The DiveListModelBase::clear() implementation was flawed: It cleared
the data in the core, but left the data in the model untouched.
The code was relying on the fact that the caller would reset the
model manually. Not a good idea.
Therefore, clear the internal data to keep the model consistent at
all times.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The way this was accessed via Qt's model semantics was horrible.
This gives arguably more readable code, since we don't have to
shoehorn things through QVariants.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Split out the actual filtering from the MultiFilterSortModel.
Create a DiveFilter class that does the actual filtering.
Currently, mobile and desktop have their own version of this
class, though ultimately we may want to merge them.
The idea here is that the trip-model and undo-commands have
direct access to the filter-function and thus can take care
of keeping track of the number of shown dives, etc.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Introduce a DiveTripModelBase::clear() function that cleanly
clears all dive data inside a beginResetModel()/endResetModel()
pair. Thus, the UI will be cleanly reset and we can remove
explicit calls to
- graphics->setEmptyState()
- mainTab->clearTabs()
- mainTab->clearTabs()
- diveList->reload()
from MainWindow::closeCurrentFile().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
DiveTripModelTree::topLevelChanged() has pretty complex code, as
it has to handle the fact that when adding/removing a dive from
a trip, the trip can change its position.
The code did not account for the fact that when moving an object
back in the top level list, one has to subtract one from the new
index, because the object was removed somewhere in the front of
the list.
To make matters worse, when an entry stayed where it was, this
was realized by moving the entry right behind itself, which of
course means that it stays where it is. But this meant that in
the by far most common case (no moving) the wrong entry was
updated.
Fix this by subtracting 1 from the new index when moving an
entry to the back.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
In the future we might want to use undo-commands for mobile as
well (even if not implementing undo).
Therefore, move the undo-command source from desktop-widgets
to their own commands top-level folder.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
To test whether an entry is a trip, we passed a pointer to the
trip through a QVariant and tested that for null-ity.
Passing pointers through QVariants has given us myriads of
problems in QML, therefore introduce a bool IS_TRIP_ROLE
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Instead of accessing the cylinder table directly, use the get_cylinder()
function. This gives less unwieldy expressions. But more importantly,
the function does bound checking. This is crucial for now as the code
hasn't be properly audited since the change to arbitrarily sized
cylinder tables. Accesses of invalid cylinder indexes may lead to
silent data-corruption that is sometimes not even noticed by
valgrind. Returning NULL instead of an invalid pointer will make
debugging much easier.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Instead of using fixed size arrays, use a new cylinder_table structure.
The code copies the weightsystem code, but is significantly more complex
because cylinders are such an integral part of the core.
Two functions to access the cylinders were added:
get_cylinder() and get_or_create_cylinder()
The former does a simple array access and supposes that the cylinder
exists. The latter is used by the parser(s) and if a cylinder with
the given id does not exist, cylinders up to that id are generated.
One point will make C programmers cringe: the cylinder structure is
passed by value. This is due to the way the table-macros work. A
refactoring of the table macros is planned. It has to be noted that
the size of a cylinder_t is 64 bytes, i.e. 8 long words on a 64-bit
architecture, so passing on the stack is probably not even significantly
slower than passing as reference.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
When editing the dive site, for certain fields a divesChanged signal
was emitted so that the dive-list can be updated.
Arguably it is wrong to decide which fields are relevant to the
dive list in the undo-command code. Therefore, let the list
catch the dive-site-edited signal and decide itself.
But the actual reason for this commit is that if the dive-site
field of a dive changes, we might have to reload the dive-location-model
because suddenly a new dive site appears. Now if this is done
in QML context on some Qt version (notably 5.9) we get crashes
later on. But that can happen if the user moves a flag. So in that
case only send a diveSiteChanged signal.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
When pressing F2 in the dive list, the number can be edited.
Make this action undoable by implementing a EditNumber command.
This command is differs from the other undo commands, as not the
currently selected dives are changed. This means that the EditCommand
needs an alternative constructor taking a single dive. This constructor
was implemented in the base class so that all edit commands can now
be called with a single dive.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
For historic reasons, there where three distinct signals concerning
dive-selection from the undo-machinery:
1) divesSelected: sent newly selected dives
2) currentDiveChanged: sent if the current dive changed
3) selectionChanged: sent at the end of a command if either the selection
or the current dive changed
Since now the undo-commands do a full reset of the selection, merge these
three signals into a single signal.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Some commands tried to retain the current selection on undo/redo,
others set the selection to the modified dives.
The latter was introduced because it was easier in some cases, but
it is probably more user-friendly because the user gets feedback
on the change.
Therefore, unify to always select the affected dives on undo()/redo().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Since the default view is batched by trips, signals were sent trip-wise.
This seemed like a good idea at first, but when more and more parts used
these signals, it became a burden. Therefore push the batching to the
part of the code where it is needed: the trip view.
The divesAdded and divesDeleted are not yet converted, because these
are combined with trip addition/deletion. This should also be detangled,
but not now.
Since the dive-lists were sorted in the processByTrip function, the
dive-list model now does its own sorting. This will have to be
audited.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
We have a very fundamental problem with data-duplication in
core and qt-models. In a particular case, this led to an easily
reproducible crash:
1) An undo command moved the last dive of a trip to another.
2) When an undo-command removed the last dive of
a trip to a different trip, the dive was removed from the
trip in the core. Then, the model was updated.
3) That lead at first to a rearrangement of the trips, because
the trip with the added dive is moved before the trip with
the removed dive.
4) In such a case, the filter-model checks the visibility of
the trip.
5) Since the trip with the removed dive has no dives in the core,
visibility was determined as false.
6) From this point on the mappings of the QSortFilterProxyModel
were messed up. Accesses led to crashes. It is unclear
whether this is a Qt bug or only a QOI issue.
As a quick-fix, cache the visibility flag of trips directly
in the Qt-models. Don't set the visibility directly in the
core, but go via the Qt-models. Thus, a more clear layering
is achieved.
In the long run, we can hopefully get rid of the data-duplication
in the models.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
These functions were spread out over dive.c and divelist.c.
Move them into their own file to make all this a bit less monolithic.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Make dive.h a bit slimmer. It's only a drop in the bucket - but at
least when modifying tag functions not the *whole* application is
rebuilt anymore.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This is just minor fixes that are not user-visible:
Fix a few erroneous comments and a debug message. These are
copy & paste mistakes and mistakes introduced during code-
refactoring.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
At some places we use UTF8 string literals. Therefore, we effectively
only support UTF8 build systems. We might just as well remove all
the other UTF_* macros and use direct string literals.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The filter code is strange: it actually only checks the
dive->hidden_by_filter flag. Thus, before propagating the dive
changed signal, this flag has to be updated. Do this in the
DiveTripModel. Ultimately, this should be refactored.
Moreover, if the filter-flag changed notify the frontend
of a changed trip so that the trip is hidden / unhidden.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
In tree-mode, the trip locations are displayed. Update the corresponding
entries if the trip changed, by hooking into the tripChanged() signal.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DiveTripModel was used to represent both, trip and list views.
Thus many functions had conditionals checking for the current mode
and both modes had to be represented by the same data structure.
Instead, split the model in two and derive them from a base class,
which implements common functions and defines an interface.
The model can be switched by a call to resetModel(), which invalidates
any pointer obtained by instance(). This is quite surprising
behavior. To handle it, straighten out the control flow:
DiveListView --> MultiFilterSortModel --> DiveTripModelBase
Before, DiveListView accessed DiveTripModelBase directly.
A goal of this commit is to enable usage of the same model by mobile
and desktop.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Commit 911edfca71 changed the dive list
on desktop to update positions of trips when adding/removing dives.
A very unlikely case, but necessary for consistency.
For a trip to be moveable down, its index has to be one-less than
the maximum index, which is "items - 1". The code was doubly wrong:
it forget the "1" and checked for less-or-equal instead less-than.
Thus this was effectively an off-by-two error. Fix it.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
These functionality was used by the desktop filter. To unify desktop
and mobile, move it into two new functions in divelist.c
Since one of them is the only caller of is_same_day() move that
likewise into divelist.c and make it of static linkage.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Just as we did for pointer to struct dive_site, make pointers to
struct dive and struct dive_trip "Qt metatypes". This means that
they can be passed through QVariants without taking a detour via
void *.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
If the date of a dive changed, it might be necessary to reorder
the trips, as the date of the trip changed. Although this seems
like an odd usecase, move the trip if necessary, for consistency's
sake.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The when field gives the time of the first dive. Instead of keeping
this field in sync, replace it by a function that determines the time
of the first dive.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
addDivesToTrip() had one level of indentation too much owing
to a copy-and-paste error. Remove.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
To make sorting more controlled, move all sorting functions into
the core. For this, introduce a "dive_or_trip" structure, which
represents a top-level item. Adapt the DiveTripModel accordingly.
There are now three sorting functions:
1) dive_less_than
2) trip_less_than
3) dive_or_trip_less_than
These should be used by all sorting code. By moving them to a
single place, the mess can hopefully be cleaned up.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The dives of each trip were kept in a list. Replace this by a
struct dive_table. This will make it significantly easier to
keep the dives of a trip in sorted state.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Traditionally, the DiveTripModel has its data sorted in opposite
direction to the core-data (chronologically descending vs. ascending).
This bring a number of subtle problems. For example, when filling
the model, trips are filled according to the *last* dive, whereas
later insertion points are according to the ->when value from the
core, which depends on the *first* dive.
As a start of fixing these subtleties, change the sort direction
to reflect the core-data. Ideally, this should lead to a removal
of the redundant data-representation.
Since the model is now sorted in ascending order, sorting has to
be enabled in the DiveListView constructor to reflect the
default-descending order.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The old code always sorted by "ascending" by default. But
because users typically want their new dives top, "ascending"
was defined for NR and DATE, such that it is actually descending.
Turn these around and intitialize these two fields as
default-descending.
This is possible using the Qt::InitialSortOrderRole role
in DiveTripModel::headerData().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DiveTripModels are sorted in *reverse* chronological order.
Therefore, when comparing a dive against a trip, the dive has
to be inserted if the dive has a *later* date. Change the
comparison accordingly.
Reported-by: Jan Mulder <jlmulder@xs4all.nl>
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The generic addInBatches() function is used to add batches of
contiguous sets of dives to the dive-list models. The loop
searching for the end of the batch used the wrong index and
would therefore not properly cut the batches.
Fix this.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The dive list was sorted using the default-sorter of
QSortFilterProxy model. This is mighty inflexible as it
considers only one column. This has the funky effect that
for rows with identical elements, the sort order depends
on the previous sorting.
Implement a lessThan() function in the MultiFilterSortModel,
which simply hands the sorting down to the actual model.
This might be considered a layering violation, but it makes
things so much easier.
Sadly, it seems like the column-to-be-sorted is transported
in the provided indices. Therefore, the comparison is chosen
using a switch for *every* comparison. It would seem much
more logical to set a function pointer once and use that.
Further investigations are necessary.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The data-flow from C-core to list-view is as follows:
C-core --> DiveTripModel --> MultiSortFilterModel --> DiveListView
The control-flow, on the other hand, differs as DiveListView
accesses both MultiSortFilterModel and DiveTripModel, whereas
MultiSortFilterModel is mostly unaware of its source model.
This is in principle legitimate, as the MultiSortFilterModel might
be used for different sources. In our particular case, this is
not so. MultiSortFilterModel is written for a particular use case.
Therefore, model control-flow follow after data-flow: Let MultiSortFilterModel
set its own source model and DiveListView access the MultiSortFilterModel,
which then manages its source model.
This is not bike-shedding, but will enable a more flexible and
higher-performance sorting.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This accessor was never used. This is a small step in splitting
the DiveTripModel in two (list & tree), which means that the
layout is moved up to the view.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
On change of the filter, the headers of non-extended trips were not
updated. Therefore, on filter-finish-event loop over all trips
in DiveTripModel and signal data-changed.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DiveItem and TripItem classes were wrappers around dive * and
dive_trip * used to extract tabular data. With the rework of
DiveTripModel they lost all their state besides the pointer itself.
The usage was:
DiveItem item(d);
item.data(...);
This can now be simplified to the much more idiomatic
diveData(d, ...);
and analoguously for TripItem.
While adapting the data() function to be part of DiveTripModel, change
the
QVariant ret
switch(...) {
...
case ...:
ret = ...;
break;
...
}
return ret;
style to
switch(...) {
...
case ...:
return ...;
}
Not only is this shorter and easier to reason about, it generally also
improves the generated code. The compiler can directly construct the
return value in the buffer provided by the caller. Though modern
compilers start to be very good at avoiding unnecessary copies.
In total this cleanup results in a net-reduction of 190 lines of code.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Ultimately, we want to use a single dive-list and not replicate
it in the Qt-model code. To this goal, let's start with using
the same sort function.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The current code cheats when it comes to move dives inside
a trip or move dives between trips: Instead of using the
*MoveRows() functionality, the dives are removed from and
re-added to the respective trips. This loses the selection.
Therefore, remember which of the moved dives are selected
and select them manually after they are re-added.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The command-objects select a current item, but this selection
was not propagated to the front-end. The current item is the
base for keyboard-navigation through the dive-list and therefore
should be set correctly.
It took some experimentation to get the flags right:
QItemSelectionModel::Current
Hopefully, these are the correct flags across all supported
Qt versions!
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Select the proper dives after the add, remove, split and merge
dives commands on undo *and* redo. Generally, select the added
dives. For undo of add, remember the pre-addition selection.
For redo of remove, select the closest dive to the first removed
dive.
The biggest part of the commit is the signal-interface between
the dive commands and the dive-list model and dive-list view.
This is done in two steps:
1) To the DiveTripModel in batches of trips. The dive trip model
transforms the dives into indices.
2) To the DiveListView. The DiveListView has to translate the
DiveTripModel indexes to actual indexes via its QSortFilterProxy-
model.
For code-reuse, derive all divelist-changing commands from a new base-class,
which has a flag that describes whether the divelist changed. The helper
functions which add and remove dives are made members of the base class and
set the flag is a selected dive is added or removed.
To properly detect when the current dive was deleted it
became necessary to turn the current dive from an index
to a pointer, because indices are not stable.
Unfortunately, in some cases an index was expected and these
places now have to transform the dive into an index. These
should be converted in due course.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Don't delesect dives, when unregistering them from the backend.
If a previously selected dive is added, select it in the dive-list.
For this purpose introduce a SELECTED_ROLE to query the DiveTripModel
for selected dives.
Unfortunately, when adding multiple selected dives, current_dive_changed
is called for each of them, making this very slow. This will have
to be fixed in subsequent commits.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Previously, each dive-list modifying function would lead to a
full model reset. Instead, implement proper Qt-model semantics
using beginInsertRows()/endInsertRows(), beginRemoveRows()/
endRemoveRows(), dataChange().
To do so, a DiveListNotifer singleton is generatated, which
broadcasts all changes to the dive-list. Signals are sent by
the commands and received by the DiveTripModel. Signals are
batched by dive-trip. This seems to be an adequate compromise
for the two kinds of list-views (tree and list). In the common
usecase mostly dives of a single trip are affected.
Thus, batching of dives is performed in two positions:
- At command-level to batch by trip
- In DiveTripModel to feed batches of contiguous elements
to Qt's begin*/end*-functions.
This is conceptually simple, but rather complex code. To avoid
repetition of complex loops, the batching is implemented in
templated-functions, which are passed lambda-functions, which
are called for each batch.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
AddDivesToTrip, CreateTrip, AutogroupDives, RemoveAutogenTrips
and MergeTrips basically all did the same thing as RemoveDivesFromTrip,
which was already implemented. Thus, factor our the common functionality
and hook it up to make all these functions undo-able.
Don't do the autogroup-call everytime the dive-list is rebuilt
(that would create innumberable undo-actions), but only on dive-load /
import or if expressly asked by the user [by switching the autogroup
flag].
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The dive list is fed data by means of a sorted "DiveTripModel".
There are two modes: list and tree. This was implemented rather
elegantly with a general "TreeModel", which can represent trees
of arbitrary depths.
Nevertheless, we have at most two levels and on the second level
only dives can reside. Implementing proper model-semantics
(insert, delete, move) will be quite a challenge and implementing
it under the umbrella of a very general model will not make it
easier.
Therefore, for now, hardcode the model:
At the top-level there are items which may either be a trip
(can contain multiple dives) or a dive (contains exactly one dive).
Thus, we can completely de-virutalize the DiveItem and TripItem
classes, which are now trivial wrappers around dive * and dive_trip *.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
