DMARC-Filter: OpenDMARC Filter v1.4.2 delorie.com 5B5IYVmL1411676
DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 2142C4C3188B
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3826.700.81.1.3\))
Subject: Re: Sporadic GNU make jobserver unavailable
Date: Fri, 5 Dec 2025 19:33:35 +0100
References: <7D93A1F2-C293-4289-94D6-B8E0C3458CF9 AT unified-streaming DOT com>
 <1ae0d609-dd05-48aa-b2bf-c3ccf39b86da AT SystematicSW DOT ab DOT ca>
 <3AC1B525-423D-4EC2-91AC-CB0E4AA01375 AT unified-streaming DOT com>
To: cygwin AT cygwin DOT com
In-Reply-To: <3AC1B525-423D-4EC2-91AC-CB0E4AA01375@unified-streaming.com>
Message-Id: <152C1323-5BAA-4AE1-A925-33A0360B0B7E@unified-streaming.com>
Precedence: list
From: Dimitry Andric via Cygwin <cygwin AT cygwin DOT com>
Reply-To: Dimitry Andric <dimitry AT unified-streaming DOT com>
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Errors-To: cygwin-bounces~archive-cygwin=delorie DOT com AT cygwin DOT com
Sender: "Cygwin" <cygwin-bounces~archive-cygwin=delorie DOT com AT cygwin DOT com>

On 5 Dec 2025, at 19:26, Dimitry Andric <dimitry AT unified-streaming DOT com> wrote:
> 
> On 5 Dec 2025, at 14:37, Brian Inglis via Cygwin <cygwin AT cygwin DOT com> wrote:
>> 
>> 
>> On 2025-12-05 02:18, Dimitry Andric via Cygwin wrote:
>>> We're using Cygwin make to build several projects in parallel mode, e.g.
>>> with -j 4 or sometimes higher settings, depending on the core count of
>>> the machine.
>>> Sporadically, and then I'm talking about "once a week" or so, random
>>> jobs will fail with a non-repeatable error, and in _every_ case these
>>> failures are preceded by a warning from make:
>>>  make[4]: warning: jobserver unavailable: using -j1.  Add '+' to parent make rule.
>>> Obviously the "Add '+' to parent rule" is nonsensical, since simply
>>> running exactly the same job again almost always works fine. But there
>>> is something flaky about make's jobserver under Windows.
>>> FWIW one machine I just observed this on has the make-4.4.1-2 package,
>>> which should use the "pipe" jobserver, if I interpret
>>> https://cygwin.com/pipermail/cygwin-announce/2023-March/010972.html
>>> correctly.
>>> Does anybody know a way to further debug this issue? It must be some
>>> rare race condition...
>> 
>> https://stackoverflow.com/questions/60702726/warning-jobserver-unavailable-using-j1-add-to-parent-make-rule
> 
> Yeah, I've read all those FAQs. The problem isn't in our Makefiles,
> since they run just fine on Linux. It's only on Windows under Cygwin
> that they break. :)
> 
> I'm fairly certain it's a make bug, but whether it's Windows specific,
> or just a Cygwin issue, is not yet clear.

I just read https://cygwin.com/pipermail/cygwin-announce/2023-March/010972.html,
which says:

> The default version [4.4.1-1] is built with jobserver using FIFO and
> the test version [4.4.1-2] using pipe.
> In same complex corner case the previous 4.4-1 was reported to
> fail and it is unclear if the pipe version is the solution
> or just a side coincidence.
> Please report any problem on any version.

From Marco's post it is not really clear what the failure was, but our
Windows systems all run 4.4.1-2 since that is the latest version, so
I'll try reverting some boxes to 4.4.1-1 to see if that helps.

-Dimitry


-- 
Problem reports:      https://cygwin.com/problems.html
FAQ:                  https://cygwin.com/faq/
Documentation:        https://cygwin.com/docs.html
Unsubscribe info:     https://cygwin.com/ml/#unsubscribe-simple