On 5 June 2015 at 23:43, Karl Berry <[email protected]> wrote: > The best option, I think, is to run "stty sane" at the end. > > Sorry to make things more complicated, but ... if we are running under > that shell on that system, I think ... changing terminal settings when > it's unnecessary does not sound good.
Better to upset whatever terminal settings the user had beforehand than to leave them apparently unable to type anything in. (Running the test script under ksh itself, instead of under bash, is even worse: every keystroke is greeted with "continued >". I had to disconnect my ssh session to recover.) > Another thought: if under this shell+system, exec some other shell. I > had to do this in run_parser_all.sh a few days ago in order to make the > tests workable on Solaris 5.10 (sparc anyway); see comments at top. I'll have a look at it, but I don't know how we're supposed to check if the shell is broken or not. > Another thought: ask Dagobert for advice. Maybe there is some special > setting needed to get rid of UTF-8. > > (Also, I would think that all tests > should be run in the C locale anyway -- I seem to recall you were doing > that already.) Yes, they are meant to be. > According to the ksh93 change log on > http://www2.research.att.com/sw/download/, someone fixed this bug on > 22-9-2011. > > Unfortunately, that doesn't mean much. Upstream fixes may or may not > ever make it into distros. Especially Solaris. E.g., I would hope that > Dagobert's system is up to date wrt system patches. The date from ksh --version is 2011-02-08, so that would be from before that patch. (I couldn't replicate the problem with the ksh on my own system, which is the one with Slackware 14.1 - that one is slightly newer, at 2011-08-01.) Gavin
