Hi Ralf, > Hello Eric-1, ;-)
I blame nabble for that. Too bad that gmane doesn't know how to let people reply nicely to cross-posted threads. > > +# A little hack, as AT_CHECK_M4 already uses `stderr' to normalize it. > > +AT_CHECK_AUTOCONF([2>&1], [], [stdout]) > > > > - ' stderr]], [0],[$4])]) > > + ' stderr >&2]], [0], [], [$4])]) > > This triggers the multiple redirection bug on Ultrix (see > `(autoconf.info)File Descriptors'); I don't know if other, non-museum > systems are affected, too. But anyway we disallow it in the manual, so > iff we decide that this is ok now, then at least both those doc parts > need updates, too: My patch is no worse than yours at violating this rule, and we already violate it elsewhere in at least tests/torture.at. Does anyone still have access to Ultrix to see if there are some cases that work? The example in the manual only lists: $ eval 'echo matter >fullness' >void illegal io $ eval '(echo matter >fullness)' >void illegal io $ (eval '(echo matter >fullness)') >void Ambiguous output redirect. but one could argue that the first two examples involve no subshell, and in the third, because the subshell is invoking a shell builtin with no unquoted metacharacters in the argument, that Ultrix took a shortcut. Maybe it would work correctly if the two redirections of the same fd are in different levels of subshells, with the inner not being a shell builtin? Or is it not worth worrying about, as this would only cause spurious testsuite failures, but not impact autoconf proper? -- Eric Blake-1 ;) -- View this message in context: http://www.nabble.com/Re%3A-Fix-chdir-long.m4-caching-%28was%3A-megatest-fallout%29-tf2335952.html#a6534335 Sent from the Gnu - Autoconf - Patches mailing list archive at Nabble.com.
