This bug is worse than I expected.
I have a set of ten packages affected by this problem. Every time I
test them, I get a different subset of reproducible and unreproducible
packages among them. This is the data I collected today:
aiozmq No No Yes
debmake No No No
doit No Yes Yes
npm2deb No Yes No
pafy Yes No No
pudb Yes Yes No
pygrib No Yes No
python-3to2 Yes Yes No
python-odf No Yes Yes
python-softlayer No No No
The first column of Yes/No was this morning. The second column was
today around 14:20 UTC, the third one was today around 19:20 UTC.
"Yes" means "the package was built twice and in both cases the created .deb
packages were byte-by-byte identical".
"No" means "the package was built twice and the .deb packages were
So, this still seems a bug in dh-python to me, but there should be
also some "random" ingredient in its inner working or in the affected
packages (parallel build?) to explain why dh-python does not seem to
Reproducible-builds mailing list