Hangs in cygwin 3.5.5-1: should 3.5.5-1 be rolled back?

[Takashi Yano]

On Tue, 14 Jan 2025 08:57:36 +0900
Takashi Yano wrote:
> Hi Michael,
> 
> On Fri, 10 Jan 2025 08:48:00 +0100
> Michael Soegtrop wrote:
> > sending again, since this did not appear in the archives ...
> > 
> > Dear Cygwin Team,
> > 
> > I wanted to discuss the status of the hangs in cygwin 3.5.5-1 discussed 
> > in various threads here. I maintain the multi platform distribution of 
> > the Rocq/Coq Proof Assistant, and the day cygwin 3.5.5 was released (Dec 
> > 20) our CI broke. Success rate of individual jobs is now roughly 50% 
> > with half failing with hangs at random points in the build. I verified 
> > that between the last good and the first bad CI run only cygwin itself 
> > changed (no other packages).
> > 
> > I looked into this locally with procmon and to me it looks like a call 
> > to make is finished, shuts down its threads but then never terminates. I 
> > can reproduce this fairly easily - a few minutes after an initial setup 
> > of about 1/2 hour, which can be reused. But it is not reproducible in a 
> > single run - it is a random effect. To make it reproducible, I have to 
> > repeat a build step until it fails. But this always happens fairly soon.
> > 
> > IMHO cygwin 3.5.5 is severely broken. Either a fix for these hangs also 
> > others experience should be released soon, or 3.5.5 should be rolled 
> > back until it is fixed.
> > 
> > I definitely cannot make a release for Rocq/Coq Platform with cygwin 3.5.5.
> > 
> > I want to take the opportunity to emphasize that in general cygwin is 
> > very stable and such issues are rare - I am running a nightly CI of 
> > Rocq/Coq Platform - a complex project which takes hours to build - with 
> > the latest cygwin since about 7 years and there were only few issues in 
> > this time. Thank you for your continued efforts!
> > 
> > Do you think it would make sense to have a collection of CIs of large 
> > projects running daily on latest cygwin? If several of these break after 
> > a release of cygwin, this would give you an early warning that things 
> > might not be in good shape. I usually don't suspect cygwin if my CI 
> > breaks, so it usually takes me quite a while to figure this out. But if 
> > several projects break, this is more obvious and faster to analyze and 
> > to fix.
> 
> We are very sorry for inconvinience.
> Personally, I personally prefer releasing 3.5.6 ASAP. However, we
> are not sure that we have already fixed all the major problems in
> 3.5.5.
> 
> Can you please test latest cygwin 3.6.0 (TEST) whether the your
> CI issue still happens?


As a side note, you can install cygwin 3.6.0 (TEST) by specifying
the -t option to setup-x86_64.exe.

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>