publish_to_dockerhub.yml duplicated build.yml and run_tests.yml in many
ways. We always want to build docker images, it's just that we don't push them
on every occasion and there is some additional logic around the latest
tag. We also want to publish multiarch images now, so all of this logic better
be consolidated in one workflow. It doesn't depend on tests, since we only
publish on release or on manual run, so it's known to be good.
We build multiarch linux/amd64 and linux/arm64, because MacOS runner can't
build docker images and even if it could that'd be linux/amd64 while we want
linux/arm64 for Apple CPUs.
Unfortunately, given the way GitHub workflows work we can't avoid using a
Makefile helper, there is no easy way to set variables conditionally and/or
use some logic to affect their contents.
We reintroduce build_image_wsc as well here because Windows images can't be
built with buildx using GitHub runners.
Unfortunately, the default MacOS runner produces amd64 binaries, therefore we
need to set GOARCH appropriately. At the same time, docker image will be
linux/arm64, so we can build it as well.
In case there are no returns in the inlined function, jumps point to the
next instruction and can be omitted. This optimization can be extended
to handle other cases, here we just make sure that already existing code
stays the same.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
It makes a copy of the resulting set, so the lock can be released
earlier. This helps a lot with iterators that keep Seek() unfinished for a
long time,
Change stack items before marshaling them which makes code in result package
much simpler and not requiring interop, iterator and storage dependencies that
clients shouldn't care about.
This also changes SessionBackedByMPT behavior, now instead of waiting for
traverseiterator call it'll rerun the script immediately if a new session is
created.
It doesn't add anything useful to regular Go types and actually native types
are always better to use in the Client. Especially given that this type is
not used by any code outside of the Client itself.
We've got parameters here and usually we name them Raw when they're
represented by json.RawMessage which is not the case here, so make it a bit
more friendly (the type itself is only used in client internals, so rename is
not a huge problem).
It's absolutely irrelevant for the client and request/response packages should
only contain code that is useful on both sides of the conversation. It's OK
for client tests to reuse this code, but the package is used by external
developers and they shouldn't be bothered with it. Nothing changed
functionally here except WSClient simplification. Fixes#2236.
