OpenBLAS build failure with MAKEFLAGS='-j4 -l4'

[zimmermann6]

I get an error when compiling Sage 8.9 from source on my workstation (Intel(R) Core(TM) i5-4590 under Debian GNU/Linux bullseye/sid). The compilation of openblas fails (see attached file):

/usr/bin/ld: cannot find -l4

Upstream (OpenBLAS) issue:

• Build with MAKEFLAGS='-j4 -l4' OpenMathLib/OpenBLAS#2673

Depends on #29913

Upstream: Fixed upstream, in a later stable release.

CC: @dimpase @slel

Component: packages: standard

Reviewer: Samuel Lelièvre

Issue created by migration from https://trac.sagemath.org/ticket/28768

[embray]

comment:1

Attachment: openblas-0.3.6.p0.log

Ticket retargeted after milestone closed

[dimpase]

comment:3

this looks very obscure.

How about using Sage 9.0 or even 9.1betas with openblas from the system?

[embray]

comment:4

Googling, and the only reference I can find to -l4 would be if you passed the flag -l4 to make, in which case if you got something like this it would be some kind of flag confusion. I can't imagine otherwise what this would mean.

[embray]

comment:5

I'm also looking at the openblas build system and the spkg-install and I can't for the life of me figure out where this could be coming from. Can you post your full environment?

[zimmermann6]

comment:7

Replying to @embray:

I'm also looking at the openblas build system and the spkg-install and I can't for the life of me figure out where this could be coming from. Can you post your full environment?

Erik, I guess you found it. I have in my environment:

export MAKEFLAGS="-j4 -l4"

which means to use 4 threads for make, but with a global limit of 4 for the cpu load.
I guess the MAKEFLAGS variable is used somewhere in the Sage and/or OpenBLAS makefile,
and not properly protected.

Paul

[zimmermann6]

comment:8

I did extract the OpenBLAS-0.3.6 tarball shipped with Sage 8.9, did "make" in upstream/OpenBLAS-0.3.6, and the same issues appears:

gfortran -O2 -Wall -frecursive -m64  -mavx2  -o sblat1 sblat1.o ../libopenblas_haswellp-r0.3.6.a -lm -lpthread -lgfortran -lm -lpthread -lgfortran -L/usr/lib/gcc/x86_64-linux-gnu/9 -L/usr/lib/gcc/x86_64-linux-gnu/9/../../../x86_64-linux-gnu -L/usr/lib/gcc/x86_64-linux-gnu/9/../../../../lib -L/lib/x86_64-linux-gnu -L/lib/../lib -L/usr/lib/x86_64-linux-gnu -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/9/../../..  -l4 -lc  
/usr/bin/ld: cannot find -l4

Thus I believe the issue is due to upstream.

[zimmermann6]

comment:9

I've reported the issue to the openblas developers. They say:

Thank you for the report.

I think we didn't test MAKEFLAGS="-j4 -l4" case.

[zimmermann6]

Upstream: Reported upstream. Developers acknowledge bug.

[mkoeppe]

comment:10

We upgraded openblas to 0.3.9, is it fixed?

[zimmermann6]

comment:11

the issue is still present with openblas 0.3.9 (downloaded from https://github.com/xianyi/OpenBLAS/releases).

[zimmermann6]

comment:12

please can someone with a github account create the issue on
https://github.com/xianyi/OpenBLAS/issues ?

[dimpase]

comment:13

comment:9 says that a report has already been made - surely, if a github issue is to be filed, it would be useful to mention this.

With all respect, setting a gihub account takes a couple of minutes, what's the point of contacting them via an intermediary? Is it forbidden in INRIA?

[zimmermann6]

comment:14

it is just that I'm reluctant to open a github account for personal reasons. Indeed I reported the issue to Zhang Xianyi on January 17, and sent him a reminder today. His answer was that one could create an issue on github.

[slel]

comment:16

I opened a ticket at OpenBLAS on Paul's behalf.

• Build with MAKEFLAGS='-j4 -l4' OpenMathLib/OpenBLAS#2673

[zimmermann6]

comment:17

thank you Samuel!

[slel]

Changed upstream from Reported upstream. Developers acknowledge bug. to Fixed upstream, in a later stable release.

[slel]

Reviewer: Samuel Lelièvre

[slel]

Dependencies: #29913

[slel]

comment:18

There was already a relevant ticket and fix upstream:

• Don't parse gcc -v / gfortran -v blindly OpenMathLib/OpenBLAS#2616
• Add workaround for unhandled gmake jobserver flags in c_check/f_check OpenMathLib/OpenBLAS#2617

The fix is part of OpenBLAS 0.3.10 released last week.

Paul confirmed to me he could no longer observe
the problem with OpenBLAS 0.3.10.

Upgrading to OpenBLAS 0.3.10 is now tracked at #29913.