Created attachment 238668 [details] [review]
Patch for the bug

I suppressed error messages of completion. Please review it.

I'm not really convinced. The error message is telling you about a mistake in your commandline, after all. Might as well get it early, rather than completing things and then learn that your command won't work after all.

As you say, errors might make users aware of their mistake. But I think that shell commands, not only gsettings, should not interfere in users when they are typing. They don't expect errors will appear while typing. They don't know that the completion system internally runs gsettings. Probably they will be confused about what happened by sudden errors.

In addition, some parts have already suppress errors:
gio/completion/gsettings:
 48       case "${COMP_WORDS[$(($coffset+1))]}" in
 49         set)
 50           choices="$(gsettings $schemadir list-keys ${COMP_WORDS[$(($coffset+2))]} 2> /dev/null | sed -e 's.$. .')"
 51           ;;
 52 
 53         get|range|reset|writable|monitor)
 54           choices="$(gsettings $schemadir list-keys ${COMP_WORDS[$(($coffset+2))]} 2> /dev/null)"
 55           ;;
 56       esac
 57       ;;
 58 
 59     4)
 60       case "${COMP_WORDS[$(($coffset+2))]}" in
 61         set)
 62           range=($(gsettings $schemadir range ${COMP_WORDS[$(($coffset+2))]} ${COMP_WORDS[$(($coffset+3))]} 2> /dev/null))

I agree with Jiro Matsuzawa; I think it’s appropriate to squash the error messages while doing command-line completion, since it basically breaks the completion. Nobody can really be expected to change their command line on the fly as a result of an interposed error message in the completion.

I’ll push the patch.