The profile was using a Qt-model to access its data. This means
routing everything through Qt's QVariants and lead to verbose
code such as
double prev_y = dataModel.index(i-1, vDataColumn).data().toReal();
Instead of storing a data-column and do access via a template,
simply store accessor functions. The code from above now reads as
double prev_y = accessor(data[i-1]);
This should also be distinctly faster for the ns-optimizers among
us.
Only one case was somewhat nasty to convert: The accessors for
the 16 tissues are now generated via a recursive template. Thanks
to C++17's constexpr if, such a template is pleasantly easy
to follow, though.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This indicated the data of the horizontal axis. It was (obviously)
always the time axis. Remove.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This label is put to the right of the corresponding curve,
so it should arguably be centered vertically. At least to
me this looks more natural.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The time axis might need some space and the average depth item
puts a formatted depth at to right of the profile. Consider
these when calculating the right border.
Since I found no way to turn of the average depth, this creates
a permanent border, which might or might not be a good thing.
Contains some refactoring of the label-size functions provided
by DiveTextItem.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The old code used the maximum / minimum values of nine-minute
intervals to indicate maximum / minimum depths. This does not
work well when zooming, since the labels will get sparse.
Instead implement a primitive peak finding algorithm, that
searches for the deepest peak in the whole plot and then
repeats the procedure for the right and left sides, leaving
out a certain distance to the origninal peak. This is repeated
until there are no more peaks found.
Only peaks of a certain prominence are considered, which
conveniently gives us the valleys.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Use variable intervals for printing temperature and heart
beat labels. Obviously, so that the labels don't become
sparse on zooming, but also to make them not too crowded
on mobile / small screens.
This doesn't work for depth labels, because these labels
use data provided from the profile.c core that doesn't
know about the size of the chart.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Avoid "overshooting" of the profile items by linearly clipping
the first and last segment to the boundaries of the time-axis.
Sadly, quite a lot of code, because every profile item is
slightly different.
In particular the pressure-segment handling was rewritten.
It now stores the begin and end of each segment to draw
the appropriate text items.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Due to bit-rot the gas pressure and gas type were displayed on top
of each other. I don't understand the meaning of the old code
[log10(log10(axisRange))] (!). Therefore let's just add the height
of the label to separate the labels.
Probably needs some fine-tuning.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Only plot the zoomed range. Currently this passes the sample
before and after the range, so it generally "overshoots" by
one sample in each direction. The plan is to do clipping
on the first and last polygon segment later on.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The tissue percentages were realized as 16 independent polygons.
That didn't work at all with the new absolute scaling.
Reimplement the item and blast it onto a pixmap. Not only is
this artifact-free, it also should (hopefully) be quite a bit
more efficient than painting numerous lines.
In contrast to the old code, this does access the plot_info
structure directly instead of using the model. Not so much
for performance reason, but rather to make things more robust:
We have a strongly typed language. Why would we shoehorn data
through the weakly typed QVariant and mess with wierd
index-arithmetics. Makes no sense to me. Qt-model have to
be used for interfacing with Qt. They are terrible for
intra-application data transfer.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The text and the brush are the two properties of text items
that change dynamically. To avoid complexities concerning
redrawing, set them concurrently instead of in two separate
calls.
Since setting one of the properties requires a full redraw,
there is no performance advantage in setting them individually.
This fixes a theoretical bug: the colors of axis labels were not
updated appropriately. However, it seems like value-dependent
labels weren't used anyway.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Alignment and scale of DiveTextItems are never changed. Therefore,
pass them at construction time. This makes things much easier
if we want to cache the rendered text [currently the text is
rerendered at every paint() event].
This also removes the "parent=0" default parameter of the
constructor, because inadvertently leaving out the last argument
led to a subtle bug.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The printFontScale is used to scale up fonts (and icons) when
rendering to high-DPI devices. With absolute scaling, this
will also be used to scale the size of different chart
regions, line thickness, etc. Therefore, give it an more
appropriate name. "Device pixel ratio", which is a well
established term, seems to appropriately describe the
concept.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Instead of intializing the text fields and then changing the
font scale via signal-rigmarole, pass down the font-scale
at construction time.
Since the fontPrintScale is only set in print mode, we also
can access it directly instead of testing for printMode.
Since the DiveTextItem is not updated using signals anymore,
the connected flag can be removed.
The commit is larger than I had hoped for, but this makes
things ultimately less brittle.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This was just a stub to make the setVisible() function a "slot".
Since there are no more signals using it, remove it.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The visibility is set on every redraw, which is called when
the preferences change. No need for these signals.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DiveCalculatedCeiling had a back-pointer to the profileWidget.
This was used for weird control-flow shenanigans, which were
removed in 975c123a30.
Remove this now useless member variable.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The "in_planner" condition was inadvertently inverted in
c6d78bc134 and therefore the wrong data was used to draw
the line (density instead of SAC). Revert to original.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
To remove reliance on global state, pass an "in_planner" argument
to AbstractProfilePolygonItem::replot(). Thus, calls to in_planner()
can be removed.
This is a bit sad, since the in_planner argument is now passed
to numerous replot() reimplementations of classes derived
from AbstractProfilePolygonItem. However, it is only needed
for one, viz. DiveGasPressureItem. Well, perhaps in the future
more features will depend on the planner mode...
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Instead of accessing the global displayed_dive variable,
pass the dive to the various profile items. This is a
step in making the profile code reentrant.
This removes the last user of the displayed_dc macro,
which can now be removed.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The profile item that shows the ceilings adds a warning event
if the ceiling is violated. This is very unfortunate.
Improve this situation by adding the event up to the function
that calculates the ceiling. This is still not how it should
be - the display layer should not modify the dive that it
displays.
To make this clear, add a comment that details that this
is a contract between planner and display layer: The planner
uses a dive that can be trampled upon by the profile.
Still, this should be solved differently.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This was used to force a replot on preferences changes.
However, the profile now does a replot in such a case
by itself. This can be removed.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
With the same argument as for DivePercentageItem, move access
to live data out of the paint() function. Instead, calculate
colors in replot(), where the other data are calculated.
This is slightly more complicated than in DivePercentageItem,
since there are multiple polygons. Therefore, replace QPolygonF
by a vector of structures contained the position and color
of the data point.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DivePercentageItem is a polygon-item with a custom paint()
method. Calculation of the polygon is done once in replot(),
but calculation of the corresponding colors is done in every
paint() call. The problem is, we have no control over paint().
It is called whenever Qt feels like. Therefore using live
dive data is a dangerous proposition if we ever want to get
rid of the global displayed_dive.
Do all the calculations in replot(). Store the colors in an
additional array of the same size as the polygon.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The old mechanism to replot the profile items was to listen
to model-change signals. Then the code checked whether it
actually had to update anything by looking at the changed
model-indices.
However, the crucial replot was always initialized with
emitDataChanged(), which simple invalidated the full model
and therefore shouldCalculateStuff() always returned true.
Since now the replot() is called explicitly, remove the whole
logic and simply rename modelDataChanged() to replot().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Instead of listening to the dive-data-model changed and
axis changed signals, update the profile items explicitly
once per plot() call. This avoids double replotting of the
dive items.
The old code had at least two replots per plot() call:
one after profileYAxis()->setMaximum() and one after
dataModel->emitDataChanged().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
In contrast to most other items, which are cleared in the
setEmptyState() function, the profile items are cleared
indirectly via a signal from the model. Very hard to follow
and indeed, I thought I could just remove the slot.
Do this explicitly instead for deterministic code.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
settingsChanged() is a virtual function, which is called
when the preferences dialog signals changes. In most derived
classes, the function does nothing.
In two classes, DiveProfileItem and DiveCalculatedTissue, it
replots the item respectively changes its visibility.
However, these two flags are *not* controlled by the preferences
dialog. Indeed, the functions are also connected to finer-grained
qPref signals. Therefore, settingsChanged() can be removed.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The profile items had a "setModel()" function to set
the DivePlotDataModel post creation. The model is never
changed. It does however mean that the model might be
null in a short period between construction and setting
the model.
To simplify reasoning about this code, set the model
in the constructor. To drive the point home that the
can never change and cannot be null, turn it into a
reference.
Yes, this is gratuitous bike-shedding, but it helps
me analysis the code.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The DiveCalculatedCeiling profile-item has a recalc()
function, which calls "dataModel->calculateDecompression()".
This is a questionable reversal of control-flow: The
profile-item should paint the model-data not change it.
The code was supposed to be called under two conditions:
1) The value of the calcceiling3m preferences flag changed.
This code was buggy for two reasons: Firstly, the cached
value was always initialized to false, which means that
sometimes the first call was missed. Secondly, the
settingsChanged() functions was only called when closing
the preferences window, not when changing the flag in the
profile widgets.
2) The datetime of the dive changed. The whole control-flow is
pretty absurd (due to "bit rot"):
- The replan-dive command sends a date-time changed signal.
- The main tab changes the date-time and informs the profile.
- The profile sends a signal to the item.
- The item instructs the model to recalculate the
decompression.
- The model causes the profile to be redrawn.
In any case, the whole thing is moot, because the decompression
is recalculated for *every* profile plot in create_plot_info_new().
Let's remove the code from the DiveCalculatedCeiling profile-item
and the calculateDecompression() function, which is now not
used anymore.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
So far, the PreferencesDialog emitted a settingsChanged signal.
This meant that models that listened to that signal had to
conditionally compile out the code for mobile or the connection
had to be made in MainWindow.
Instead, introduce a global signal that does this and move
the connects to the listeners to remove inter-dependencies.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Now, that we have this helper function that should have been
introduced long ago, we can make some more expressions
more idiomatic.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
This adds a tab for dive log - related preferences.
A suitable test programs is still required.
Signed-off-by: willemferguson <willemferguson@zoology.up.ac.za>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
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>
The goal here is to make it possible to detach the pressure related
data from the plot_info structure. Thus, the pressure related data
can be allocated independently depending on the number of cylinders
per dive.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
When keeping track of cylinder related data, the code was using
static arrays of MAX_CYLINDERS length. If we want to use dynamically
sized cylinder arrays, these have to be dynamically allocated.
In C++ code, this is trivial: simply replace the C-style arrays
by std::vector<>. Don't use QVector, as no reference counting or
COW semantics are needed here. These are purely local and unshared
arrays.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The planner can produce negative cylinder pressures when
more gas is used than available. Let's color the pressure
graph in a highly visible color to alert the user of the
fact that current gas planning is insufficient.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
There is absolutely no reason to use a macro here.
The only argument that can be made is consistency with
the other pressure-macros, but those too are questionable.
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>
Remove a few cases of
void fun() {
...
}
While touching these functions, fix a few other whitespace
coding style violations.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Commit f5b11daffd changed gasmix
arguments and return values to be passed by value instead of
using pointers.
Notably, get_gasmix() is fed a default-value and returns a
new value. In the old code, NULL was passed in in a first
loop iteration and non-NULL was always returned in the first
iteration. Thus, an equality comparison of passed-in an
returned gasmix would always fail in the first loop iteration.
The new code passed in air as default. Now if air was also
returned, then the matching gases were not calculated in
calculate_sac(). To revert to the old behavior, pass in
an invalid gasmix.
Moreover, give names to the invalid and air gasmixes.
Reported-by: tormento <turment@gmail.com>
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
These were pointers into the global prefs object. The user must
not use these to modify the settings, therefore make them
pointers-to-const.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This is another entry in the series to make more things
"const-clean" with the ultimate goal of merge_dive() take
const pointers.
This concerns functions taking pointers to events and
the fallout from making these const.
The somewhat debatable part of this commit might be
that get_next_event() is split in a two distinct
(const and non-const) versions with different names,
since C doesn't allow overloading. The linker should
recognize that these functions are identical and remove
one of them.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
