[mythtv-users] INSERT INTO recordedmarkup fails w/duplicate; recording status changes [Workaround]

[Ross Boylan]

All of the weird behavior I've been  seeing seems to have gone away
when I downgraded all mariadb packages from the current stable version
in Debian, 1:10.5.19-0+deb11u1, to the previous stable version,
1:10.5.18-0+deb11u1.*  I believe the key difference is in the version
information provided in
https://sources.debian.org/src/mariadb-10.5/1%3A10.5.19-0%2Bdeb11u1/libmariadb/libmariadb/mariadb_lib.c/,
which follows the new scheme used in Maria 10.6, vs
https://sources.debian.org/src/mariadb-10.5/1%3A10.5.18-0%2Bdeb11u1/libmariadb/libmariadb/mariadb_lib.c/,
which follows the old scheme.

See this proposed patch (against 10.3, however) to undo the transition
to the new scheme, and discussion in
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1031863.
Apparently the maria 10.6 scheme reports a major version of 3, while
the old one reports a major version of 10, and that is throwing other
things off.  In particular, Qt's SQL library (libqt5sql5 on Debian)
gets thrown by this.  There are various proposals to patch Qt floating
around.

This change in versioning seems likely to screw up the conditional SQL
code we discussed a little earlier as well--in fact it makes me wonder
which parts of it got run.

Note that this maria versioning explanation is not necessarily in
conflict with those centering about the handling of a trailing Z in
the datetimestamp; the version problems may be causing the
datetimestamp problems.

This is a hack because downgrading is a hack.  Even if it works my
package tools want to upgrade immediately; presumably I could pin them
to that version.  And the newer release fixes some bugs and security
holes, as later releases will too.

More fundamentally, MariaDB is moving to the new versioning scheme,
and so other software using it will need to adapt.  Note there are
likely 3 different places that adjustments could be made to work
around this problem:
1. myth (though it's not clear to me it can work around problems in Qt).
2. Qt's SQL library.
3. Maria.

Apart from the version info maria reports, its handling of trailing
Z's in datetimes could be more forgiving, particularly in keys.  Some
reports suggest later versions of maria can handle times with Z OK,
and possibly earlier versions did as well.

My experience seems to imply that the upgrade to maria 10.5.19, which
happened on 3/11 according to my logs, did not take full effect until
the system restarted on 4/4.  Which seems surprising, particularly
since my recent downgrade became effective immediately.  For the
recent downgrade I stopped, myth, the SQL sessions I had open, and
maria before downgrading.  Maybe I missed some of those steps earlier
(though I thought the packaging system itself stops maria before
changing it). Perhaps myth checks maria version info when it starts,
and I didn't restart myth?

Ross

*I downgraded them all together out of caution and deference to prior
admonitions in another thread to do them all together.  It might be
that downgrading libmariadb3 alone, which I believe is where the
version reporting code lives, would suffice.  But that's riskier, and
the packaging system might not even support it.  BTW, "stable" is an
attribute of the entire distribution, not a particular package.
"stable" is the mainline version of Debian = Debian 11 = bullseye.
Packages in it are supposed to keep a consistent API even if upgraded
for critical fixes.  The maria packages obviously have not managed
that.