1 # Some overall notes on how this works
3 # * We smoke using the system provided latest, and custom built "oddball perls"
4 # The reason for not having a blanket matrix is to conserve travis resources
5 # as a full DBIC depchain isn't cheap
7 # * Minimum perl officially supported by DBIC is 5.8.3. This *includes* the
8 # basic depchain. On failure either attempt to fix it or bring it to the
9 # attention of ribasushi. *DO NOT* disable 5.8 testing - it is here for a
12 # * The matrix is built from two main modes - CLEANTEST = [true|false].
13 # - In the first case we test with minimal deps available, and skip everything
14 # listed in DBIC::OptDesps. The modules are installed with classic CPAN
15 # invocations and are *fully tested*. In other words we simulate what would
16 # happen if a user tried to install on a just-compiled virgin perl
17 # - Without CLEANTEST we bring the armada of RDBMS and install the maximum
18 # possible set of deps *without testing them*. This ensures we stay within
19 # a reasonable build-time and still run as many of our tests as possible
21 # * The perl builds and the DBIC tests run under NUMTHREADS number of threads.
22 # The testing of dependencies under CLEANTEST runs single-threaded, at least
23 # until we fix our entire dep-chain to safely pass under -j
25 # * The way .travis.yml is fed to the command controller is idiotic - it
26 # makes using multiline `bash -c` statements impossible. Therefore to
27 # aid readability (our travis logic is rather complex), the bulk of
28 # functionality is moved to a script. More about the problem (and the
29 # WONTFIX "explanation") here: https://github.com/travis-ci/travis-ci/issues/497
33 # Smoke only specific branches to a) not overload the queue and b) not
34 # overspam the notification channels
36 # Furthermore if the branch is ^topic/ - the custom compiled smokes will
37 # not run at all, again in order to conserve queue resources
39 # Additionally bleadperl tests do not run on master (but do run on smoke/*)
50 - "irc.perl.org#dbic-smoke"
52 - "%{branch}#%{build_number} by %{author}: %{message} (%{build_url})"
60 # Temporary - if it proves to be too noisy, we'll shut it off
61 #- dbix-class-devel@lists.scsys.co.uk
76 # minimum supported with threads
80 - BREWOPTS="-Duseithreads"
83 # minimum supported without threads
91 - perl: bleadperl_thr_mb
94 - BREWOPTS="-Duseithreads -Dusemorebits"
97 # check CLEANTEST of minimum supported
101 - BREWOPTS="-Dusemorebits"
104 # this is the perl suse ships
105 - perl: 5.10.0_thr_dbg
108 - BREWOPTS="-DDEBUGGING -Duseithreads"
111 # this particular perl is quite widespread
115 - BREWOPTS="-Duseithreads -Dusemorebits"
119 # Do not make this part of the env-matrix
120 # different boxes we run on have different amount of hw threads
121 # hence why we need to query
122 # result is 1.5 times the physical threads
123 - export NUMTHREADS=$(( ( $(cut -f 2 -d '-' /sys/devices/system/cpu/online) + 1 ) * 15 / 10 ))
126 # Build and switch to a custom perl if requested
127 # Set the environment based on CLEANTEST
128 # Preinstall/install deps
130 # sourcing the file is *EXTREMELY* important - otherwise
131 # no envvars will survive
132 - source maint/travis-ci_prepare_env
135 - export HARNESS_TIMER=1 HARNESS_OPTIONS=c:j$NUMTHREADS
137 # either a plain 'make test' OR a shuffled prove torture
138 # use the random order test plan unless CLEANTEST
139 # prepare_env may have short-circuited the test entirely
140 - test -n "$SHORT_CIRCUIT_SMOKE" || (test "$CLEANTEST" = "true" && make test || prove -lrswj$NUMTHREADS t xt)