archive: mark EMetaBlob denc incompatible release before 60730271d4
in
https://github.com/ceph/ceph/commit/60730271d49a3c16bee6ff60c402b924a0280efc,
the encoding schema version of EMetaBlob bumped from 4 to 5. the older
release containing this change was v0.58
in
https://github.com/ceph/ceph/commit/d62c67ca2041797fc7aa212ae4a37ba91e5620e7,
the struct_v of EMetaBlob::fullbit::encode was bumped from 7 to 8, , the
oldest release containing this change was hammer rc1, i.e., v0.93.
in
https://github.com/ceph/ceph/pull/35931/commits/0127f2b01928b515bbbb7811976012ec97d3a957,
the backward compatible code of EMetaBlob::fullbit was dropped, so it
does not decode any EMetaBlob::fullbit whose struct_v belongs to [2, 8)
anymore.
in this change, all EMetaBlob* corpus created before 9.3 or 0.93 are
marked forward incompatible, so the denc tests with them are skipped if
the dencoder version is higher than those versions.
this change helps us to drop the backward compatible code to decode
EMetaBlob. as it is not neeed anymore after 2 releases.
Sage Weil [Mon, 27 Feb 2017 21:32:54 +0000 (16:32 -0500)]
add ScrubMap[::object] incompat at 10.2.0
We're removing decoding ability for older instances because we no longer
need to read anything pre-jewel (this is an on-wire message and upgrades
pass through jewel).
Sage Weil [Mon, 27 Feb 2017 21:22:52 +0000 (16:22 -0500)]
drop 0.46-add-osd-uuid/objects/OSDMap/9
This is the one OSDMap instance with stray trailing data, which as of 3f3a72b42e6410c13dc98d6b05f53ec9da93c4fe ceph-dencoder no longers likes
for OSDMap. Easier to remove this ancient test instance than add the new
macros needed for STRAYDATA + NOCOPY.
Sage Weil [Sun, 11 Jan 2015 05:18:21 +0000 (21:18 -0800)]
drop coll_t 'foo' and 'bar'
These are no longer valid, but we want to keep testing all of the other
old encoded instances. Just drop them since we don't care about whether
they decode in real clusters.
Sage Weil [Mon, 13 Feb 2012 20:40:17 +0000 (12:40 -0800)]
note forward_incompat changes from osd_peer_stat_t at 0.41-306-gfbbbd01
osd_peer_stat_t isn't in the old corpus, but it was contained by messages
that used it; mark those as forward_incompat (meaning new code won't
understand the old data).
We did this because it's an on-wire incompatibility only, and a new
feature bit protects us there.