I am not certain exactly how it happened. There was no indication in the
recent versions of MW that there was a problem with the database.
What took a bit to figure out is that the problem was with the values in
the revision table. When I looked at the info page for the affected pages,
the content model was valid. The values in the page table in the database
were fine too.
Of course the software cannot catch every bogus value in the database, but
perhaps the diagnostic could be clearer that the revision table is the
source of the difficulty.
David
On Thu, Jun 21, 2018 at 1:24 AM [[kgh]] <mediawiki(a)kghoffmeyer.de> wrote:
Thanks a lot for sharing your solution to the problem.
However, this
sounds like a bug to me that should be fixed.
I however do not know how you ended up with 0 instead of NULL and if
this could also have happened to other users of MW using the special
page in the past to change the content model.
I understand that some developers are on this list too who can easily at
least assess if there is an issue. Let's wait for their reply.
Cheers Karsten
Am 21.06.2018 um 02:19 schrieb David Bohman:
I got past the problem. It appears that MW 1.31
is not as permissive
about
the value of rev_content_model in the revision
table. I had entries in
the
table which were 0, and MW 1.30 did not have a
problem with it. However,
1.31 only accepts a value of NULL, otherwise it barfs.
Anyhow, I fixed it by forcing the 0 values in the revision table to their
default value. The logic seems to be to be to use the content model from
the revision if it is set to non-null, otherwise use the value in the
page
table. I suspect that earlier versions of MW
accepted a revision content
model of 0 as the same as NULL.
David Bohman