A strange and buggy dive where time goes backwards (right now easy to
create with the dive plan editor) can cause us to run out of plot info
elements.
This prevents that from causing memory corruption by refusing to go back
in time.
Reported-by: Dirk Hohndel <dirk@hohndel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Now that the pressure_time calculations are done in our "native"
integer units (millibar and seconds), we might as well keep using
integer variables.
We still do floating point calculations at various stages for the
conversions (including turning a depth in mm into a pressure in mbar),
so it's not like this avoids floating point per se. And the final
approximation is still done as a fraction of the pressure-time values,
using floating point. So floating point is very much involved, but
it's used for conversions, not (for example) to sum up lots of small
values.
With floating point, I had to think about the dynamic range in order
to convince myself that summing up small values will not subtly lose
precision.
With integers, those kinds of issues do not exist. The "lost
precision" case is not subtle, it would be a very obvious overflow,
and it's easy to think about. It turns out that for the pressure-time
integral to overflow in "just" 31 bits, we'd have to have pressures
and times that aren't even close to the range of scuba cylinder air
use (eg "spend more than a day at a depth of 200+ m").
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
I fixed the pressure-time calculations to use "proper" units, but
thinking about it some more, it turns out that units don't really
matter. As long as we use the *same* unit for calculating the
integral, and then re-calculating the step-wise entries, the units
will cancel out.
So we can simplify the "pressure_time()" function a bit, and use
whatever units are most natural for our internal representation. So
instead of using atm, use "mbar".
Now, since the units don't matter, this patch doesn't really make much
of a difference conceptually. Sure, it's a slightly simpler function,
but maybe using more "natural" units for it would be worth it. But it
turns out that using milli-bar and seconds has an advantage: we could
do all the pressure_time integral using 32-bit integers, and we'd
still be able to represent values that would be equivalent to staying
at 24 bar for a whole day.
This patch doesn't actually change the code to use integers, but with
this unit choice, we at least have that possibility.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This splits up the function to create the estimated pressures for
missing tank pressure information.
The code now has a separate pass to create the beginning and ending
pressures for segments that lack them, and fill them in to match the
overall SAC-rate for that cylinder.
In the process, it also fixes the calculation of the interpolated gas
pressure: you can see this in test-dive 13, where we switch back to the
first tank at the end of the dive. It used to be that the latter
segment of that cylinder showed in a different color from the first
segment, showing that we had a different SAC-rate. But that makes no
sense, since our interpolation is supposed to use a constant SAC-rate
for each cylinder.
The bug was that the "magic" calculation (which is just the pressure
change rate over pressure-time) was incorrect, and used the current
cylinder pressure for start-pressure calculation. But that's wrong,
since we update the current cylinder pressure as we go along, but we
didn't update the total pressure_time.
With the separate phase to calculate the segment beginning/ending
pressures, the code got simplified and the bug stood out more.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The code was using bar, not atm to calculate the pressure_time
multiplier. But SAC-rate is relative to atm.
We could do the correction at the end (and keep the pressure_time in
"bar-seconds"), but let's just use the expected units during the
integration. Especially since this also makes a helper function to do
the calculations (with variables to keep the units obvious) instead of
having multi-line expressions that have the wrong units.
This fixes what I thought were rounding errors for the pressure graphs.
They were just unit confusion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This splits up the (very complex) function that calculates all the plot
info data, so that the gas pressure logic is in several helper
functions, and the deco and partial pressure calculations are in a
function of their own.
That makes the code almost readable.
This also changes the cylinder pressure calculations so that if you have
manually set the beginning and end pressures, those are the ones we will
show (by making them fake "sensor pressures"). We used to shopw some
random pressure that was related to the manually entered ones only
distantly (through various rounding phases and the SAC-rate calculations).
That does make the rounding errors more obvious in the graph, but we can
fix that separately.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This simplifies - and improves - the code to generate the plot info
entries from the samples.
We used to generate exactly one plot info entry per sample, and then -
because the result doesn't have high enough granularity - we'd
generate additional plot info entries at gas change events etc.
Which resulted in all kinds of ugly special case logic. Not only for
the gas switch, btw: you can see the effects of this in the deco graph
(done at plot entry boundaries) and in the gas pressure curves.
So this throws that "do special plot entries for gas switch events"
code away entirely, and replaces it with a much more straightforward
model: we generate plot entries at a minimum of ten-second intervals.
If you have samples more often than that, you'll get more frequent
plot entries, but you'll never get less than that "every ten seconds".
As a result, the code is smaller and simpler (99 insertions, 161
deletions), and actually does a better job too.
You can see the difference especially in the test dives that only have
a few entries (or if you create a new dive without a dive computer,
using the "Add Dive" menu entry). Look at the deco graph of test-dive
20 before and after, for example. You can also see it very subtly in
the cylinder pressure curves going from line segments to curves on
that same dive.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
For dives with no samples, we crate a fake dive computer with a set of
made-up samples and use those to display the profile.
However, the actual calculations to do the maximum duration and depth
etc were always done with the "real" dive information, which is empty.
As a result, the scale of the plot ended up being bogus, and part of
the dive would be missing.
Trivially fix by just passing the same dive computer information to
calculate_max_limits() that we use for everything else.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This comes with absolutely no gui - so the plan literally needs to be
compiled into Subsurface. Not exactly a feature, but this allowed me to
focus on the planning part instead of spending time on tedious UI work.
A new menu "Planner" with entry "Test Planner" calls into the hard-coded
function in planner.c. There a simple dive plan can be constructed with
calls to plan_add_segment(&diveplan, duration, depth at the end, fO2, pO2)
Calling plan(&diveplan) does the deco calculations and creates deco stops
that keep us below the ceiling (with the GFlow/high values currently
configured). The stop levels used are defined at the top of planner.c in
the stoplevels array - there is no need to do the traditional multiples of
3m or anything like that.
The dive including the ascents and deco stops all the way to the surface
is completed and then added as simulated dive to the end of the divelist
(I guess we could automatically select it later) and can be viewed.
This is crude but shows the direction we can go with this. Envision a nice
UI that allows you to simply enter the segments and pick the desired
stops.
What is missing is the ability to give the algorithm additional gases that
it can use during the deco phase - right now it simply keeps using the
last gas used in the diveplan.
All that said, there are clear bugs here - and sadly they seem to be in
the deco calculations, as with the example given the ceiling that is
calculated makes no sense. When displayed in smooth mode it has very
strange jumps up and down that I wouldn't expect. For example with GF
35/75 (the default) the deco ceiling when looking at the simulated dive
jumps from 16m back up to 13m around 14:10 into the dive. That seems very
odd.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Without this the cairo_close_path call could do silly looking things
(intersecting polygons...).
Reported-by: "Robert C. Helling" <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The old implementation was broken in several ways.
For one thing the GF values are percentages, so they should normally be
0 < GF < 1 (well, some crazy people like to go above that).
With this most of the Bühlmann config constants were wrong.
Furthermore, after we adjust the pressure tolerance based on the gradient
factors, we need to convert this back into a depth (instead of passing
back the unmodified depth - oops).
Finally, this commit adds closed circuit support to the deco calculations.
Major progress and much more useful at this stage.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This also initializes the N2 tissue saturations to correct numbers
(setting them to zero was clearly silly).
With this commit we walk back in the dive_table until we find a surface
intervall that's longer than 48h. Or a dive that comes after the last one
we looked at; that would indicate that this is a divelist that contains
dives from multiple divers or dives that for other reasons are not
ordered. In a sane environment one would assume that the dives that need
to be taken into account when doing deco calculations are organized as one
trip in the XML file and so this logic should work.
One major downside of the current implementation is that we recalculate
everything whenever the plot_info is recreated - which happens quite
frequently, for example when resizing the window or even when we go into
loup mode. While this isn't all that compute intensive, this is an utter
waste and we should at least cache the saturation inherited from previous
dives (and clear that number when the selected dive changes). We don't
want to cache all of it as the recreation of the plot_info may be
triggered by the user changing equipment (and most importantly, gasmix)
information. In that case the deco data for this dive does indeed have to
be recreated. But without changing the current dive the saturation after
the last surface intervall should stay the same.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Usually dive computers show the ceiling in terms of the next deco stop -
and those are in 3m increments. This commit also adds the ability to chose
either the typical 3m increments or the smooth ceiling that the Bühlmann
algorithm actually calculates.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This is on top of the deco information reported by the dive computer (in a
different color - currently ugly green). The user needs to enable this via
the Tec page of the preferences.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The existing code had the somewhat retarded Ctrl-C binding for displaying
the next divecomputer and no way to go back to the previous one. With this
commit we use our keyboard grab to map Left and Right to previous and next
divecomputer. Much nicer.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This clarifies/changes the meaning of our "cylinderindex" entry in our
samples. It has been rather confused, because different dive computers
have done things differently, and the naming really hasn't helped.
There are two totally different - and independent - cylinder "indexes":
- the pressure sensor index, which indicates which cylinder the sensor
data is from.
- the "active cylinder" index, which indicates which cylinder we actually
breathe from.
These two values really are totally independent, and have nothing
what-so-ever to do with each other. The sensor index may well be fixed:
many dive computers only support a single pressure sensor (whether
wireless or wired), and the sensor index is thus always zero.
Other dive computers may support multiple pressure sensors, and the gas
switch event may - or may not - indicate that the sensor changed too. A
dive computer might give the sensor data for *all* cylinders it can read,
regardless of which one is the one we're actively breathing. In fact, some
dive computers might give sensor data for not just *your* cylinder, but
your buddies.
This patch renames "cylinderindex" in the samples as "sensor", making it
quite clear that it's about which sensor index the pressure data in the
sample is about.
The way we figure out which is the currently active gas is with an
explicit has change event. If a computer (like the Uemis Zurich) joins the
two concepts together, then a sensor change should also create a gas
switch event. This patch also changes the Uemis importer to do that.
Finally, it should be noted that the plot info works totally separately
from the sample data, and is about what we actually *display*, not about
the sample pressures etc. In the plot info, the "cylinderindex" does in
fact mean the currently active cylinder, and while it is initially set to
match the sensor information from the samples, we then walk the gas change
events and fix it up - and if the active cylinder differs from the sensor
cylinder, we clear the sensor data.
[Dirk Hohndel: this conflicted with some of my recent changes - I think
I merged things correctly...]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This commit makes deco handling in Subsurface more compatible with the way
libdivecomputer creates the data. Previously we assumed that having a
stopdepth or stoptime and no ndl meant that we were in deco. But
libdivecomputer supports many dive computers that provide the deco state
of the diver but with no information about the next stop or the time
needed there. In order to be able to model this in Subsurface this adds an
in_deco flag to the samples. This is only stored to the XML file when it
changes so it doesn't add much overhead but will allow us to display some
deco information on dive computers like the Atomic Aquatics Cobalt or many
of the Suuntos (among others).
The commit also removes the old event based deco code that was commented
out already. And fixes the code so that the deco / ndl information is
stored for the very last sample as well.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
We only store the model/deviceid/nickname for those dive computers that
are mentioned in the XML file. This should make the XML files nicely
selfcontained.
This also changes the code to consistently use model & deviceid to
identify a dive computer. The deviceid is NOT guaranteed to be collision
free between different libdivecomputer backends...
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Having it there with the model information seemed to make sense but on
second thought it's the wrong spot to keep that information, especially
since we were storing it in the XML file in every single dive.
This change removes the nickname member from the divecomputer and makes
the rest of the code reasonably self consistent. It does not add much of
the new code for the new design to handle nicknames.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This adds the capability to actually view all your dive computers, by
adding a menu item under "Log"->"View"->"Next DC" to show the next dive
computer.
Realistically, if you actually commonly use this, you'd use the
accelerator shortcut. Which right now is Ctrl-C ("C for Computer"),
which is probably a horrible choice.
I really would want to have nice "next/prev dive" accelerators too,
because the cursor keys don't work very well with the gtk focus issues.
Being able to switch between dives would also make the "just the dive
profile, maam" view (ctrl-2) much more useful.
The prev/next dive in the profile view should probably be done with a
keyboard action callback, which also avoids some of the limitations of
accelerators (ie you can make any key do the action). Some gtk person,
please?
Anyway, this commit only does the dive computer choice thing, and only
using the accelerators.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This changes two things to improve the appearance of the profile:
- the partial pressure scale is now in 0.5 increments if the total is <= 4
and in 1.0 increments if it is > 4.
- the depth marker lines end slightly below the depth chart so that we no
longer have overlap between the depth scale and the partial pressure
scale.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
In profile.c:create_plot_info(), store the last address in which
memory was allocated for the plot data entries in the static
variable "last_pi_entry". If "last_pi_entry" isn't a NULL
pointer in each call to create_plot_info(), free memory at that
address.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
We maintain a list of dive computers that we know about (by deviceid) and
their nicknames in our config. If the user downloads dive from a dive
computer that we haven't seen before, we give them the option to set a
nickname for that dive computer. That nickname is displayed in the profile
(and stored in the XML file, assuming it is not the same as the model).
This implementation attempts to make sure that it correctly deals with
utf8 nicknames.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Showing the depth scale all the way to the bottom of the profile plot
looks strange when there are partial pressure graphs down there. So
instead we only plot down to the next marker below the maximum depth of
the actual dive.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
I foolishly changed visible_columns in both the (ill-named) cns branch and
master...
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Conflicts:
divelist.c
gtk-gui.c
profile.c
This splits the dive time, depth, pressure and temperature maxima and
minima setup from the per-dive-computer "create_plot_info()" function
into one setup function that walks _all_ the dive computers, so that we
have a global maxima and minima.
That way the graph scaling we set up will now fit the data from all dive
computers rather than just the particular one we are plotting. So if
you switch back-and-forth between computers, the scale (which is defined
by the extremes) remains the same.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Now we can simply remember the state of all the preferences at the
beginning of preferences_dialog() and restore them if the user presses
'Cancel'.
Fixes#21
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Instead provide a scale on the right in a highly transparent grey and rely
on the tooltip available with mouse-over to pinpoint the value at certain
spots with much better accuracy.
Fixes#30
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Some dive computers appear to tell us the gas used in a gaschange event
right at the beginning of the dive. We arbitrary have a cut-off that says
"a gas change in the first 30 seconds shouldn't get a marker".
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Regardless what the dive computer tells us, don't believe that pO2 was
higher than the ambient pressure. This gives much more realistic values.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This now takes the He percentage into account when matching tanks. Given
the encoding of fO2 and fHe in the GASCHANGE2 event, we can simply mask
and shift as if we have a GASCHANGE2 event and things will automatically
work correctly for the regular GASCHANGE event.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Tec divers seem to like a little bit more precision - and the dive
computers certainly provide it (or we can calculate it).
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
While having the background "come down" seemed like a good visualization
of the ceiling, some divers appear to prefer something more dramatic. This
adds an option to the Tec Settings to have the ceiling shown in red
instead of the default background color.
Suggested-by: Jan Schubert <Jan.Schubert@GMX.li>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This way they don't clash with the dive computer model information that
was added by commit a23ec27ca7bb "Add dive computer name to the dive
plot".
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
.. and then allocate just the plot-info entry array dynamically.
We want to have a longer lifetime for the basic plot_info data
structure, because we want to do computer selection and maximum
time/depth/temperature computations *before* we start plotting anything,
and before we allocate the plot entry array.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
I want to have some way to show multiple dive computers, so start off by
adding the name of the current one. So if we change dive computers,
we'll see it.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This commit changes the code that was recently introduced to deal with
deco ceilings. Instead of handling these through events we now store the
ceiling (which in reality is the deepest deco stop with all known dive
computers) and the stop time at that ceiling in the samples.
This also adds support for NDL (non stop dive limit) which both dive
computers that appear to give us ceiling / deco information appear to
give us as well (when the diver isn't in deco).
If the mouse hovers over the profile we now add support for displaying the
NDL, the current deco obligation and (if we are able to tell from the
data) whether we are at a safety stop.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
I was a little too eager to add the deco feature to Subsurface. Jef and I
went back and forth a few more times and the definition of those events
changed. I guess I shouldn't have commited that code until the
corresponding libdivecomputer code had been pushed.
This commit now brings us in sync with the current master of
libdivecomputer (but should compile with 0.2 as well - only deco events
won't work then).
One issue that I see is that deco / ndl aren't really a good fit for the
event model. I actually disabled the drawing of the little yellow
triangles for ndl events as for example on the Uemis those events are
created whenever the remaining non stop time changes - and that can be
every few seconds.
The correct solution may be to treat this as a function of the samples,
but for now this works and is tested with both OSTC and Uemis SDA.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This actually triggers for one of our insane test dives (test15): it has
no samples, so we created a fake dive computer entry with the a fake
profile in it, but we didn't copy the events over.
Having a dive with no samples, yet having events from the dive computer,
sounds pretty bogus. But that test-case did show that when that bogus
situation happens, we had two independent buglets: (a) we didn't insert
the entries in the fake dive computer entry we used and (b) we would
then mix up the events of the fake dive computer entry with the first
dive computer of a dive.
Fix this, just to make test15 happy again. And eventually, when we
actually plot the information for multiple dive computers, fixing case
(b) would become necessary even for real dives.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
For now we only have one fixed divecomputer associated with each dive,
so this doesn't really change any current semantics. But it will make
it easier for us to associate a dive with multiple dive computers.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Prervent tiny temperature changes from being exaggerated in the plot.
Also, shift pressure plot around a bit (if necessary) to prevent it from
ending in the same space as the temperature plato on the profile graph.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The code in commit f99e1b476b18 "Trim the dive to exclude surface time at
beginning and end" failed rather badly if a dive has no samples at all -
which is true for many of our test dives.
This makes sure that we don't exclude data points if we never set up start
and end times.
Reported-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
