Bug#991451: redis breaks python-fakeredis autopkgtest: AssertionError

[Jochen Sprickerhof]

* Chris Lamb <la...@debian.org> [2021-07-25 16:37]:

Chris Lamb wrote:

Sure thing -- I've forwarded this upstream here:

  https://github.com/redis/redis/issues/9273

Okay, so the latest reply there suggests that this is (now) the
expected and behaviour of Redis going forward.

I still don't quite grasp what it is that fakeredis is testing though,
so I can't state with any authority whether it definitely is fakeredis
that needs to be addressed, but reverting this behavioural change in
Redis does not seem the right way to go at all (!).

--- test/test_hypothesis.py.org 2021-07-26 07:54:36.151227815 +0200
+++ test/test_hypothesis.py     2021-07-26 07:54:38.599225613 +0200
@@ -257,8 +257,8 @@
 set_commands = (
     commands(st.just('sadd'), keys, st.lists(fields,))
     | commands(st.just('scard'), keys)
-    | commands(st.sampled_from(['sdiff', 'sinter', 'sunion']), st.lists(keys))
-    | commands(st.sampled_from(['sdiffstore', 'sinterstore', 'sunionstore']),
+    | commands(st.sampled_from(['sdiff', 'sunion']), st.lists(keys))
+    | commands(st.sampled_from(['sdiffstore', 'sunionstore']),
                keys, st.lists(keys))
     | commands(st.just('sismember'), keys, fields)
     | commands(st.just('smembers'), keys)


Cheers Jochen

I have no idea about Redis/Fakeredis, adding Ondřej as he did all the uploads, lately. What I got from looking at the code was that Hypothesis is a library to do extensive testing, so testing with 0 or empty strings sounds like a common way. I haven't looked how much work it would be to adopt Fakeredis to the new Redis behavior, but in case we want to only silence the tests for bullseye this would be a minimal change:

signature.asc
Description: PGP signature