isn't Linux-specific. > > For what it's worth, this isn't completely true. Bash can also use > FIFOs for process substitution [1]. FreeBSD does not use fdescfs by > default, but it can be mounted, and using fdescfs for process > substitution is provided as an optional feature in shells/bash. It > would be nice to not grow a dependency on fdescfs for dtc. =) > > [1] https://www.gnu.org/software/bash/manual/html_node/Process-Substitution.html sorry, you're right. i didn't fully trace the bash source. that man page matches the current sources in that either named fifos or /dev/fd are needed. Simon suggested i just use tempfiles myself. i had avoided them because the current method makes it easy to avoid leaks (for whatever reason), but if it isn't a big deal, that'd be an easy fix. -mike TMP1= TMP2= trap 'rm -f "${TMP1}" "${TMP2}"' EXIT TMP1=$(mktemp) TMP2=$(mktemp) source_and_sort "$1" "${TMP1}" source_and_sort "$2" "${TMP2}" diff -u "${TMP1}" "${TMP2}"
Attachment:
signature.asc
Description: Digital signature
