Target ar71xx is deprecated and removed in the master branch and makes SDK
not available anymore. Travis fails because of that.
It was superseded by target ath79. These devices have 4 MB flash and/or 32 MB RAM.

However, ath79 is being used by CircleCI if you have it configured for
your repository and if you are contributing to this repository. It
is not good to have two CI for the same target. Let's use powerpc.

Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>

Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

$SDK doesn't need to be updated everytime the filename changes because of
a gcc update

Signed-off-by: Alexander Couzens <lynxis@fe80.eu>

Signed-off-by: Alexander Couzens <lynxis@fe80.eu>

OpenWrt build now outputs WARNING on dependencies issues, for exemple
"WARNING: Makefile 'package/feeds/packages/coova-chilli/Makefile' has a dependency on 'libmatrixssl', which does not exist"

Relax the check for now

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

TRAVIS_PULL_REQUEST can be:
- undefined (local usage)
- false (branch build)
- PR number (if in a PR)
it's never set to true

use TRAVIS_PULL_REQUEST_SHA instead, so we can use normal tests (-z/-n)

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

This started out as me trying to fix my Python build
in a local Travis CI env.
After 40+ minutes, the build still did not stop:
https://travis-ci.org/commodo/packages/builds/310413628



There is a bug, where the `.travis_do.sh` script does exit early,
and enters an endless loop, when doing a branch build.
It tries to continuously fetch 50 more revisions.
So, I decided to add some limited support for branch builds.

This also allows a simpler testing with Travis before
opening a PR.

This scope of testing is limited to whatever the last
commit touches, but it's still better than nothing.

Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

commit gpg keys into the repo (it sometimes timeout)
trust the keys ultimately

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

More precisely, remove this message
./.travis_do.sh: line 16: 13736 Terminated              while :; do
    sleep 590; echo "still running (please don't kill me Travis)";
done

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

"fatal: Invalid revision range XXX...YYY"

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Rework exec_status to use "tee" to continuously output logs
(so we don't trigger travis timeouts)

Only launch the compile test if everything else is ok
(spare Travis builders a bit)

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

git merge-base does not work as I expected

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Paul Spooren <paul@spooren.de>

restarting a build doesn't update .travis.yml but does update .travis_do.sh ...

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

if the merge-base is not in the initial clone,
git diff / git rev-list will return funky results

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

- use 'git diff --name-only'
- unpack the sdk only once as if there is no errors we don't need clean sdk
  fresh sdk mean new git checkout of the feeds each time
- only include base, packages and luci,
  as we should not have dependencies on other feeds
- use github for feeds
- continue to test on error
- add a bit of color
- use bash

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>

Travis will run the following steps for every new or modified package:

- mkdir tempdir
- unpack SDK
- make download
- make check

make download will download the source code.
make check will do certain checks. Atm only checking the validity of
PKG_HASH.

Signed-off-by: Alexander Couzens <lynxis@fe80.eu>