While working on reproducible builds for openSUSE, I found that
semi-xemacs randomly (20-50% of the time) had variations 
in custom-load.el
or custom-load.elc
even when disabling all build variations.


Steps to Reproduce:
osc co openSUSE:Factory/semi-xemacs && cd $_
osc build --noservice
osc chroot --offline
cd /home/abuild/rpmbuild/BUILD/semi-1.14.6
rm custom-load.el
xemacs -batch -q -no-site-file -l SEMI-MK -f compile-semi-package &&
  md5sum custom-load.el

and then run the last command again


Actual Diff:

+(custom-add-loads 'mime-view '("mime-play" "mime-view"))
+(custom-add-loads 'pgg-parse '("pgg-parse"))
+(custom-add-loads 'smime '("smime"))
+(custom-add-loads 'mime-edit '("mime-edit"))
+(custom-add-loads 'pgg '("pgg-def" "pgg-gpg" "pgg-parse" "pgg-pgp" "pgg-pgp5"))
+(custom-add-loads 'pgg-gpg '("pgg-gpg"))
+(custom-add-loads 'pgg-pgp5 '("pgg-pgp5"))
+(custom-add-loads 'mime '("mime-edit" "mime-view" "pgg-def" "semi-def" "smime"))
+(custom-add-loads 'pgg-pgp '("pgg-pgp"))

and a long hash line having 23 vs 24 elements.

Could this diff cause bugs at runtime?

In the simplified reproducer above, the result remains the same after the 2nd call until the .el file is removed again. This hints at a race condition occuring during the build, even in a 1-core VM.