This attempts to allow sharing a host directory across multiple builds, target
use case is to have a shared source directory on a VM host and be able to build
from that in a number of VMs without those builds stepping on top of each
other.
Instead of subsurface/build, subsurface/mobile-build,
subsurface/libdivecomputer/build, use a prefix path to allow having true out of
tree builds.
The one shortcoming is that the autotools need to be run in the libdivecomputer
directory - that means this will be run on the first system that starts a
build. But that seems to cause no harm in my testing.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Sadly, there's an explicit change in the sources to of QtConnectivity
that requires this workaround when running the build on a case sensitive
file system.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Right now in the Ubuntu 14.04 base image for Docker the file
/etc/apt/preferences.d/ubuntu-esm-infra-trusty doesn't exist.
Subsequently, apt-get update enables ESM, but since we don't have a license to
use that upgrades / installs from ESM fail.
This workaround simply ensure that there is such a file pinning ESM to never be
used. With that, the creation of our image should succeed again.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Based on ideas from Anton - both the basic building of containers in the first
place as well as the workaround for the 6h build limit.
Because GitHub Actions are limited to 6 hours we split the creation of the MXE
container into two steps and push the intermediary container after stage 1 to
docker hub. Right now each of the steps takes about 3.5 hours, so hopefully
even with changes in the future this will continue to work.
This commit also introduces use of docker hub instead of GitHub's own registry
(since strangely right now GitHub actions cannot run containers from GitHub's
private registry).
In order for this to work, we need to have the docker credentials in secrets in
GitHub. As a result, only people who can create branches in our repository can
easily test changes to the container images. Others can modify the code to use
a different docker hub account and provide those secrets in their own GitHub
account. Not ideal, but of course we cannot allow every pull request to
potentially overwrite docker images in our "official" docker hub account.
Suggested-by: Anton Lundin <glance@acc.umu.se>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
libssh2 depends on openssl, therefore it is important that openssl is
build before libssh2.
The old get-deps would cause errors in 2 situations:
1) In a clean build, make of libssh2 would fail
2) In a normal build, where openssl changed version, make of libssh2 would
depend on old build.
Signed-off-by: Jan Iversen <jan@casacondor.com>
This feature is in beta right now and might change without notice, but instead
of dealing with the broken Travis Mac builds, this does seem progress.
The build artifact seems to work, but it's a bit more painful to get to. Go to
https://github.com/Subsurface-divelog/subsurface/actions and click on the
corresponding run - it's then in the top right corner under Artifacts. The one
oddity is that after unzipping the file you need to manually make
Contents/MacOS/Subsurface executable.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Also make sure Grantlee still compiles with Qt 5.13 by cherry picking a commit
that was added after the v5.1.0 release.
In order to identify this commit as comming from the build automation we
temporarily override the user name and email address. As a side effect this
also makes this work on Travis.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
QML has ways to style icons - and we use that for the main theme color,
but it doesn't seem to work (anymore?) for the edit and save icons.
Instead of tracking down what changed there, simply switch between icons
with different foreground color, depending on theme.
All the other icons seem to work well in all three themes.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This reverts commit 60e63afb82.
I merged this to early without paying attention to the fact that this
needed an updated build container as well.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Kirigami isn't picking up our font for the Back entry in sub menus.
Also, we still don't get a back button icon on Android. This will
allow us to work around that.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This looks extremely fishy to me, but it does seem sufficient to
get the forward and backward buttons to show up in the toolbar.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Whatever I try, the toolbar background is always based on the 'active'
color set in qtquickcontrols2.conf, not on anything that I can set in
QML code. So in an effort to brute-force the issue, this hardcodes the
subsurfaceTheme value in the toolbar UI code of Kirigami.
To make this easier, this (and one of the other hacks) is added to the
existing kirigami.diff.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The hack to remove the action button caused situations where the action
button didn't return. Let's skip that for now. All the other fixes
appear to still be needed.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Remove hidapi from manually built components and use the mxe based one instead.
Remove libzip as that is handled by mxe packages.
Update version of grantlee used to build with qt 5.13.1.
Also hide vscode files from git.
[Dirk Hohndel: combined two commits, cleaned up the commit message and removed
one now incorrect comment line from mxe-based-build.sh]
Signed-off-by: Paul Buxton <paulbuxton.mail@googlemail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
When building dbus within the appimage, cmake picks up the installation
path of various files dbus uses through the GNUInstallDirs package,
however this doesn't work under the appimage build.
So we replace the variable with the normal location of this file.
Signed-off-by: Paul Buxton <paubuxton.mail@googlemail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
For this we need our own hand crafted trusty container with Qt 5.12,
including QtWebKit and an updated cmake and libdbus, as well as already
build googlemaps plugin, grantlee and libgit2.
At the same time stop uploading the Subsurface AppImage in the
traditional trusty build.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
With all upgrading, the build apks now show up in a slightly different
location. Correct this in the scripting. Notice that this is debug
building only. Release building is outside the repo.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
Trivial. The final touch command was missing the proper quotes, so it
created a bunch of strangely names files from the date command. Just
good for the developers that like to peek into the docker image.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
The main reason for upgrading of the Qt version is the hunt for a broken
BT/BLE stack, preventing downloads from BT/BLE enabled DCs, in relation
to arm64 architecture builds. (And the absolute need for an arm64 build
in relation to the publication of the Android app in Googles Play
store).
In addition, Qt 5.12.4 starts supporting OpenSSL 1.1.1c, and trying to
use our current OpenSSL 1.0 series is highly discouraged by Qt (and
OpenSSL itself).
So, upgrade both in unison. But ... be careful bisecting issues on this
commit, as it does break our build. That will be fixed in the next
commit.
This fixes the BT/BLE download for arm64!
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
The .app.zip should once again run on any Mac (ignoring the security issue of
unsigned binaries). The Qt binaries in that archive include the jpeg and png
libraries that were missing in the Qt 5.11.1 binaries we used until now.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
0.9 docker image includes static libraries to build mdbtools so there is
no need for an aditional tarball.
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Add -a parameter to tee to avoid overwriting build.log when building
static libraries for smtk2ssrf
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Passing an argument on the docker build command line avoids the need to
modify the Dockerfile for each image build.
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
mdbtools only builds static under mxe.
This should add static build of glib to the container with the mxe
libraries.
[Dirk Hohndel: merged with latest version of Dockerfile]
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Instead of trying to do it all in one step rely on --squash to do its
job. Don't try to be so aggressive in removing things, it saves very
little space and caused builds to fail.
This results in version 0.9 of the MXE build container
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Printing never worked, none of this was ever included in test builds. Also, now
that there are official releases of QtWebKit again, this just doesn't seem worth
carrying along anymore.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Just like Android, Windows binaries are best created in a container.
I still need to push the latest version to docker hub and use it on
Travis, but this way at least the Dockerfile is here.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
... and remove install of the default (old) libgit2 from OS. That old
(0.24.0) libgit2 will be replaced by a newer anyway, so useless to
install.
But the real change to get this Travis build running again is using
the well known openssl instead of libressl.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
The install was missing curl.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl> Removed upgrade to newer libgit2.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
We currently require a minimal version of libgit2 of 0.24.0. From
issue #1926 it seems that this version is too old. So, a simple test on
Linux to see the behaviour with such an old libgit2, I tried that.
Interestingly, with the current version of openssl that old libgit2
version does not even compile from source (known error in libgit2).
So, bump our minimal version of libgit2 to 0.26.0. That is also the
version we currently use on the Travis and official builds, so well
tested.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
The scripts/build.sh script has an option --skip-googlemaps. Introduced
in 2017 at a moment the Travis Mac build failed on this. Interestingly,
when Mac building of the maps plugin was possible again (commit 79e3f69f48)
the --skip-googlemaps stayed. Obviously, this hack was never intended
to be used for anything else then getting it passed Travis on
some point in time for a specific Mac build.
So, remove this option.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
With this we have working arm and arm64 images (except that the arm64
image crashes when using Bluetooth).
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
cmake 3.10 (which comes with Ubuntu 18.04) in combination with Qt 5.12
and the current qt-android-cmake causes an odd bug. Paths are set with a
double slash at the start '//' and later in the process this causes
garbled path names for some of the objects which in return causes the
APKs built in the container to fail.
Upgrading the cmake inside the container to 3.13.2 fixes that problem.
All the credit for identifying the problem and figuring out a solution
goes to Jan Mulder.
The resulting container was pushed to Docker hub as version to 5.12.03.
Reported-by: Jan Mulder <jlmulder@xs4all.nl>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Again, this is relevant for developers that do local docker android
builds, and normal android builds. A normal build uses the directory
subsurface-mobile-build-arm(64), and when doing a docker android build
this directory is shared between host and container. That sharing is
good, as it nicely exposes the build tree to the host (for easy compare,
inspection, etc.). But reusing the same tree as the local one is
inconvenient (and possibly dangerous due to all kinds of caching
issues).
So, give the docker build its own output tree for the shared
subsurface-mobile-build-arm(64) build output.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
Users that use docker locally for Windows style build and Android style
builds will (probably) not like that we use the same name for both
docker containers. So, give the android builder its own name.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
As explained in commit 449d4ee33d ("Android build: add explanation for
huge hack").
It seems reasonable to add this to our Travis image as that is custom
made just to build our Android binaries.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This should make our mac builds on Travis faster.
This also switches to the latest xcode / VM image which helps speed things up
(less to update for Homebrew). It turned out that that app directories that we
were creating here didn't run for people, anyway, so why even bother with an
old image.
We still create / upload that image (simply in order to be able to peek into it
in case something goes wrong).
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Initial implementation/prototype of copy-paste support for
Subsurface-mobile. The UI part is really lacking; right now the copy
button is initially visible and paste is achieved by long press on a
dive and clicking the paste button when it appears. Delete is currently
not possible at all, as I just failed to layout the buttons properly
using QML. It just sounds so simple, to put all the copy-paste-delete
buttons next to each other...
The data to be copied is currently hard-coded. A dialog to choose
inteded fields would be nice, but it'll take quite a bit effort to get
used to QML enough to be able to hack something together.
Anyway, this seems to work, even though the UI is not always reflecting
the paste without switching dives (when testing on laptop).
Signed-off-by: Miika Turkia <miika.turkia@gmail.com>
And use that to have our Travis build still work with the existing MXE
build container as well as the even older, pre-compiled MXE binaries
used in the windows build.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Ooops. Forgot to fix this before sent the patches, as this part doesn't
works on my travis builds.
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Enable building of SmartTrak divelogs importer.
A new, lighter, tarball for mxe static libraries has been built, as it
seems impossible to build mdbtools with shared libraries (see mxe's
build matrix). The tarball doesn't include prebuilt mdbtools and we
build from source via build script.
Signed-off-by: Salvador Cuñat <salvador.cunat@gmail.com>
Starting with Xcode 10, system headers are located inside the
macOS SDK.
Add this location to the check for command line tools.
Signed-off-by: Murillo Bernardes <mfbernardes@gmail.com>
More specifically, don't upload them from the old Windows build - we
just keep that one around for the smtk2ssrf binaries. The Subsurface
binaries are now created in the container based Windows build.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
I expect this to become the default way to test Windows builds and
create installers on Travis. The idea is that instead of downloading the
pre-built MXE binaries we might as well use a container that has all
this installed and can be used locally to test if things fail on Travis;
which will allow us to have the exact same environment for testing
locally as runs on Travis.
At this point the container used is way too big - more effort needs to
be spent on shrinking it.
Right now this only deals with Subsurface and not with smtk2ssrf.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
That qmake -query was added for debugging a long time ago.
Since the comment clearly indicate that the edit of the Makefile is only
for Travis, let's only do it when running on Travis.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
It wasn't documented in the first place (magic first argument, anyone?).
This used to be available for quite a few of the dependency and had
somehow kept around only for Grantlee and Googlemaps. Let's just kill
this and be consistent for all dependencies.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
In some Mac environments autotools somehow think that we have clock_gettime(),
even though it isn't supported. Somehow the previous workaround stopped working
as make ended up re-running ../configure and overwriting our change. This tries
to work around that problem.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Turns out that Jan found an issue with the latest Kirigami, so let's go back to
the known good one.
This reverts commit 17ec95e70c.
Suggested-by: Jan Mulder <jlmulder@xs4all.nl>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Turns out that Jan found an issue with the latest Kirigami, so let's go back to
the known good one.
This reverts commit 40766db459.
Suggested-by: Jan Mulder <jlmulder@xs4all.nl>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Update to the master of today, and no issues detected on mobile-on-desktop
and Android.
Only, the ugly border is back as the magic hack of 0b16b547ae failed
due to the patch file that errored. So that is fixed too.
Signed-off-by: Jan Mulder <jlmulder@xs4all.nl>
Build systems that run from tar balls and not git fail to create valid
.appdata.xml This solves the problem for tar balls that we create for
OBS via our own make-package script. It doesn't solve the problem for
Arch or Gentoo who I believe take our tar files created via git archive.
One way to fix this would be to change the process by which I create
those tar files, I guess.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Instead of editing appdata/subsurface.appdata.xml in place, switch to our usual
pattern of modifying a .in file and add the resulting file to .gitignore.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Use correct format, create script to update the version and release date in the
appdata.
[Dirk Hohndel: call said script during the build process]
Signed-off-by: Alexander Wilms <f.alexander.wilms@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Switching to GitHub as source for libzip means that we need to encode
the version number differently. Newer versions of libzip don't compile
cleanly on Android and this one seems new enough.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This also switches us to libzip's new official home on GitHub, and takes into
account that libzip no longer supports autotools and instead now is cmake
based.
Building against that, on my Mac build system, Subsurface once again correctly
opens DLD files downloaded from divelogs.de.
Fixes#1534
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Clearly something has changed here. When I first tried to use Homebrew on
Travis the update would take often so long that the build would time out. Now
it is nice and fast. So instead of wasting time with the cache let's just use
Homebrew directly.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Doing that confuses the build setup and as a result the library isn't found at
runtime without some fixups.
Fixes#1469
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
But simply ignore when building outside of Travis.
Of course since we are building Android in a container, we need to first pass
the environment variable to the container...
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
These come originally from https://github.com/travis-ci/travis-build and are
available when running on Travis, but not when running inside a Docker
container on Travis as we do in order to build for Android.
The goal is to provide a quasi heart-beat on STDOUT during very long running
commands - without this the wget to download Qt often times out, so that's
where we are going to use this.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
There are actually more datatypes in bash than just strings. One can for
example hold a list of strings in a list, and use that to keep track of
what we're expected to do.
Signed-off-by: Anton Lundin <glance@acc.umu.se>
The previous code assumed everything was good to go if just the tar ball
was there, but if it wasn't unpacked, it all went sideways.
This makes it more robust and to actually handle that the tarball might
just be there.
Signed-off-by: Anton Lundin <glance@acc.umu.se>
We want to be able to roll our own banner image, logo, title, and other
information in the GlobalDrawer - but Kirigami adds an ugly margin
around that. This attempts to remove that margin (but for some reason
there is still a margin on the left side).
This requires the patch command to be installed, but because of the
context sensitivity of the changes, I couldn't figure out how to do
this with sed or perl (which we already require).
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
git clone/checkout code was mostly the same copied over and over.
Move code to a single common function
Signed-off-by: Murillo Bernardes <mfbernardes@gmail.com>
For unknown reasons cloning from github is slower to start if .git
is omitted.
Add .git to all "git clone" statements
Signed-off-by: Jan Iversen <jani@apache.org>
Only mobilecomponents.sh download breeze-icons and kirigami.
This patch is merely to avoid confusion.
Before get-dep-lib in ios/download would download those, and
mobilecomponents would see that they were already downloaded.
Signed-off-by: Jan Iversen <jani@apache.org>
It seems that sometimes the AppImage tooling gets confused and doesn't
create a .zsync file. That should cause Travis to fail.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
When building with -builddeps on a "virgin" mac, configure
of libgit2 could not find libcurl and libssh2
moved building of libcurl and libssh2 in front of libgit2
Signed-off-by: Jan Iversen <jani@apache.org>
Specifically, don't conflate needing libgit2 with the Mac -builddep
argument, and when determining if we need to build libgit2 on Linux,
make sure to also check for a version that we may have built in a
previous run of the build.sh script.
This commit is much easier to understand with
git show -w
as it contains quite a bit of simple indentation change.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The goal is to reduce the overall wall clock time of our test builds.
This secondary Linux target uses the no-sudo container setup with a Qt
5.10.1 backport and tests the build without BT support as well as the
mobile build (but not the full desktop build and doesn't create an
AppImage as the pre-built Qt5.10.1 is missing QtWebKit).
In exchange we remove the no-bt and mobile build from the existing linux
target.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
These are the latest, supported versions for OpenSSL. I have not tested
if Android would work with OpenSSL 1.1.
The lower versio for libzip is still required as libzip 1.2.0 doesn't
compile for Android.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Android build uses wget and not curl.
Currently android uses older versions of some libs, due to that
a new switch is added temporary.
Old version tested for curl and if not found set the different versions,
however this is not a stable method for obvious reasons
Signed-off-by: Jan Iversen <jani@apache.org>
get-dep-lib has been integrated into build.sh, which
ensures different platforms use the same versions etc.
travis is a frequent user of build.sh, but on a mac it runs
without -build-deps and instead used cached versions of the
library. This setup is alo supported
Signed-off-by: Jan Iversen <jani@apache.org>
Secured git work only happens when cloning a dir.
There is a version for hidapi, but scripts/build.sh
(which are the source for that part of depend) has a
line:
# there is no good tag, so just build master
and then continues with master, so it should be good
enough for get-dep-lib as well.
Signed-off-by: Jan Iversen <jani@apache.org>
Since about a year ago qt-android-cmake shifted to using gradle instead
of ant, and the android sdk's stopped supporting ant to.
Signed-off-by: Anton Lundin <glance@acc.umu.se>
This gets delete dive working properly.
Kirigami passive notification ends up hijacking area where the
"Add dive" or "Delete" or "Discard" buttons are shown. So after
deleting a dive the "Undo" button from the notification
keeps handling the touch events even when not visible.
Signed-off-by: Murillo Bernardes <mfbernardes@gmail.com>
For reasons I cannot explain, running configure for libdivecomputer claims that
certain feature tests pass, even though those features demonstrably aren't
there. This is happening for two compiler warning flags (-Wrestrict &
-Wno-unused-but-set-variable) as well as the test for clock_gettime.
To work around this, we manually edit the config.h file and the created
Makefile before building libdivecomputer.
This happened on macOS 10.11.6 with clang-800.0.42.1 (part of Xcode 8.2.1).
Tangentially related to:
See #1263
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
