On 25/09/2026 07:51, Thomas Klausner wrote:
On Thu, Sep 24, 2026 at 05:17:28PM +0100, Collin Funk wrote:
Pádraig Brady <[email protected]> writes:
On 24/09/2026 22:18, Paul Eggert wrote:
On 2026-09-24 07:52, Collin Funk wrote:
It has been a while since I looked at pipe-f2.sh, but I remember coming
to the conclusion that it was a NetBSD bug [1]
[1]https://lists.gnu.org/archive/html/coreutils/2026-04/msg00129.html
There do seem to be problems in that area. See NetBSD PR kern/57659
<https://gnats.netbsd.org/57659>, along with test cases added yesterday
(!) trying to catch a similar bug with fifos and ptys, e.g.,
<https://github.com/NetBSD/src/commit/abad86f745946a5718e2b9c6559c7513f56edc01>.
Interesting. I also see the following which seems like it might be related:
https://github.com/NetBSD/src/commit/b736ec86d
I've not looked at those in detail,
but I suspect that may have regressed against the POSIX
requirement for fifos, that...
“If no process has the pipe open for writing, read() shall return 0 to indicate
end-of-file.”
I.e. the subsequent blocking after the first read() returning EOF should not
happen.
Yes, that was the conclusion that I came to a few months ago it seems
[1].
Collin
[1] https://lists.gnu.org/archive/html/coreutils/2026-04/msg00129.html
Sorry, I'm confused by the timing here.
Do you (Pádraig) think the patch in
https://github.com/NetBSD/src/commit/b736ec86d is a POSIX regression?
Or do you (Collin) think it fixes the problem you saw months ago? (The
patch is three days old, and I'm not running it on my machine yet.)
Thomas
Sorry I was confused.
Checking my old test logs that I made against NetBSD 10,
shows that I did actually hit the issue there.
Also I assumed the fact you were on 11.99 meant you were testing trunk
and it seemed too coincidental that those related patches landed
in the last few days.
But if you're not running with those patches, then it does
in fact seem to be a coincidence,
So those new patches may in fact fix the issue,
but as mentioned previously I'll also make the patch to GNU tail
as it seems safe and is more defensive/consistent.
cheers,
Padraig
