Because we want to avoid local replace directives (as they make it easy
to screw up things for external applications that use lnd as a library
without us noticing), we're going to switch over to a new process.
See section "Use of Golang submodules" in the code contribution
guideline.
As is customary with new GitHub actions, they don't work on first
attempt if you don't test them *sigh*
We need to use a more recent version of golang than is pre-installed to
avoid the "unknown directive: retract" error message.
It turns out we were using the wrong matrix variable in the actual
make command and ran the same itest 6 times with no arguments, all
resulting in running the btcd test.
To avoid uploading too many files in individual requests, we zip them
first before uploading the zip itself.
To reduce the likelyhood of the linter OOMing on the GitHub runner, we
exclude any generated code from being inspected and also tune the golang
garbage collector to be a bit more agressive.
In this commit, we add a simple bash script to parse out the current PR
number from an environment variable in the GH actions context, and use
that to check to see if the PR has been referenced in the release notes
or not. This isn't 100% fool proof, but it should catch most of the
common cases.
Because we now build a docker image for the RPC compilation, we can save
some execution minutes if we run the mobile RPC and code compilation check in the
same step of the CI workflow.
Because we now have conditionally compiled code that depends on the
architecture it is built for, we want to make sure we can build all
architectures that we also release. Since GitHub builds are very fast,
we can easily do this instead of only compiling for certain select
architectures.
We add a GitHub action to our workflow that makes sure all command line
flags of lnd that are available with the default build tags are
contained in the sample-lnd.conf file.
To free up build in Travis, we decided to run the non-flaky parts of
the CI pipeline in GitHub Workflows/Actions only. The integration tests
on the other hand are removed from GitHub because individual actions
cannot be restarted there which caused us to restart the whole workflow
if one test was flaky.
This split should give us the best of both worlds: Fast run of small
checks, linting and unit tests with an easy overview of what failed in
the PR directly. And more free build slots on Travis to do more advanced
integration tests on other architectures and/or operating systems. And
the option to restart a single flaky integration test on Travis.
Checkout v1 has a known flake:
https://github.com/actions/checkout/issues/23#issuecomment-572688577.
For our linter to pass, we need to checkout our full history (default
depth is 1 commit). We could set fetch-depth, but we will eventually
move that depth past the linter's start point commit and need to
implement another fix. Indead, we add an extra step in our linter to
fetch full history so that the linter reference commit is found.
The continue-on-error was added to make sure the log files of the
failed itests would always be uploaded. But this has the side effect
of marking the whole job successful, even if the itest job itself
failed. The failure condition in the log file steps already solve
that, so the continue-on-error is not needed anymore.
Elle Mouton