All-numeric domains are now considered invalid, since they introduce confusion: one can assume
that an all-numeric domain is a port, and the server should bind to `localhost` with that port.

The specific change to domain parsing happens in a future commit. This commit fixes the tests in
preparation for it.

Note that all-numeric top-level domains are considered invalid according to RFC 952 and RFC 1123.
In fact, the top-level domain label must be alphabetic, but we do not place that restriction, since
local domains could have arbitrary format.

Checking for specific error messages is fragile. Any cosmetic change to the message,
such as fixing a typo or enabling coloured output, causes test breakage. It also
prevents us from designing any better errors, or including backtrace or source trace
in error messages. We also shouldn't depend on the specific error being returned. For
example, connecting to an invalid host can return various errors, including timeout
errors or host resolution errors, and the errors may depend on the specifics of OS and
network connectivity (e.g. a network error can change a DNS resolution error into a
DNS timeout).

Instead, use regular expressions to check for presence of errors, without depending on the
specific text.

Mutex is held for the duration of the whole tests, and poisons if the test panics, making all
other env tests fail with PoisonError. Panicking tests is expected behaviour during development.
We don't introduce any inconsistency by clearing the poison, since the only invariant that the
mutex protects is that tests don't concurrently access the environment (which must be blocked for
the whole test duration, since tests depend on the specific change order of env variables) and that
the environment is unconditionally restored when the test terminates, including panics.

Антон Фетисов authored 1 month ago and Антон Фетисов committed 1 month ago

This fixes the `static-mut-refs` lint in sbroad.

When specifying the precision for localtimestamp, the digits to the right are zeroed out. However, if there were already zeros in those positions in the original value, the test may fail even when precisions are different. This can lead to flaky tests so test is removed.

EmirVildanov authored 1 month ago and Dmitry Ivanov committed 1 month ago

printlns were previously used in order to fix expected explain
more easily, this commit removes them and also refactors some code

EmirVildanov authored 1 month ago and Dmitry Ivanov committed 1 month ago

Before this commit we wasn't recalculating types after parameters
binding. This commit add additional tree traversal in order to fix it.

EmirVildanov authored 1 month ago and Dmitry Ivanov committed 1 month ago

Previously we were mistakenly giving default Integer type
to the Null value or to the column with no type. This commit
fixes such behavior and also adds support for type absence
during its calculation on SQL expressions

Denis Smirnov authored 2 months ago and Dmitry Ivanov committed 1 month ago

Stop using tarantool types in virtual table columns and derive them
from IR plan.

* fae9c5bbe fix: timeouts for on_shutdown fibers used to be computed incorrectly
* e2911d68e feat: auto set box.cfg{checkpoint_enabled=false} on split-brain
* 4cd0b4ae0 feat: add box.cfg.checkpoint_enabled to toggle snapshots on/off
* 3352da865 fix: ssl iostream used to return errors without setting them via diag_set
* cdf7a646d fix: drop WAL directory lock upon exec()

Антон Фетисов authored 1 month ago and Антон Фетисов committed 1 month ago

Criterion 0.5 has disabled default features of Clap, but those include help & usage messages.
Added Clap to explicit dependencies with default features, so that it works properly for benches.

Kurdakov Alexander authored 1 month ago and Alexander Kurdakov committed 1 month ago

Since it's deprecated option, it should not appear anywhere
except of explicit user intention.

Антон Фетисов authored 1 month ago and Dmitry Ivanov committed 1 month ago

Closes #1085
Closes #1084

Виталий Шунков authored 1 month ago and Виталий Шунков committed 1 month ago

fix: rule for build-base-image

Антон Фетисов authored 1 month ago and Антон Фетисов committed 1 month ago

This reduces build errors for sbroad tests when the `mock` feature isn't enabled.
This doesn't fully fix the build, linking still fails with an error (strange, tarantool-module
is listed in dev dependencies, but somewhy isn't linked properly). However, at least there are
fewer errors now.

cfg(test) on inner test modules is useless, since the containing module is already conditionally
included.

 - checkpoint_interval
 - checkpoint_count
 - max_concurrent_messages

max_concurrent_messages were renamed to iproto_net_msg_max

Антон Фетисов authored 1 month ago and Georgy Moshkin committed 1 month ago

Closes !915