Fix R installation errors related to gfortran

[mkoeppe]

As reported

• for 9.0.rc1 in https://groups.google.com/d/msg/sage-release/YaedKjwMhzc/d44U0MM3BwAJ
• for 9.1.beta3 in https://groups.google.com/d/msg/sage-release/xxEfql2Isgg/bV-FQN8VFQAJ
• for 9.1.beta8 in https://groups.google.com/d/msg/sage-release/eMQIlhglN98/bTwDGD9HBQAJ

https://developer.r-project.org/Blog/public/2019/05/15/gfortran-issues-with-lapack/index.html

Upstream: Not yet reported upstream; Will do shortly.

CC: @dimpase @EmmanuelCharpentier @kiwifb @orlitzky @timokau

Component: packages: standard

Author: Matthias Koeppe

Branch/Commit: 5fbea80

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/29170

[dimpase]

comment:1

gcc fix here (linked from the R link) https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90329

not clear what they actually did - new option(s) added to gfortran?

[orlitzky]

comment:3

Ok, I read that whole thread so no one else has to. Ultimately, this is due to broken code that needs to be updated. The GCC developers have done two things:

• Added -fc-prototypes-external, to help people fix that broken code.
• In the meantime, they have added a workaround in the form of the (on by default) -ftail-call-workaround flag (https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html). That keeps the broken code working for now, but the option will eventually be turned off and go away.

The workaround was backported all the way back to gcc-7.x, so really what we should be doing is telling everyone to use a new-enough gfortran. The workaround is in gcc-9.2, and based on the release dates, I presume it's also in 7.5, 8.4, and everything newer than that.

We can probably just update gfortran's spkg-configure.m4 to ensure a new-enough version. Maybe for now it would suffice to check if the -ftail-call-workaround flag is supported. That would rule out older compilers from before things were broken, but is easy to implement.

[mkoeppe]

comment:4

See also: #29379 Upgrade R to 3.6.3 (which does not fix these errors)

[mkoeppe]

comment:5

See also: #29378 Update OpenBLAS to 0.3.9 (which also does not fix these errors)

[dimpase]

comment:6

Replying to @mkoeppe:

See also: #29378 Update OpenBLAS to 0.3.9 (which also does not fix these errors)

IIRC, the bugs in question (related to passing strings) are in "canonical" Fortran Lapack routines, and OpenBLAS does not amend Fortran code

[mkoeppe]

Description changed:

--- 
+++ 
@@ -1,6 +1,7 @@
 As reported 
 - for 9.0.rc1 in https://groups.google.com/d/msg/sage-release/YaedKjwMhzc/d44U0MM3BwAJ
 - for 9.1.beta3 in https://groups.google.com/d/msg/sage-release/xxEfql2Isgg/bV-FQN8VFQAJ
+- for 9.1.beta8 in https://groups.google.com/d/msg/sage-release/eMQIlhglN98/bTwDGD9HBQAJ
 
 https://developer.r-project.org/Blog/public/2019/05/15/gfortran-issues-with-lapack/index.html
 

[mkoeppe]

comment:9

Our R 3.6.2 already contains checks in configure that are intended to address the problem with the Fortran ABI. On one of the failing platforms, however, I see:

[r-3.6.2.p0] checking if need -fno-optimize-sibling-calls for gfortran... yes
[r-3.6.2.p0] checking for type of 'hidden' Fortran character lengths... 
[r-3.6.2.p0] checking for xmkmf... no

which looks suspicious (note - no result shown for the type)

[mkoeppe]

comment:10

from config.log there:

configure:20320: checking for type of 'hidden' Fortran character lengths
/usr/bin/ld: conftestf.o: relocation R_X86_64_32 against `.rodata' can not be used when making a PIE object; recompile with -fPIC
/usr/bin/ld: final link failed: nonrepresentable section on output
collect2: error: ld returned 1 exit status
configure:20384: result: 

[dimpase]

comment:11

bug in R's ./configure ?

[mkoeppe]

comment:12

In any case missing error handling. I'm on it

[mkoeppe]

comment:13

Adding -fPIC to CFLAGS & FCFLAGS seems to do the trick.

[mkoeppe]

Branch: u/mkoeppe/fix_r_installation_errors_related_to_gfortran__or_downgrade_r_to__experimental_

[mkoeppe]

Author: Matthias Koeppe

[mkoeppe]

Commit: 5fbea80

[mkoeppe]

New commits:

5fbea80

build/pkgs/r/spkg-install.in: Work around a failing R configure check for hidden Fortran character lengths