On Tue, Oct 7, 2014 at 3:05 PM, Dave Mitchell <davem@iabyn.com> wrote: > On Mon, Oct 06, 2014 at 09:38:11PM +0200, Ævar Arnfjörð Bjarmason wrote: >> On Mon, Oct 6, 2014 at 2:23 PM, Dave Mitchell <davem@iabyn.com> wrote: >> > On Sun, Oct 05, 2014 at 07:47:24PM +0200, Ęvar Arnfjörš Bjarmason wrote: >> >> I got access to an IBM pSeries POWER7 Fedora Linux box. It's currently >> >> failing these tests on blead: >> >> >> >> * io/eintr.t >> >> * op/sprintf2.t >> >> * porting/libperl.t >> >> >> >> It seems the first is skipped in v5.20.1 >> > >> > io/eintr.t is skipped on non-devel builds. It tests what happens when a >> > sig handler does IO (e.g. closing a file handle) in the middle of IO (it >> > used to SEGV). On many plaforms such IO isn't interruptible, so the test >> > can't be done. Looks like your system is one of those. There's a big >> > 'skip_all' near the top of the test file that presumably needs tweaking. >> >> It's a recent Linux system, I don't see why stuff like that wouldn't >> work just because you're on a different processor. > > Ah, sorry; I completely failed to spot the obvious in your OP. > >> -$buf = "a" x 1_000_000 . "\n"; # bigger than any pipe buffer hopefully >> +$buf = "a" x 1_000_000_000 . "\n"; # bigger than any pipe buffer hopefully > > Interesting. Are you able to find a smaller value that still blocks? > I should imagine a 1Gb string might choke small systems. The magic number that makes this test pass is 0x100000. I just pushed 1e02895 to make the test use that.Thread Previous | Thread Next