New branch considers all files changed. We don't want to rebuild all
docker images for new branches; that's just a waste of time and energy.

There is a better way for future -- a much better fine-grained control
may be obtained by running secondary pipelines.

fixes #212

Possible commands are stored as keywords, each keyword has its own structure.
The last acceptable keyword structure contains string with hint. But when the hint was printed only direct child
of the base keyword was considered. If it was multi keyword command, the first child did not carry any hint to print,
so it was ignored.

Now, if we don't find a hint in a child, we recursively search in grandchildren.

Jana Babovakova authored 2 weeks ago and Ondřej Zajíček committed 1 week ago

- Licence to License - also in code comments.
- copyright date until now.
- updated license text from gnu.org
- added command to build bird documentation.

Maria Matejka authored 4 months ago and Ondřej Zajíček committed 1 week ago

The build-netlab job was side-effecting the test-* jobs,
and if for some reason Gitlab scheduled build-netlab before
other pipeline's test-* jobs finished, these jobs got a wrong
binary, possibly failing. Solved by using explicit artifacts, which is
not the fastest way to do this (we could keep the binaries named there)
but it's the gitlab-right way to do this.

(Re-committed, was accidentally removed by one of previous commits)

Forgetting to do proper checks, I merged a buggy release tools update
not detecting the right version (mostly if tagged).

Fixed and added current distros, deleted certain EOL distros,
sorted things in the CI file.

With some heavy lifting and comments by Maria.

Signed-Off-By: Maria Matejka <mq@jmq.cz>

Fedora builds for 32-41, Debian up to 12, Ubuntu up to 24.10.

Includes autorebuilder if a Dockerfile changes.

The CI file is obnoxiously boilerplaty. TODO: generate it from a list?

Also version detection is done by a separate script, not a magic line in
the Makefile.

This is expected to hit a webhook inside our CI to trigger
automatic deployment to testweb.

Some minor changes were done in the original documentation to allow for
easier conversion, and also to make the documentation a little bit more
strictly valid.

The build-netlab job was side-effecting the test-* jobs,
and if for some reason Gitlab scheduled build-netlab before
other pipeline's test-* jobs finished, these jobs got a wrong
binary, possibly failing. Solved by using explicit artifacts, which is
not the fastest way to do this (we could keep the binaries named there)
but it's the gitlab-right way to do this.

Reconfiguration does not work properly for this case, disable it for now.

Reworked and finalized version of TCP-AO.

Katerina Kubecova authored 9 months ago and Ondřej Zajíček committed 2 weeks ago

Uses Linux TCP-AO. Tested against JUNIPER.

Any change in BFD iface configuration should trigger session
reconfiguration, as config is copied into the bfd_session structure
and not just accessed through the bfd_iface structure.

As bfd_session now contains a pointer to the password list allocated
from the configuration, forgetting to update the bfd_session causes
use-after-free.

Thanks to Lexi Winter for the bugreport.

(was accidentally removed by the previous commit)

By default, the Babel protocol will advertise IPv4 routes with an IPv6
next hop address only on interfaces which do not have an IPv4 address.

Add option 'next hop prefer' to control whether IPv4 routes should use
IPv4 or IPv6 next hop when both are available.

Based on the patch from Chris Webb, thanks!

(For some reason, it is not enabled in CI)

When a BGP session is established using link-local next hop addresses,
there is no global IPv6 address for next hop. Implementations differ on
how to encode such next hop. This leads to interoperability problems.

Add the option 'link local next hop format' to specify which format to
use when encoding such next hops.

Based on a patch from Andrey V. Elsukov and Georgy Kirichenko, submitted
by Aleksandr Stepanov, thanks!

Some package build tests were broken due to apkg changes,
disable them temporarily.