Also reported upstream.

While working on reproducible builds for openSUSE, I found that
our pari-2.13.2 package build results varied between -j1 and -j4 builds.

From the logs I found that in the parallel build case, gp.o was compiled a 2nd time without our CFLAGS but instead with flags that come from 
pari-2.13.2/config/get_cc


[1]
difflog | grep -- -o.gp.o
- /usr/bin/gcc  -c -I. -I../src/headers -I../src/language -DDL_DFLT_NAME=NULL -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -Werror=return-type  -fno-strict-aliasing -o gp.o ../src/gp/gp.c
+ /usr/bin/gcc  -c -I. -I../src/headers -I../src/language -DDL_DFLT_NAME=NULL -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -Werror=return-type  -fno-strict-aliasing -o gp.o ../src/gp/gp.c
+ /usr/bin/gcc  -c -I. -I../src/headers -I../src/language -DDL_DFLT_NAME=NULL -O3 -Wall -fno-strict-aliasing -fomit-frame-pointer    -o gp.o ../src/gp/gp.c


pari-2.13.2/Olinux-x86_64/init.o also varied, likely from the same root cause.

This is probably some problem with dependencies in the build system.

We build thusly:
https://code.opensuse.org/package/pari/blob/08424e1789878449eb0ac61c5cafde2ff296a565/f/pari.spec#_104

That .o files are sometimes built without our hardening CFLAGS is a bug independent of reproducible builds.

The previous version 2.13.1 was building reproducibly today with the old spec file and current toolchain. But 2.13.1 also became unreproducible when building with the current .spec file.

This indicates that one of the .spec changes triggered the new problem. The diff is
https://code.opensuse.org/package/pari/c/08424e1789878449eb0ac61c5cafde2ff296a565

I think, the problem came from
make [...] all docpdf
because changing that to
make [...] all
make [...] docpdf

made the output reproducible again.

[1] https://github.com/bmwiedemann/reproducibleopensuse/blob/master/difflog