These get created when running kresd without changing
to a different rundir.

The KRU-based garbage collection reduced the LMDB size a bit
for a particular overall cache size setting.
Apple Silicon uses 16 KiB pages instead of the usual 4 KiB,
so the usage is a bit shifted and that's why we didn't notice earlier
(though before the KRU GC we adjusted in commit adaac913)

For example failure, see "not ok" lines in
https://cache.nixos.org/log/1452wsll1fxh38p0fbqa0vw1p7ki6v55-knot-resolver_6-6.0.17.drv

Fixes #714

There's no ./static dir to even install.
Apparently this snippet came from modules/http.
Discovered thanks to meson reporting:
  NOTICE: Future-deprecated features used:
   * 0.60.0: {'install_subdir with empty directory'}

https://mesonbuild.com/Reference-manual_functions.html#summary
- has nicer output
- less code in meson files
- allows defining summary parts closer to where they originate

Also short pieces of meson code are moved around.

I'm not sure how much info to put into those summaries;
this commit keeps the amount about the same.
Also note that (while not documented in meson)
a "User defined options" is appended, so the summary()
may end up repeating entries from that section.

This gets rid of a warning * 0.52.0: {'priority arg in test'}

There's still Ubuntu 20.04 with meson 0.53.2,
though its standard support ended a few months ago.
Debian 11 with meson 0.56.2 is similar in this respect.

Still well supported: RHEL 9 derivatives with meson 0.63.3
and Ubuntu 22.04 with meson 0.61.2.

Support for Python 3.14

See merge request !1767

The problem was that in Arch the clang-tidy package
updates at random moments, so we have to resolve
the issues when doing urelated refreshes of CI images.

Aleš Mrázek authored Dec 05, 2025 and Vladimír Čunát committed Dec 08, 2025

PEP 649 & PEP 749: Deferred evaluation of annotations
The annotations on classes(ConfigSchema classes in our case) are no longer evaluated eagerly.  They are evaluated only when necessary.

It might happen that we have a stale PID file, e.g. after a crash,
but some other process has taken the freed PID already.

Release 6.0.17

See merge request !1762

It's just removing an assertion, and it does make sense to me.
Also expose kr_rule_local_data_ins() in Lua, like other similar functions.

Otherwise we get use-after-free when attempting to close the cache
after this failure causes the process to shut down.

In particular, I was now interested in `cache-gc`
and incorrectly assumed that it's merged with `cache`
(the C code didn't help, as they appear merged on that level).

With clang they'd cause lots of complaints:
  warning: '_Static_assert' with no message is a C23 extension [-Wc23-extensions]

Also simplify the related code a little.
I don't think this has any practical impact.

The code clearly assumes it isn't, and I checked that it never is called
that way (and it would crash anyway).

Affects just clang, apparently.  This is just a copy from knot-dns changes,
doing a bit more than this resolver commit title says.  See:
knot-dns@ec4a2567
knot-dns@509d9d82

It seemed to bring more complexity than benefit.
In many parts this meant revert to code before a few commits.

It's slightly tricky that qry or req may be NULL in some cases,
but I believe I've caught all of them by conditions, and moreover
they're cases where it doesn't make sense to update kr_cache_top_*

Move the function up unchanged and use it on two places
which have the exact lines as the function.
Also fix its description.