external/Makefile race-condition

[jsarenik]

Issue and Steps to Reproduce

The issue is already mentioned in #1069 and I noticed it a couple of times in the past. Now I have a 100% reproduction on both Alpine Linux and Ubuntu 18.04. A full log with the error follows, then a full log with the same thing, just with extra --recurse-submodules additional parameter to the git clone line - this fixes all (and is the same as cloning without it and then cd into cloned directory and calling git submodule update --init):

Error full log

https://gist.github.com/jsarenik/98636bd46547a47dda959c4cdf8748d8#file-failure-txt

Success full log

https://gist.github.com/jsarenik/98636bd46547a47dda959c4cdf8748d8#file-success-txt

Proposed fix

Get rid of git commands called by make: #1542

getinfo output

Not related :-) We're not there yet.

[jsarenik]

So I did a git-bisect on Ubuntu 18.04. Here is one of the results:

cc1: all warnings being treated as errors
<builtin>: recipe for target 'gossipd/gossip.o' failed
make: *** [gossipd/gossip.o] Error 1
make: *** Waiting for unfinished jobs....
make: *** wait: No child processes.  Stop.
94581c6e0725575e490ed84c4a028112d263b6fb is the first bad commit
commit 94581c6e0725575e490ed84c4a028112d263b6fb
Author: Rusty Russell <rusty@rustcorp.com.au>
Date:   Mon Jun 4 13:52:25 2018 +0930

    gossipd: wire up infrastructure to generate query_short_channel_ids msg.
    
    Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

:040000 040000 09de552302c29346a6e6793a36af5a449055c8f5 18bb780f5ec650eea4d65eed3c170fa3807fcbe1 M      gossipd
:040000 040000 e7b93ef2648d0a8aff6f3f65a3f137d36a403cfc aefd9250c4b30a0225b314388fe75264ed6ab900 M      lightningd
bisect run success

I did it like this:

cat > ~/bisect.sh <<EOF
#!/bin/sh

{
git clean -xfd
git submodule deinit --all -f
./configure || true
make -j4
} >/dev/null 2>&1 && echo Success || { echo FAIL; exit 1; }
EOF
chmod a+x ~/bisect.sh
git clone https://github.com/ElementsProject/lightning
cd lightning
git remote add rusty https://github.com/rustyrussell/lightning
git fetch rusty
git checkout guilt/configure
git bisect start HEAD 27a186b --
git bisect run ~/bisect.sh
git bisect reset

Your mileage may vary. I still think this is a race condition so the bisect is not really showing when it was introduced and will oscillate on multiple commits. But one thing is clear to me, the latest commit on rustyrussel/lightning@guilt/configure is failing a lot. Try it please.

I used debootstrap to set up a fresh Ubuntu 18.04 chroot directory and chsys script to get into it.

mkdir ubuntu18
debootstrap --variant=buildd --arch=amd64 \
  bionic ./ubuntu18 http://archive.ubuntu.com/ubuntu/
./chsys ubuntu18 /bin/bash
# then as root in chroot
apt install autoconf automake build-essential git libtool libgmp-dev libsqlite3-dev python python3 net-tools zlib1g-dev

[jsarenik]

rustyrussell@5d77c6a fixes the issue. Thank you! Closing.