We still need to support Qt 5.9 for Ubuntu 18.04 / Bionic. This uses
deprecated calls to qrand() - but then qrand() wasn't deprecated in 5.9.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The old test was broken in many ways and kept failing for a number of
reasons. Some of them were addressed in the previous commits (the
missing HEAD ref being the main one), the other one was that the tests
kept stepping on top of each other - as were potentially random users or
reviewers using the 'universal' test account.
This uses a random one of ten dedicated test accounts, and on top of
that uses a random branch name (instead of the fixed email address
associated with the account).
This also rewrites several of the tests dealing with offline changes to
correctly model going offline (but keeps one that directly writes to the
local cache).
Finally this change also tries to do a much better job cleaning up after
itself and not leaving data behind the the next run could stumble over.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This will be used by the test to clean up test branches that are created
on the server. Since we aren't testing that functionality (it's not
something that Subsurface itself ever does) the helper prints out errors
it encounters, but doesn't report them back to the caller.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
If you had one of the unfortunate local git caches with a local HEAD
just pointing to 'master', this will make note of that and then fix it
up to use the proper branch name in the cache repository.
[Dirk Hohndel: demoted from error to fprintf as most users won't care]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
In create_and_push_remote(), we set up the remote tracking etc to use
the proper branch name, but never actually set up the initial local
branch for the new cache repository at all. So the repository would end
up with the default 'master' branch, instead of the branch name it
should have.
This went unnoticed, because most setups start by initializing the git
caches by cloning from the cloud, and that worked fine.
Debugged-by: Dirk Hohndel <dirk@hohndel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some profilewidget classes hat protected members which can
be made private as there is no subclassing.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
There were a number of protected member functions in ProfileWidget2.
However no class subclassed ProfileWidget2, so this appears to have
been an artifact. Therefore, make these functions private.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
In 9310d72943, resetting of the dive
sites was moved to DiveTripModelBase::reset(). This seemed like a
good idea, because it means that the location list is reloaded
every time the dive list is reset.
Unfortunately that function is only used on mobile, thus on desktop
the dive site model is not updated. Do that in
MultiFilterSortModel::resetModel(), because this is always called
when the dive list is reset. Desktop differs from mobile in that
two different models are used depending on whether we are in list
or in tree mode.
Fixes#2749
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This fixes a crash: when the undo commands removed a dive from
the list, the fulltext cache was not cleared. If now the divelist
is reset and then the undo-command deleted, deletion of the owned
dive tries to remove it's fulltext cache, which doesn't exist
anymore.
For reasons of symmetry, when readding the dive, its fulltext
has to be registered.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Calling qApp->processEvents() in QMLManager::setNotificationText()
caused crashes, because it could lead to the context-menu that
initialized the call being deleted. Something that QML apparently
doesn't like.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
On startup, we showed progress of the population of the fulltext
and listmodels for every 100th dive. This worked by overwriting
the last line if the new line started with '\r'. Since we don't
do that anymore, we can remove this special case.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
When initializing the fulltext-cache and the dive-list, every
100 dives a notification was shown. I had a feeling that this
made startup significantly slower, but that could have been
purely psychological.
Therefore I measured and indeed, removing the fine-grained
notification, it becomes *significantly* faster. For a 3500
dives test log with mobile-on-desktop:
Initialization of the fulltext: 1350 ms -> 730 ms (-46%)
Initialization of the divelistmodel: 689 ms -> 113 ms (-83%)
Let's remove the fine-grained notification. There *is* a visual
indication of work-in-progress anyway.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The compiler complained about this and it seems the
function does not need it.
Additional-test-suggested-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Signed-off-by: Robert C. Helling <helling@atdotde.de>
The planner does not know about events except gas
changes. But if the dive comes from the log, we
should preserve the dive computer events. At least
those that happend before we started to delete
waypoints to let the planner take over.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
When creating a TankInfoDelegate editor, reploting of the profile
was disabled to avoid replotting when the user scrolls through
the tank-info list. Since the code was changed to only set the
tank-info when the editor is closed, this became unnecessary
(hopefully). Indeed the clearing of the flag was removed in a
previous commit. This means that we also have to remove the setting
of the flag. Since this is all the TankInfoDelegate::createEditor()
function was doing, we can remove the whole function.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
MainWindow::on_actionAddDive_triggered() updated the profile after
calling Command::addDive(). However, that is redundant because the
undo-machinery does the profile reload. Remove the call.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This fixes a bug: when deleting a picture when multiple dives
were selected, possibly the wrong dive was invalidated.
Thus, the dive wouldn't have been saved to the git repository.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DivePictureModel kept a pointer to picture for each entry.
Firstly, this is dangerous from a data-consistency point of view.
Secondly, the entry wasn't even used anywhere. Remove it.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This isn't really a useful performance improvement, but it's still better,
IMHO, because we don't have a less specific match later on potentially change
an already executed match.
Because of our coding style the comment covering multiple cases of Pelagic dive
computers now is associated just with the first of those entries. I don't see a
way to do this differently without being in violation of our coding style, so
I'll just keep it like this.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Both Shearwater Petrel and Petrel 2 identify as 'Petrel' as their BT and BLE
names. But only the Petrel 2 supports BLE, thus only the Petrel 2 shows up in
the list of known dive computers on iOS (which supports only BLE but not
BT-only). By switching this around to always pick Petrel 2 we now correctly
detect such a dive computer on iOS.
Fixes#2739
Reported-by: Rick Holcombe <wrh@nc.rr.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Qt hates empty ranges, and even for a non-empty range, this is better
implemented as a reset than a remove.
This fixes a crash that I have been able to create on iOS by rescanning
for devices on the download page.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Instead of assigning to a QVariant ret and returning at the end,
return directly in the various switch-cases. This makes the code
more readable, and is more idiomatic C++, as it avoids unnecessary
copies.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
There is the free_picture() function with the same functionality.
The compiler/linker should recognize that and remove the duplicate
code, but still...
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This never made sense and I think I just forgot to complete this code
when I first worked on it. Now we can see which version of Subsurface or
Subsurface-mobile created a merge.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
When pasting cylinders, the destination dive got a verbatim copy
of the cylinders of the source dive. This is not what users want:
for example, the start and stop pressures from the dive computer
as well as the gas mix may be overwritten.
There seems to be no perfect solution, since some times users may
want to paste the gas-mix.
Therefore, let's choose a heuristic approach for now (in the future
we might implement a UI-flag):
When copying over existing cylinders, only copy type and maximum
pressure.
When adding new cylinders (i.e. from-dive has more cylinders than
to-dive), copy everything, but reset start- and top-pressure to 0,
which represents unkown.
Moroever, in the latter case, set "manually added" to true, since
obviously this wasn't added by a dive computer.
Fixes#2676
Reported-by: Miika Turkia <miika.turkia@gmail.com>
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
While having the local user information in the repo on Linux seemed
clever when we implemented it, it's inconsistent with all the other
platforms. Let's just not do that unless the user has indeed set
a global name/email pair for git.
Instead indicate if this was Subsurface or Subsurface-mobile.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This function was not meant to be called with already existing data.
However, if it was, it cleared the words without clearing the fulltext
caches of the dives. This lead to crashes.
Be more resilient by not clearing the words: Already existing dives
are unregistered during the process of populating anyway. So this
now *should* work if new dives are added to the dive list and then
this function is called.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This function was named improperly: it was only used on freshly
loaded data. Indeed, attempts to use it to actually reload lead
to crashes.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
parse_file() refused to load from a git repository if we already
had that repository and there were no changes. However, this only
checked the global divelist_changed flag, which does not track
undo-commands. Thus, after editing dives the user couldn't reload
from git.
Remove this check. It is somewhat questionable that the io layer
refuses to load from a repository anyway. Let the caller decide.
There appears to be a check_git_sha function for that purpose(?).
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The comment states that Qt treats TAB as cancel when in the combobox.
However, testing shows that this use-case works without this hack.
Since it caused weird behavior (the data was set *after* the editor
was closed, leading to inconsistent state), remove it.
Note: this overrides the previous commit, which is therefore redundant
from a history point of view. However, I'll leave the previous commit
in so that if something turns out to break, we can figure out which
of the two changes it was.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
fixTabBehavior() set the editor text *after* closing the editor.
This left us in an inconsistent state where we thought that the
editor is active. By reversing two connects, this problem is resolved.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Parts of the code were not compiled on mobile, because they used
the undo-command infrastructure. However, since mobile now also
compiles that, we might as well remove the conditional compilation.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
