Do not use pipes for building & uploading packages because:
* They spawn more jobs on Bintray for metadata generation - each upload is
done in a separate session.
* Bintray doesn't handle well concurrent uploads and metadata can be corrupted.
This happened in all recent builds of platform independent packages - in
each upload task file had a same name and that triggered problem with metadata.
* Using delays for workaround for a problem described above works, but isn't
safe.
CI periodically shows error like
Failed to decrypt some environment variables
so it will be better to stop immediately when environment is inconsistent
for unknown reason (e.g. due to Github's problems).
All platform independent packages affected by this issue. Results in
persistent apt error like:
Writing more data than expected (22610650 > 22610408)
Hashes of expected file:
- SHA256:08abc5cffb9e1967fab487a471bb547f5ae706c48bf8e59d1443b2ad4af9caaf
- SHA1:36ded1e207135bdeb8b7301bc03309ebced3cd32 [weak]
- Filesize:22610408 [weak]
In certain cases Bintray may return errors like `invalid version a.b.c` for
specific tasks. This probably happens because we now running package uploading in
parallel.
Retrying upload one more time should fix the problem.
* Move package building/uploading implementation from config to a separate script.
* Implement new tag '%ci:no-build' to allow immediately stop build with status 'passed'.
%ci:no-build
Keep bare minimum of scripts & configuration files for CI.
Set Travis target language to generic, remove 'Gemfile' and
'Rakefile' since we building *.deb packages and not ruby ones.
Code that determining changes in git repository is reimplemented
in bash.
Disabled packages are excluded from CI tracking as they frequently
cause errors.
its-pointless
Leonid Plyushch