[Moving this subthread back to the autoconf mailing list, it's not relevant to glibc.] On Thu, Aug 27, 2020 at 2:22 PM Paul Eggert <eggert@xxxxxxxxxxx> wrote: > [...] it's better for FD 0 to be write-only and FDs > 1+2 to be read-only, as that's more likely to prevent buggy programs from > misbehaving further. For Autoconf, I suggest: > > (exec 3<&0) 2>/dev/null || exec 0>/dev/null > (exec 3>&1) 2>/dev/null || exec 1</dev/null > (exec 3>&2) || exec 2</dev/null I tested this. It makes configure scripts emit floods of "printf: write error" messages, and exit unsuccessfully despite having run the entire series of configuration tests, when fd 1 is closed on startup. This is because configure scripts frequently write to AS_MESSAGE_FD (which defaults to 1) without bothering to check whether it's writable first, but they do detect the write error and turn it into a delayed failure. (They finish successfully when run with --quiet.) I could argue that configure _should_ fail when fd 1 is closed on startup, unless --quiet is on the command line. However, it should do so _promptly_, rather than wasting people's time running all the tests first. Unfortunately I don't see a low-risk way to make that happen. So I think that for 2.70 we should make fd 0 read-only and 1,2 write-only here, and revisit this afterward -- when we're not in a release freeze we can think about things like turning on set -e mode. zw
