This is simpler than parsing /sys/devices/system/cpu/online ourselves
and also works under macOS.
Reduce the number of jobs, 1.5*CPU number is too much, so use more
conservative +1 instead.
Also rename TCI_NUMTHREADS (which was another reference to Travis CI) to
just num_cpus.
Use python3-venv package which should work on all Ubuntu versions,
including Xenial where this script currently runs, and which doesn't
have python3.4-venv.
This might help with sporadic failures in Travis CI Firebird build,
which sometimes fails with weird errors like this:
Unable to complete network request to host "localhost".
Failed to establish a connection.
unable to open database
Error setting new SYSDBA password
Note that dpkg-reconfigure does not return an error in this case, so we
need to check the password file itself to see if it was updated.
"services" can't be used with macOS runners and there is no way to limit
them to Linux runners only (i.e. "if:" is not supported here), so we'd
have to use 2 different jobs and duplicate all the other steps if we
wanted to continue using them.
Instead, just install PostgreSQL manually under Linux. This also means
that it isn't done for the builds that don't need it uselessly any more.
It is installed there but not started by default, so just launch it
ourselves.
Note that we must not use PGHOST etc variables under Mac, as the server
seems to be configured to only listen on the local socket by default, so
don't define them globally and do it only for Linux manually instead.
It doesn't make sense to call these scripts "before install" because
they do actually install things themselves.
In fact the whole separation into "install" and "before build" is
probably not very useful, but keep it as long as we keep using Travis CI
at all, as it maps better to its config file keys.
No real changes.
"readline" frontend doesn't work without a controlling TTY as can be
seen by setting DEBCONF_DEBUG=user before running dpkg-reconfigure, so
use "teletype" instead.
Also note that DEBIAN_FRONTEND environment variable unexpectedly seems
to override the value explicitly passed via dpkg-reconfigure -f option,
so we need to reset it (and then we don't need the -f option at all).
Handle Xenial which is similar to Trusty and also Focal, where we need
to install Firebird 3.0 server and tweak expect command to work with it.
Note that we still use Xenial for GitHub Actions Firebird CI build as
using Focal version results in memory leaks inside Firebird libraries
and there doesn't seem to be a debug symbol package for Firebird 3.0
that would be required to write a suppression file for them.
This fixes the problem with psqlodbca.so not being found by unixODBC in
GitHub Actions environment.
It remains unclear why was this unnecessary on Travis CI and why it
doesn't work out of the box, but without this change we were always
getting
[unixODBC][Driver Manager]Can't open lib 'psqlodbca.so' : file not found (SQL state 01000)
when trying to connect to the database.
Start the service explicitly as this is not done by default.
Also explicitly use password with mysql in GitHub Actions environment,
both for the root and for the current user.
This wasn't necessary on Travis CI but is required in this environment,
so specify the root password explicitly and also create the MySQL user
(which is not hardcoded "travis" any longer) and record its password in
~/.my.cnf.
This fixes Oracle tests after 325ee4e9 (Split build and test stages of
the CI builds, 2021-03-16) which resulted in oracle.sh not being sourced
when running the tests any longer.
This workflow reuses the same CI scripts as the existing Travis CI
config file and, for now, uses Ubuntu 16.04 (Xenial) runner to be as
close to Travis configuration as possible.
While this doesn't matter for Travis CI, it makes sense to separate them
for the other CI systems using more structured approach to the build
steps.
Add a few test_*.sh scripts for the backends that need to do something
special when running the tests and just use run_test function for all
the others.
The term "script" is too generic and really doesn't mean anything,
especially outside of Travis CI context, where "script" is used as the
config file key for the build script to run.
Travis hosts use PostgreSQL PPA which is problematic for a few reasons:
1. We don't actually test in the same environment as most SOCI users,
who would just use the standard Ubuntu repositories and not this PPA.
2. The environment is not even stable, e.g. recently the version of
psqlodbc has changed from 12 to 13 in Travis builds, breaking them
due to new leak reports from ASAN.
3. We don't have debug symbols for these packages as the PPA doesn't
provide them and the ones from Ubuntu repositories don't match. This
prevents us from even creating precise suppressions for ASAN.
Solve all these problems at once by just sticking to the stock Xenial
version, which doesn't even trigger any leak reports, so there is
nothing to suppress any more and the changes of 457ec97e (Suppress
reports about memory leaks in PostgreSQL ODBC test, 2020-03-29) can be
reverted.
Define SOCI_DEFAULT_CMAKE_OPTIONS once and use it in almost all builds
instead of repeating the same options many times everywhere.
For the remaining build (Valgrind) still reuse a couple of options which
it has in common with the other ones.
Unfortunately we don't get any information about these leaks, the entire
report looks like this:
==6706==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 24 byte(s) in 1 object(s) allocated from:
#0 0x7f6473f80602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
#1 0x7f646f89ba82 (/usr/lib/x86_64-linux-gnu/odbc/psqlodbca.so+0x1ba82)
Indirect leak of 25 byte(s) in 1 object(s) allocated from:
#0 0x7f6473f80602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
#1 0x7f646f89ba9a (/usr/lib/x86_64-linux-gnu/odbc/psqlodbca.so+0x1ba9a)
SUMMARY: AddressSanitizer: 49 byte(s) leaked in 2 allocation(s).
So we have no choice but to suppress all leaks from psqlodbca.so to let
the tests to pass on Travis CI.
Unfortunately we don't get any real information about these leaks, the
entire report is
==10667==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 176 byte(s) in 1 object(s) allocated from:
#0 0x7ff24b649602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
#1 0x7ff23f0ef551 (<unknown module>)
Indirect leak of 744 byte(s) in 3 object(s) allocated from:
#0 0x7ff24b649602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
#1 0x7ff23f0ef551 (<unknown module>)
Indirect leak of 64 byte(s) in 1 object(s) allocated from:
#0 0x7ff24b649602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
#1 0x7ff23f456d89 (<unknown module>)
So we can only disable leaks coming from libasan.so to suppress them
which, of course, risks hiding any other leaks too. The only way to
check for them is to check that the count of suppressions in libasan.so
doesn't increase beyond 5.
If TRAVIS_BRANCH matches release/X.Y, then the once the Travs CI job
completes building the empty, it also does
- run release.sh to generate source archive for soci-X.Y.Z
- unpacks the archive
- runs a minimal test build of the source archive,
equivalent to the main build run earlier by the job.
The purpose is to have a minimal procedure verifying the source code
archives generated for release.
Remove automatic branching off of master as release/X.Y,
instead expect and require existing release branch.
Update RELEASING.md about use of the release.sh script.
Vadim Zeitlin