mirror of
https://github.com/subsurface/subsurface.git
synced 2024-11-28 05:00:20 +00:00
f3c7dcf9c9
Fix the filters for planned (i.e. has at least one dive plan attached) and logged (i.e. has at least one dive computer log attached) dives. Also refactor the respective functions for improved readability. Signed-off-by: Michael Keller <github@ike.ch>
3534 lines
96 KiB
C++
3534 lines
96 KiB
C++
// SPDX-License-Identifier: GPL-2.0
|
|
/* dive.cpp */
|
|
/* maintains the internal dive list structure */
|
|
#include <string.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <limits.h>
|
|
#include <memory>
|
|
#include "dive.h"
|
|
#include "gettext.h"
|
|
#include "subsurface-string.h"
|
|
#include "libdivecomputer.h"
|
|
#include "device.h"
|
|
#include "divelist.h"
|
|
#include "divelog.h"
|
|
#include "divesite.h"
|
|
#include "errorhelper.h"
|
|
#include "event.h"
|
|
#include "extradata.h"
|
|
#include "interpolate.h"
|
|
#include "qthelper.h"
|
|
#include "membuffer.h"
|
|
#include "picture.h"
|
|
#include "sample.h"
|
|
#include "tag.h"
|
|
#include "trip.h"
|
|
#include "structured_list.h"
|
|
#include "fulltext.h"
|
|
|
|
// For user visible text but still not translated
|
|
const char *divemode_text_ui[] = {
|
|
QT_TRANSLATE_NOOP("gettextFromC", "Open circuit"),
|
|
QT_TRANSLATE_NOOP("gettextFromC", "CCR"),
|
|
QT_TRANSLATE_NOOP("gettextFromC", "pSCR"),
|
|
QT_TRANSLATE_NOOP("gettextFromC", "Freedive")
|
|
};
|
|
|
|
// For writing/reading files.
|
|
const char *divemode_text[] = {"OC", "CCR", "PSCR", "Freedive"};
|
|
|
|
static double calculate_depth_to_mbarf(int depth, pressure_t surface_pressure, int salinity);
|
|
|
|
/*
|
|
* The legacy format for sample pressures has a single pressure
|
|
* for each sample that can have any sensor, plus a possible
|
|
* "o2pressure" that is fixed to the Oxygen sensor for a CCR dive.
|
|
*
|
|
* For more complex pressure data, we have to use explicit
|
|
* cylinder indices for each sample.
|
|
*
|
|
* This function returns a negative number for "no legacy mode",
|
|
* or a non-negative number that indicates the o2 sensor index.
|
|
*/
|
|
extern "C" int legacy_format_o2pressures(const struct dive *dive, const struct divecomputer *dc)
|
|
{
|
|
int i, o2sensor;
|
|
|
|
o2sensor = (dc->divemode == CCR) ? get_cylinder_idx_by_use(dive, OXYGEN) : -1;
|
|
for (i = 0; i < dc->samples; i++) {
|
|
const struct sample *s = dc->sample + i;
|
|
int seen_pressure = 0, idx;
|
|
|
|
for (idx = 0; idx < MAX_SENSORS; idx++) {
|
|
int sensor = s->sensor[idx];
|
|
pressure_t p = s->pressure[idx];
|
|
|
|
if (!p.mbar)
|
|
continue;
|
|
if (sensor == o2sensor)
|
|
continue;
|
|
if (seen_pressure)
|
|
return -1;
|
|
seen_pressure = 1;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Use legacy mode: if we have no O2 sensor we return a
|
|
* positive sensor index that is guaranmteed to not match
|
|
* any sensor (we encode it as 8 bits).
|
|
*/
|
|
return o2sensor < 0 ? 256 : o2sensor;
|
|
}
|
|
|
|
/* warning: does not test idx for validity */
|
|
extern "C" struct event *create_gas_switch_event(struct dive *dive, struct divecomputer *dc, int seconds, int idx)
|
|
{
|
|
/* The gas switch event format is insane for historical reasons */
|
|
struct gasmix mix = get_cylinder(dive, idx)->gasmix;
|
|
int o2 = get_o2(mix);
|
|
int he = get_he(mix);
|
|
struct event *ev;
|
|
int value;
|
|
|
|
o2 = (o2 + 5) / 10;
|
|
he = (he + 5) / 10;
|
|
value = o2 + (he << 16);
|
|
|
|
ev = create_event(seconds, he ? SAMPLE_EVENT_GASCHANGE2 : SAMPLE_EVENT_GASCHANGE, 0, value, "gaschange");
|
|
ev->gas.index = idx;
|
|
ev->gas.mix = mix;
|
|
return ev;
|
|
}
|
|
|
|
extern "C" void add_gas_switch_event(struct dive *dive, struct divecomputer *dc, int seconds, int idx)
|
|
{
|
|
/* sanity check so we don't crash */
|
|
/* FIXME: The planner uses a dummy cylinder one past the official number of cylinders
|
|
* in the table to mark no-cylinder surface interavals. This is horrendous. Fix ASAP. */
|
|
//if (idx < 0 || idx >= dive->cylinders.nr) {
|
|
if (idx < 0 || idx >= dive->cylinders.nr + 1 || idx >= dive->cylinders.allocated) {
|
|
report_error("Unknown cylinder index: %d", idx);
|
|
return;
|
|
}
|
|
struct event *ev = create_gas_switch_event(dive, dc, seconds, idx);
|
|
add_event_to_dc(dc, ev);
|
|
}
|
|
|
|
/* since the name is an array as part of the structure (how silly is that?) we
|
|
* have to actually remove the existing event and replace it with a new one.
|
|
* WARNING, WARNING... this may end up freeing event in case that event is indeed
|
|
* WARNING, WARNING... part of this divecomputer on this dive! */
|
|
extern "C" void update_event_name(struct dive *d, int dc_number, struct event *event, const char *name)
|
|
{
|
|
if (!d || !event)
|
|
return;
|
|
struct divecomputer *dc = get_dive_dc(d, dc_number);
|
|
if (!dc)
|
|
return;
|
|
struct event **removep = &dc->events;
|
|
struct event *remove;
|
|
while ((*removep)->next && !same_event(*removep, event))
|
|
removep = &(*removep)->next;
|
|
if (!same_event(*removep, event))
|
|
return;
|
|
remove = *removep;
|
|
*removep = (*removep)->next;
|
|
add_event(dc, event->time.seconds, event->type, event->flags, event->value, name);
|
|
free(remove);
|
|
invalidate_dive_cache(d);
|
|
}
|
|
|
|
extern "C" struct gasmix get_gasmix_from_event(const struct dive *dive, const struct event *ev)
|
|
{
|
|
if (ev && event_is_gaschange(ev)) {
|
|
int index = ev->gas.index;
|
|
// FIXME: The planner uses one past cylinder-count to signify "surface air". Remove in due course.
|
|
if (index == dive->cylinders.nr)
|
|
return gasmix_air;
|
|
if (index >= 0 && index < dive->cylinders.nr)
|
|
return get_cylinder(dive, index)->gasmix;
|
|
return ev->gas.mix;
|
|
}
|
|
return gasmix_air;
|
|
}
|
|
|
|
// we need this to be uniq. oh, and it has no meaning whatsoever
|
|
// - that's why we have the silly initial number and increment by 3 :-)
|
|
extern "C" int dive_getUniqID()
|
|
{
|
|
static int maxId = 83529;
|
|
maxId += 3;
|
|
return maxId;
|
|
}
|
|
|
|
extern "C" struct dive *alloc_dive(void)
|
|
{
|
|
struct dive *dive;
|
|
|
|
dive = (struct dive *)malloc(sizeof(*dive));
|
|
if (!dive)
|
|
exit(1);
|
|
memset(dive, 0, sizeof(*dive));
|
|
dive->id = dive_getUniqID();
|
|
return dive;
|
|
}
|
|
|
|
/* copy an element in a list of dive computer extra data */
|
|
static void copy_extra_data(struct extra_data *sed, struct extra_data *ded)
|
|
{
|
|
ded->key = copy_string(sed->key);
|
|
ded->value = copy_string(sed->value);
|
|
}
|
|
|
|
/* this is very different from the copy_divecomputer later in this file;
|
|
* this function actually makes full copies of the content */
|
|
static void copy_dc(const struct divecomputer *sdc, struct divecomputer *ddc)
|
|
{
|
|
*ddc = *sdc;
|
|
ddc->model = copy_string(sdc->model);
|
|
ddc->serial = copy_string(sdc->serial);
|
|
ddc->fw_version = copy_string(sdc->fw_version);
|
|
copy_samples(sdc, ddc);
|
|
copy_events(sdc, ddc);
|
|
STRUCTURED_LIST_COPY(struct extra_data, sdc->extra_data, ddc->extra_data, copy_extra_data);
|
|
}
|
|
|
|
static void dc_cylinder_renumber(struct dive *dive, struct divecomputer *dc, const int mapping[]);
|
|
|
|
/* copy dive computer list and renumber the cylinders
|
|
* space for the first divecomputer is provided by the
|
|
* caller, the remainder is allocated */
|
|
static void copy_dc_renumber(struct dive *d, const struct divecomputer *sdc, struct divecomputer *ddc, const int cylinders_map[])
|
|
{
|
|
for (;;) {
|
|
copy_dc(sdc, ddc);
|
|
dc_cylinder_renumber(d, ddc, cylinders_map);
|
|
if (!sdc->next)
|
|
break;
|
|
sdc = sdc->next;
|
|
ddc->next = (divecomputer *)calloc(1, sizeof(struct divecomputer));
|
|
ddc = ddc->next;
|
|
}
|
|
ddc->next = NULL;
|
|
}
|
|
|
|
static void free_dive_structures(struct dive *d)
|
|
{
|
|
if (!d)
|
|
return;
|
|
fulltext_unregister(d);
|
|
/* free the strings */
|
|
free(d->buddy);
|
|
free(d->diveguide);
|
|
free(d->notes);
|
|
free(d->suit);
|
|
/* free tags, additional dive computers, and pictures */
|
|
taglist_free(d->tag_list);
|
|
free_dive_dcs(&d->dc);
|
|
clear_cylinder_table(&d->cylinders);
|
|
free(d->cylinders.cylinders);
|
|
clear_weightsystem_table(&d->weightsystems);
|
|
free(d->weightsystems.weightsystems);
|
|
clear_picture_table(&d->pictures);
|
|
free(d->pictures.pictures);
|
|
}
|
|
|
|
extern "C" void free_dive(struct dive *d)
|
|
{
|
|
free_dive_structures(d);
|
|
free(d);
|
|
}
|
|
|
|
/* copy_dive makes duplicates of many components of a dive;
|
|
* in order not to leak memory, we need to free those .
|
|
* copy_dive doesn't play with the divetrip and forward/backward pointers
|
|
* so we can ignore those */
|
|
extern "C" void clear_dive(struct dive *d)
|
|
{
|
|
if (!d)
|
|
return;
|
|
free_dive_structures(d);
|
|
memset(d, 0, sizeof(struct dive));
|
|
}
|
|
|
|
/* make a true copy that is independent of the source dive;
|
|
* all data structures are duplicated, so the copy can be modified without
|
|
* any impact on the source */
|
|
static void copy_dive_nodc(const struct dive *s, struct dive *d)
|
|
{
|
|
clear_dive(d);
|
|
/* simply copy things over, but then make actual copies of the
|
|
* relevant components that are referenced through pointers,
|
|
* so all the strings and the structured lists */
|
|
*d = *s;
|
|
memset(&d->cylinders, 0, sizeof(d->cylinders));
|
|
memset(&d->weightsystems, 0, sizeof(d->weightsystems));
|
|
memset(&d->pictures, 0, sizeof(d->pictures));
|
|
d->full_text = NULL;
|
|
invalidate_dive_cache(d);
|
|
d->buddy = copy_string(s->buddy);
|
|
d->diveguide = copy_string(s->diveguide);
|
|
d->notes = copy_string(s->notes);
|
|
d->suit = copy_string(s->suit);
|
|
copy_cylinders(&s->cylinders, &d->cylinders);
|
|
copy_weights(&s->weightsystems, &d->weightsystems);
|
|
copy_pictures(&s->pictures, &d->pictures);
|
|
d->tag_list = taglist_copy(s->tag_list);
|
|
}
|
|
|
|
extern "C" void copy_dive(const struct dive *s, struct dive *d)
|
|
{
|
|
copy_dive_nodc(s, d);
|
|
|
|
// Copy the first dc explicitly, then the list of subsequent dc's
|
|
copy_dc(&s->dc, &d->dc);
|
|
STRUCTURED_LIST_COPY(struct divecomputer, s->dc.next, d->dc.next, copy_dc);
|
|
}
|
|
|
|
static void copy_dive_onedc(const struct dive *s, const struct divecomputer *sdc, struct dive *d)
|
|
{
|
|
copy_dive_nodc(s, d);
|
|
copy_dc(sdc, &d->dc);
|
|
d->dc.next = NULL;
|
|
}
|
|
|
|
/* make a clone of the source dive and clean out the source dive;
|
|
* this allows us to create a dive on the stack and then
|
|
* add it to the divelist. */
|
|
extern "C" struct dive *move_dive(struct dive *s)
|
|
{
|
|
struct dive *dive = alloc_dive();
|
|
*dive = *s; // so all the pointers in dive point to the things s pointed to
|
|
memset(s, 0, sizeof(struct dive)); // and now the pointers in s are gone
|
|
return dive;
|
|
}
|
|
|
|
#define CONDITIONAL_COPY_STRING(_component) \
|
|
if (what._component) \
|
|
d->_component = copy_string(s->_component)
|
|
|
|
// copy elements, depending on bits in what that are set
|
|
extern "C" void selective_copy_dive(const struct dive *s, struct dive *d, struct dive_components what, bool clear)
|
|
{
|
|
if (clear)
|
|
clear_dive(d);
|
|
CONDITIONAL_COPY_STRING(notes);
|
|
CONDITIONAL_COPY_STRING(diveguide);
|
|
CONDITIONAL_COPY_STRING(buddy);
|
|
CONDITIONAL_COPY_STRING(suit);
|
|
if (what.rating)
|
|
d->rating = s->rating;
|
|
if (what.visibility)
|
|
d->visibility = s->visibility;
|
|
if (what.divesite) {
|
|
unregister_dive_from_dive_site(d);
|
|
add_dive_to_dive_site(d, s->dive_site);
|
|
}
|
|
if (what.tags)
|
|
d->tag_list = taglist_copy(s->tag_list);
|
|
if (what.cylinders)
|
|
copy_cylinder_types(s, d);
|
|
if (what.weights)
|
|
copy_weights(&s->weightsystems, &d->weightsystems);
|
|
if (what.number)
|
|
d->number = s->number;
|
|
if (what.when)
|
|
d->when = s->when;
|
|
}
|
|
#undef CONDITIONAL_COPY_STRING
|
|
|
|
/* copies all events from all dive computers before a given time
|
|
this is used when editing a dive in the planner to preserve the events
|
|
of the old dive */
|
|
extern "C" void copy_events_until(const struct dive *sd, struct dive *dd, int time)
|
|
{
|
|
if (!sd || !dd)
|
|
return;
|
|
|
|
const struct divecomputer *s = &sd->dc;
|
|
struct divecomputer *d = &dd->dc;
|
|
|
|
while (s && d) {
|
|
const struct event *ev;
|
|
ev = s->events;
|
|
while (ev != NULL) {
|
|
// Don't add events the planner knows about
|
|
if (ev->time.seconds < time && !event_is_gaschange(ev) && !event_is_divemodechange(ev))
|
|
add_event(d, ev->time.seconds, ev->type, ev->flags, ev->value, ev->name);
|
|
ev = ev->next;
|
|
}
|
|
s = s->next;
|
|
d = d->next;
|
|
}
|
|
}
|
|
|
|
extern "C" int nr_cylinders(const struct dive *dive)
|
|
{
|
|
return dive->cylinders.nr;
|
|
}
|
|
|
|
extern "C" int nr_weightsystems(const struct dive *dive)
|
|
{
|
|
return dive->weightsystems.nr;
|
|
}
|
|
|
|
extern "C" void copy_used_cylinders(const struct dive *s, struct dive *d, bool used_only)
|
|
{
|
|
int i;
|
|
if (!s || !d)
|
|
return;
|
|
|
|
clear_cylinder_table(&d->cylinders);
|
|
for (i = 0; i < s->cylinders.nr; i++) {
|
|
if (!used_only || is_cylinder_used(s, i) || get_cylinder(s, i)->cylinder_use == NOT_USED)
|
|
add_cloned_cylinder(&d->cylinders, *get_cylinder(s, i));
|
|
}
|
|
}
|
|
|
|
/*
|
|
* So when we re-calculate maxdepth and meandepth, we will
|
|
* not override the old numbers if they are close to the
|
|
* new ones.
|
|
*
|
|
* Why? Because a dive computer may well actually track the
|
|
* max. depth and mean depth at finer granularity than the
|
|
* samples it stores. So it's possible that the max and mean
|
|
* have been reported more correctly originally.
|
|
*
|
|
* Only if the values calculated from the samples are clearly
|
|
* different do we override the normal depth values.
|
|
*
|
|
* This considers 1m to be "clearly different". That's
|
|
* a totally random number.
|
|
*/
|
|
static void update_depth(depth_t *depth, int new_depth)
|
|
{
|
|
if (new_depth) {
|
|
int old = depth->mm;
|
|
|
|
if (abs(old - new_depth) > 1000)
|
|
depth->mm = new_depth;
|
|
}
|
|
}
|
|
|
|
static void update_temperature(temperature_t *temperature, int new_temp)
|
|
{
|
|
if (new_temp) {
|
|
int old = temperature->mkelvin;
|
|
|
|
if (abs(old - new_temp) > 1000)
|
|
temperature->mkelvin = new_temp;
|
|
}
|
|
}
|
|
|
|
/* Which cylinders had gas used? */
|
|
#define SOME_GAS 5000
|
|
static bool cylinder_used(const cylinder_t *cyl)
|
|
{
|
|
int start_mbar, end_mbar;
|
|
|
|
start_mbar = cyl->start.mbar ?: cyl->sample_start.mbar;
|
|
end_mbar = cyl->end.mbar ?: cyl->sample_end.mbar;
|
|
|
|
// More than 5 bar used? This matches statistics.c
|
|
// heuristics
|
|
return start_mbar > end_mbar + SOME_GAS;
|
|
}
|
|
|
|
/* Get list of used cylinders. Returns the number of used cylinders. */
|
|
static int get_cylinder_used(const struct dive *dive, bool used[])
|
|
{
|
|
int i, num = 0;
|
|
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
used[i] = cylinder_used(get_cylinder(dive, i));
|
|
if (used[i])
|
|
num++;
|
|
}
|
|
return num;
|
|
}
|
|
|
|
/* Are there any used cylinders which we do not know usage about? */
|
|
static bool has_unknown_used_cylinders(const struct dive *dive, const struct divecomputer *dc,
|
|
const bool used_cylinders[], int num)
|
|
{
|
|
int idx;
|
|
const struct event *ev;
|
|
auto used_and_unknown = std::make_unique<bool[]>(dive->cylinders.nr);
|
|
std::copy(used_cylinders, used_cylinders + dive->cylinders.nr, used_and_unknown.get());
|
|
|
|
/* We know about using the O2 cylinder in a CCR dive */
|
|
if (dc->divemode == CCR) {
|
|
int o2_cyl = get_cylinder_idx_by_use(dive, OXYGEN);
|
|
if (o2_cyl >= 0 && used_and_unknown[o2_cyl]) {
|
|
used_and_unknown[o2_cyl] = false;
|
|
num--;
|
|
}
|
|
}
|
|
|
|
/* We know about the explicit first cylinder (or first) */
|
|
idx = explicit_first_cylinder(dive, dc);
|
|
if (idx >= 0 && used_and_unknown[idx]) {
|
|
used_and_unknown[idx] = false;
|
|
num--;
|
|
}
|
|
|
|
/* And we have possible switches to other gases */
|
|
ev = get_next_event(dc->events, "gaschange");
|
|
while (ev && num > 0) {
|
|
idx = get_cylinder_index(dive, ev);
|
|
if (idx >= 0 && used_and_unknown[idx]) {
|
|
used_and_unknown[idx] = false;
|
|
num--;
|
|
}
|
|
ev = get_next_event(ev->next, "gaschange");
|
|
}
|
|
|
|
return num > 0;
|
|
}
|
|
|
|
extern "C" void per_cylinder_mean_depth(const struct dive *dive, struct divecomputer *dc, int *mean, int *duration)
|
|
{
|
|
int i;
|
|
int32_t lasttime = 0;
|
|
int lastdepth = 0;
|
|
int idx = 0;
|
|
int num_used_cylinders;
|
|
|
|
if (dive->cylinders.nr <= 0)
|
|
return;
|
|
|
|
for (i = 0; i < dive->cylinders.nr; i++)
|
|
mean[i] = duration[i] = 0;
|
|
if (!dc)
|
|
return;
|
|
|
|
/*
|
|
* There is no point in doing per-cylinder information
|
|
* if we don't actually know about the usage of all the
|
|
* used cylinders.
|
|
*/
|
|
auto used_cylinders = std::make_unique<bool[]>(dive->cylinders.nr);
|
|
num_used_cylinders = get_cylinder_used(dive, used_cylinders.get());
|
|
if (has_unknown_used_cylinders(dive, dc, used_cylinders.get(), num_used_cylinders)) {
|
|
/*
|
|
* If we had more than one used cylinder, but
|
|
* do not know usage of them, we simply cannot
|
|
* account mean depth to them.
|
|
*/
|
|
if (num_used_cylinders > 1)
|
|
return;
|
|
|
|
/*
|
|
* For a single cylinder, use the overall mean
|
|
* and duration
|
|
*/
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
if (used_cylinders[i]) {
|
|
mean[i] = dc->meandepth.mm;
|
|
duration[i] = dc->duration.seconds;
|
|
}
|
|
}
|
|
|
|
return;
|
|
}
|
|
if (!dc->samples)
|
|
fake_dc(dc);
|
|
const struct event *ev = get_next_event(dc->events, "gaschange");
|
|
std::vector<int> depthtime(dive->cylinders.nr, 0);
|
|
for (i = 0; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample + i;
|
|
int32_t time = sample->time.seconds;
|
|
int depth = sample->depth.mm;
|
|
|
|
/* Make sure to move the event past 'lasttime' */
|
|
while (ev && lasttime >= ev->time.seconds) {
|
|
idx = get_cylinder_index(dive, ev);
|
|
ev = get_next_event(ev->next, "gaschange");
|
|
}
|
|
|
|
/* Do we need to fake a midway sample at an event? */
|
|
if (ev && time > ev->time.seconds) {
|
|
int newtime = ev->time.seconds;
|
|
int newdepth = interpolate(lastdepth, depth, newtime - lasttime, time - lasttime);
|
|
|
|
time = newtime;
|
|
depth = newdepth;
|
|
i--;
|
|
}
|
|
/* We ignore segments at the surface */
|
|
if (depth > SURFACE_THRESHOLD || lastdepth > SURFACE_THRESHOLD) {
|
|
duration[idx] += time - lasttime;
|
|
depthtime[idx] += (time - lasttime) * (depth + lastdepth) / 2;
|
|
}
|
|
lastdepth = depth;
|
|
lasttime = time;
|
|
}
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
if (duration[i])
|
|
mean[i] = (depthtime[i] + duration[i] / 2) / duration[i];
|
|
}
|
|
}
|
|
|
|
static void update_min_max_temperatures(struct dive *dive, temperature_t temperature)
|
|
{
|
|
if (temperature.mkelvin) {
|
|
if (!dive->maxtemp.mkelvin || temperature.mkelvin > dive->maxtemp.mkelvin)
|
|
dive->maxtemp = temperature;
|
|
if (!dive->mintemp.mkelvin || temperature.mkelvin < dive->mintemp.mkelvin)
|
|
dive->mintemp = temperature;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* If the cylinder tank pressures are within half a bar
|
|
* (about 8 PSI) of the sample pressures, we consider it
|
|
* to be a rounding error, and throw them away as redundant.
|
|
*/
|
|
static int same_rounded_pressure(pressure_t a, pressure_t b)
|
|
{
|
|
return abs(a.mbar - b.mbar) <= 500;
|
|
}
|
|
|
|
/* Some dive computers (Cobalt) don't start the dive with cylinder 0 but explicitly
|
|
* tell us what the first gas is with a gas change event in the first sample.
|
|
* Sneakily we'll use a return value of 0 (or FALSE) when there is no explicit
|
|
* first cylinder - in which case cylinder 0 is indeed the first cylinder.
|
|
* We likewise return 0 if the event concerns a cylinder that doesn't exist.
|
|
* If the dive has no cylinders, -1 is returned. */
|
|
extern "C" int explicit_first_cylinder(const struct dive *dive, const struct divecomputer *dc)
|
|
{
|
|
int res = 0;
|
|
if (!dive->cylinders.nr)
|
|
return -1;
|
|
if (dc) {
|
|
const struct event *ev = get_next_event(dc->events, "gaschange");
|
|
if (ev && ((dc->sample && ev->time.seconds == dc->sample[0].time.seconds) || ev->time.seconds <= 1))
|
|
res = get_cylinder_index(dive, ev);
|
|
else if (dc->divemode == CCR)
|
|
res = std::max(get_cylinder_idx_by_use(dive, DILUENT), res);
|
|
}
|
|
return res < dive->cylinders.nr ? res : 0;
|
|
}
|
|
|
|
/* this gets called when the dive mode has changed (so OC vs. CC)
|
|
* there are two places we might have setpoints... events or in the samples
|
|
*/
|
|
extern "C" void update_setpoint_events(const struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
struct event *ev;
|
|
int new_setpoint = 0;
|
|
|
|
if (dc->divemode == CCR)
|
|
new_setpoint = prefs.defaultsetpoint;
|
|
|
|
if (dc->divemode == OC &&
|
|
(same_string(dc->model, "Shearwater Predator") ||
|
|
same_string(dc->model, "Shearwater Petrel") ||
|
|
same_string(dc->model, "Shearwater Nerd"))) {
|
|
// make sure there's no setpoint in the samples
|
|
// this is an irreversible change - so switching a dive to OC
|
|
// by mistake when it's actually CCR is _bad_
|
|
// So we make sure, this comes from a Predator or Petrel and we only remove
|
|
// pO2 values we would have computed anyway.
|
|
const struct event *ev = get_next_event(dc->events, "gaschange");
|
|
struct gasmix gasmix = get_gasmix_from_event(dive, ev);
|
|
const struct event *next = get_next_event(ev, "gaschange");
|
|
|
|
for (int i = 0; i < dc->samples; i++) {
|
|
struct gas_pressures pressures;
|
|
if (next && dc->sample[i].time.seconds >= next->time.seconds) {
|
|
ev = next;
|
|
gasmix = get_gasmix_from_event(dive, ev);
|
|
next = get_next_event(ev, "gaschange");
|
|
}
|
|
fill_pressures(&pressures, lrint(calculate_depth_to_mbarf(dc->sample[i].depth.mm, dc->surface_pressure, 0)), gasmix ,0, dc->divemode);
|
|
if (abs(dc->sample[i].setpoint.mbar - (int)(1000 * pressures.o2)) <= 50)
|
|
dc->sample[i].setpoint.mbar = 0;
|
|
}
|
|
}
|
|
|
|
// an "SP change" event at t=0 is currently our marker for OC vs CCR
|
|
// this will need to change to a saner setup, but for now we can just
|
|
// check if such an event is there and adjust it, or add that event
|
|
ev = get_next_event_mutable(dc->events, "SP change");
|
|
if (ev && ev->time.seconds == 0) {
|
|
ev->value = new_setpoint;
|
|
} else {
|
|
if (!add_event(dc, 0, SAMPLE_EVENT_PO2, 0, new_setpoint, "SP change"))
|
|
report_info("Could not add setpoint change event");
|
|
}
|
|
}
|
|
|
|
/*
|
|
* See if the size/workingpressure looks like some standard cylinder
|
|
* size, eg "AL80".
|
|
*
|
|
* NOTE! We don't take compressibility into account when naming
|
|
* cylinders. That makes a certain amount of sense, since the
|
|
* cylinder name is independent from the gasmix, and different
|
|
* gasmixes have different compressibility.
|
|
*/
|
|
static void match_standard_cylinder(cylinder_type_t *type)
|
|
{
|
|
double cuft, bar;
|
|
int psi, len;
|
|
const char *fmt;
|
|
char buffer[40], *p;
|
|
|
|
/* Do we already have a cylinder description? */
|
|
if (type->description)
|
|
return;
|
|
|
|
bar = type->workingpressure.mbar / 1000.0;
|
|
cuft = ml_to_cuft(type->size.mliter);
|
|
cuft *= bar_to_atm(bar);
|
|
psi = lrint(to_PSI(type->workingpressure));
|
|
|
|
switch (psi) {
|
|
case 2300 ... 2500: /* 2400 psi: LP tank */
|
|
fmt = "LP%d";
|
|
break;
|
|
case 2600 ... 2700: /* 2640 psi: LP+10% */
|
|
fmt = "LP%d";
|
|
break;
|
|
case 2900 ... 3100: /* 3000 psi: ALx tank */
|
|
fmt = "AL%d";
|
|
break;
|
|
case 3400 ... 3500: /* 3442 psi: HP tank */
|
|
fmt = "HP%d";
|
|
break;
|
|
case 3700 ... 3850: /* HP+10% */
|
|
fmt = "HP%d+";
|
|
break;
|
|
default:
|
|
return;
|
|
}
|
|
len = snprintf(buffer, sizeof(buffer), fmt, (int)lrint(cuft));
|
|
p = (char *)malloc(len + 1);
|
|
if (!p)
|
|
return;
|
|
memcpy(p, buffer, len + 1);
|
|
type->description = p;
|
|
}
|
|
|
|
/*
|
|
* There are two ways to give cylinder size information:
|
|
* - total amount of gas in cuft (depends on working pressure and physical size)
|
|
* - physical size
|
|
*
|
|
* where "physical size" is the one that actually matters and is sane.
|
|
*
|
|
* We internally use physical size only. But we save the workingpressure
|
|
* so that we can do the conversion if required.
|
|
*/
|
|
static void sanitize_cylinder_type(cylinder_type_t *type)
|
|
{
|
|
/* If we have no working pressure, it had *better* be just a physical size! */
|
|
if (!type->workingpressure.mbar)
|
|
return;
|
|
|
|
/* No size either? Nothing to go on */
|
|
if (!type->size.mliter)
|
|
return;
|
|
|
|
/* Ok, we have both size and pressure: try to match a description */
|
|
match_standard_cylinder(type);
|
|
}
|
|
|
|
static void sanitize_cylinder_info(struct dive *dive)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
sanitize_gasmix(&get_cylinder(dive, i)->gasmix);
|
|
sanitize_cylinder_type(&get_cylinder(dive, i)->type);
|
|
}
|
|
}
|
|
|
|
/* some events should never be thrown away */
|
|
static bool is_potentially_redundant(const struct event *event)
|
|
{
|
|
if (!strcmp(event->name, "gaschange"))
|
|
return false;
|
|
if (!strcmp(event->name, "bookmark"))
|
|
return false;
|
|
if (!strcmp(event->name, "heading"))
|
|
return false;
|
|
return true;
|
|
}
|
|
|
|
/* match just by name - we compare the details in the code that uses this helper */
|
|
static struct event *find_previous_event(struct divecomputer *dc, struct event *event)
|
|
{
|
|
struct event *ev = dc->events;
|
|
struct event *previous = NULL;
|
|
|
|
if (empty_string(event->name))
|
|
return NULL;
|
|
while (ev && ev != event) {
|
|
if (same_string(ev->name, event->name))
|
|
previous = ev;
|
|
ev = ev->next;
|
|
}
|
|
return previous;
|
|
}
|
|
|
|
extern "C" pressure_t calculate_surface_pressure(const struct dive *dive)
|
|
{
|
|
const struct divecomputer *dc;
|
|
pressure_t res;
|
|
int sum = 0, nr = 0;
|
|
|
|
for_each_relevant_dc(dive, dc) {
|
|
if (dc->surface_pressure.mbar) {
|
|
sum += dc->surface_pressure.mbar;
|
|
nr++;
|
|
}
|
|
}
|
|
res.mbar = nr ? (sum + nr / 2) / nr : 0;
|
|
return res;
|
|
}
|
|
|
|
static void fixup_surface_pressure(struct dive *dive)
|
|
{
|
|
dive->surface_pressure = calculate_surface_pressure(dive);
|
|
}
|
|
|
|
/* if the surface pressure in the dive data is redundant to the calculated
|
|
* value (i.e., it was added by running fixup on the dive) return 0,
|
|
* otherwise return the surface pressure given in the dive */
|
|
extern "C" pressure_t un_fixup_surface_pressure(const struct dive *d)
|
|
{
|
|
pressure_t res = d->surface_pressure;
|
|
if (res.mbar && res.mbar == calculate_surface_pressure(d).mbar)
|
|
res.mbar = 0;
|
|
return res;
|
|
}
|
|
|
|
static void fixup_water_salinity(struct dive *dive)
|
|
{
|
|
struct divecomputer *dc;
|
|
int sum = 0, nr = 0;
|
|
|
|
for_each_relevant_dc (dive, dc) {
|
|
if (dc->salinity) {
|
|
if (dc->salinity < 500)
|
|
dc->salinity += FRESHWATER_SALINITY;
|
|
sum += dc->salinity;
|
|
nr++;
|
|
}
|
|
}
|
|
if (nr)
|
|
dive->salinity = (sum + nr / 2) / nr;
|
|
}
|
|
|
|
extern "C" int get_dive_salinity(const struct dive *dive)
|
|
{
|
|
return dive->user_salinity ? dive->user_salinity : dive->salinity;
|
|
}
|
|
|
|
static void fixup_meandepth(struct dive *dive)
|
|
{
|
|
struct divecomputer *dc;
|
|
int sum = 0, nr = 0;
|
|
|
|
for_each_relevant_dc (dive, dc) {
|
|
if (dc->meandepth.mm) {
|
|
sum += dc->meandepth.mm;
|
|
nr++;
|
|
}
|
|
}
|
|
|
|
if (nr)
|
|
dive->meandepth.mm = (sum + nr / 2) / nr;
|
|
}
|
|
|
|
static void fixup_duration(struct dive *dive)
|
|
{
|
|
struct divecomputer *dc;
|
|
duration_t duration = { };
|
|
|
|
for_each_relevant_dc (dive, dc) {
|
|
duration.seconds = std::max(duration.seconds, dc->duration.seconds);
|
|
}
|
|
dive->duration.seconds = duration.seconds;
|
|
}
|
|
|
|
static void fixup_watertemp(struct dive *dive)
|
|
{
|
|
if (!dive->watertemp.mkelvin)
|
|
dive->watertemp.mkelvin = dc_watertemp(&dive->dc);
|
|
}
|
|
|
|
static void fixup_airtemp(struct dive *dive)
|
|
{
|
|
if (!dive->airtemp.mkelvin)
|
|
dive->airtemp.mkelvin = dc_airtemp(&dive->dc);
|
|
}
|
|
|
|
/* if the air temperature in the dive data is redundant to the one in its
|
|
* first divecomputer (i.e., it was added by running fixup on the dive)
|
|
* return 0, otherwise return the air temperature given in the dive */
|
|
static temperature_t un_fixup_airtemp(const struct dive *a)
|
|
{
|
|
temperature_t res = a->airtemp;
|
|
if (a->airtemp.mkelvin && a->airtemp.mkelvin == dc_airtemp(&a->dc))
|
|
res.mkelvin = 0;
|
|
return res;
|
|
}
|
|
|
|
/*
|
|
* events are stored as a linked list, so the concept of
|
|
* "consecutive, identical events" is somewhat hard to
|
|
* implement correctly (especially given that on some dive
|
|
* computers events are asynchronous, so they can come in
|
|
* between what would be the non-constant sample rate).
|
|
*
|
|
* So what we do is that we throw away clearly redundant
|
|
* events that are fewer than 61 seconds apart (assuming there
|
|
* is no dive computer with a sample rate of more than 60
|
|
* seconds... that would be pretty pointless to plot the
|
|
* profile with)
|
|
*
|
|
* We first only mark the events for deletion so that we
|
|
* still know when the previous event happened.
|
|
*/
|
|
static void fixup_dc_events(struct divecomputer *dc)
|
|
{
|
|
struct event *event;
|
|
|
|
event = dc->events;
|
|
while (event) {
|
|
struct event *prev;
|
|
if (is_potentially_redundant(event)) {
|
|
prev = find_previous_event(dc, event);
|
|
if (prev && prev->value == event->value &&
|
|
prev->flags == event->flags &&
|
|
event->time.seconds - prev->time.seconds < 61)
|
|
event->deleted = true;
|
|
}
|
|
event = event->next;
|
|
}
|
|
event = dc->events;
|
|
while (event) {
|
|
if (event->next && event->next->deleted) {
|
|
struct event *nextnext = event->next->next;
|
|
free(event->next);
|
|
event->next = nextnext;
|
|
} else {
|
|
event = event->next;
|
|
}
|
|
}
|
|
}
|
|
|
|
static int interpolate_depth(struct divecomputer *dc, int idx, int lastdepth, int lasttime, int now)
|
|
{
|
|
int i;
|
|
int nextdepth = lastdepth;
|
|
int nexttime = now;
|
|
|
|
for (i = idx+1; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample + i;
|
|
if (sample->depth.mm < 0)
|
|
continue;
|
|
nextdepth = sample->depth.mm;
|
|
nexttime = sample->time.seconds;
|
|
break;
|
|
}
|
|
return interpolate(lastdepth, nextdepth, now-lasttime, nexttime-lasttime);
|
|
}
|
|
|
|
static void fixup_dc_depths(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
int i;
|
|
int maxdepth = dc->maxdepth.mm;
|
|
int lasttime = 0, lastdepth = 0;
|
|
|
|
for (i = 0; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample + i;
|
|
int time = sample->time.seconds;
|
|
int depth = sample->depth.mm;
|
|
|
|
if (depth < 0) {
|
|
depth = interpolate_depth(dc, i, lastdepth, lasttime, time);
|
|
sample->depth.mm = depth;
|
|
}
|
|
|
|
if (depth > SURFACE_THRESHOLD) {
|
|
if (depth > maxdepth)
|
|
maxdepth = depth;
|
|
}
|
|
|
|
lastdepth = depth;
|
|
lasttime = time;
|
|
if (sample->cns > dive->maxcns)
|
|
dive->maxcns = sample->cns;
|
|
}
|
|
|
|
update_depth(&dc->maxdepth, maxdepth);
|
|
if (!is_logged(dive) || !is_dc_planner(dc))
|
|
if (maxdepth > dive->maxdepth.mm)
|
|
dive->maxdepth.mm = maxdepth;
|
|
}
|
|
|
|
static void fixup_dc_ndl(struct divecomputer *dc)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample + i;
|
|
if (sample->ndl.seconds != 0)
|
|
break;
|
|
if (sample->ndl.seconds == 0)
|
|
sample->ndl.seconds = -1;
|
|
}
|
|
}
|
|
|
|
static void fixup_dc_temp(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
int i;
|
|
int mintemp = 0, lasttemp = 0;
|
|
|
|
for (i = 0; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample + i;
|
|
int temp = sample->temperature.mkelvin;
|
|
|
|
if (temp) {
|
|
/*
|
|
* If we have consecutive identical
|
|
* temperature readings, throw away
|
|
* the redundant ones.
|
|
*/
|
|
if (lasttemp == temp)
|
|
sample->temperature.mkelvin = 0;
|
|
else
|
|
lasttemp = temp;
|
|
|
|
if (!mintemp || temp < mintemp)
|
|
mintemp = temp;
|
|
}
|
|
|
|
update_min_max_temperatures(dive, sample->temperature);
|
|
}
|
|
update_temperature(&dc->watertemp, mintemp);
|
|
update_min_max_temperatures(dive, dc->watertemp);
|
|
}
|
|
|
|
/* Remove redundant pressure information */
|
|
static void simplify_dc_pressures(struct divecomputer *dc)
|
|
{
|
|
int i;
|
|
int lastindex[2] = { -1, -1 };
|
|
int lastpressure[2] = { 0 };
|
|
|
|
for (i = 0; i < dc->samples; i++) {
|
|
int j;
|
|
struct sample *sample = dc->sample + i;
|
|
|
|
for (j = 0; j < MAX_SENSORS; j++) {
|
|
int pressure = sample->pressure[j].mbar;
|
|
int index = sample->sensor[j];
|
|
|
|
if (index == lastindex[j]) {
|
|
/* Remove duplicate redundant pressure information */
|
|
if (pressure == lastpressure[j])
|
|
sample->pressure[j].mbar = 0;
|
|
}
|
|
lastindex[j] = index;
|
|
lastpressure[j] = pressure;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* Do we need a sensor -> cylinder mapping? */
|
|
static void fixup_start_pressure(struct dive *dive, int idx, pressure_t p)
|
|
{
|
|
if (idx >= 0 && idx < dive->cylinders.nr) {
|
|
cylinder_t *cyl = get_cylinder(dive, idx);
|
|
if (p.mbar && !cyl->sample_start.mbar)
|
|
cyl->sample_start = p;
|
|
}
|
|
}
|
|
|
|
static void fixup_end_pressure(struct dive *dive, int idx, pressure_t p)
|
|
{
|
|
if (idx >= 0 && idx < dive->cylinders.nr) {
|
|
cylinder_t *cyl = get_cylinder(dive, idx);
|
|
if (p.mbar && !cyl->sample_end.mbar)
|
|
cyl->sample_end = p;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Check the cylinder pressure sample information and fill in the
|
|
* overall cylinder pressures from those.
|
|
*
|
|
* We ignore surface samples for tank pressure information.
|
|
*
|
|
* At the beginning of the dive, let the cylinder cool down
|
|
* if the diver starts off at the surface. And at the end
|
|
* of the dive, there may be surface pressures where the
|
|
* diver has already turned off the air supply (especially
|
|
* for computers like the Uemis Zurich that end up saving
|
|
* quite a bit of samples after the dive has ended).
|
|
*/
|
|
static void fixup_dive_pressures(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
int i;
|
|
|
|
/* Walk the samples from the beginning to find starting pressures.. */
|
|
for (i = 0; i < dc->samples; i++) {
|
|
int idx;
|
|
struct sample *sample = dc->sample + i;
|
|
|
|
if (sample->depth.mm < SURFACE_THRESHOLD)
|
|
continue;
|
|
|
|
for (idx = 0; idx < MAX_SENSORS; idx++)
|
|
fixup_start_pressure(dive, sample->sensor[idx], sample->pressure[idx]);
|
|
}
|
|
|
|
/* ..and from the end for ending pressures */
|
|
for (i = dc->samples; --i >= 0; ) {
|
|
int idx;
|
|
struct sample *sample = dc->sample + i;
|
|
|
|
if (sample->depth.mm < SURFACE_THRESHOLD)
|
|
continue;
|
|
|
|
for (idx = 0; idx < MAX_SENSORS; idx++)
|
|
fixup_end_pressure(dive, sample->sensor[idx], sample->pressure[idx]);
|
|
}
|
|
|
|
simplify_dc_pressures(dc);
|
|
}
|
|
|
|
/*
|
|
* Match a gas change event against the cylinders we have
|
|
*/
|
|
static bool validate_gaschange(struct dive *dive, struct event *event)
|
|
{
|
|
int index;
|
|
int o2, he, value;
|
|
|
|
/* We'll get rid of the per-event gasmix, but for now sanitize it */
|
|
if (gasmix_is_air(event->gas.mix))
|
|
event->gas.mix.o2.permille = 0;
|
|
|
|
/* Do we already have a cylinder index for this gasmix? */
|
|
if (event->gas.index >= 0)
|
|
return true;
|
|
|
|
index = find_best_gasmix_match(event->gas.mix, &dive->cylinders);
|
|
if (index < 0 || index >= dive->cylinders.nr)
|
|
return false;
|
|
|
|
/* Fix up the event to have the right information */
|
|
event->gas.index = index;
|
|
event->gas.mix = get_cylinder(dive, index)->gasmix;
|
|
|
|
/* Convert to odd libdivecomputer format */
|
|
o2 = get_o2(event->gas.mix);
|
|
he = get_he(event->gas.mix);
|
|
|
|
o2 = (o2 + 5) / 10;
|
|
he = (he + 5) / 10;
|
|
value = o2 + (he << 16);
|
|
|
|
event->value = value;
|
|
if (he)
|
|
event->type = SAMPLE_EVENT_GASCHANGE2;
|
|
|
|
return true;
|
|
}
|
|
|
|
/* Clean up event, return true if event is ok, false if it should be dropped as bogus */
|
|
static bool validate_event(struct dive *dive, struct event *event)
|
|
{
|
|
if (event_is_gaschange(event))
|
|
return validate_gaschange(dive, event);
|
|
return true;
|
|
}
|
|
|
|
static void fixup_dc_gasswitch(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
struct event **evp, *event;
|
|
|
|
evp = &dc->events;
|
|
while ((event = *evp) != NULL) {
|
|
if (validate_event(dive, event)) {
|
|
evp = &event->next;
|
|
continue;
|
|
}
|
|
|
|
/* Delete this event and try the next one */
|
|
*evp = event->next;
|
|
}
|
|
}
|
|
|
|
static void fixup_no_o2sensors(struct divecomputer *dc)
|
|
{
|
|
// Its only relevant to look for sensor values on CCR and PSCR dives without any no_o2sensors recorded.
|
|
if (dc->no_o2sensors != 0 || !(dc->divemode == CCR || dc->divemode == PSCR))
|
|
return;
|
|
|
|
for (int i = 0; i < dc->samples; i++) {
|
|
int nsensor = 0, j;
|
|
struct sample *s = dc->sample + i;
|
|
|
|
// How many o2 sensors can we find in this sample?
|
|
for (j = 0; j < MAX_O2_SENSORS; j++)
|
|
if (s->o2sensor[j].mbar)
|
|
nsensor++;
|
|
|
|
// If we fond more than the previous found max, record it.
|
|
if (nsensor > dc->no_o2sensors)
|
|
dc->no_o2sensors = nsensor;
|
|
|
|
// Already found the maximum posible amount.
|
|
if (nsensor == MAX_O2_SENSORS)
|
|
return;
|
|
}
|
|
}
|
|
|
|
static void fixup_dc_sample_sensors(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
unsigned long sensor_mask = 0;
|
|
|
|
for (int i = 0; i < dc->samples; i++) {
|
|
struct sample *s = dc->sample + i;
|
|
for (int j = 0; j < MAX_SENSORS; j++) {
|
|
int sensor = s->sensor[j];
|
|
|
|
// No invalid sensor ID's, please
|
|
if (sensor < 0 || sensor > MAX_SENSORS) {
|
|
s->sensor[j] = NO_SENSOR;
|
|
s->pressure[j].mbar = 0;
|
|
continue;
|
|
}
|
|
|
|
// Don't bother tracking sensors with no data
|
|
if (!s->pressure[j].mbar) {
|
|
s->sensor[j] = NO_SENSOR;
|
|
continue;
|
|
}
|
|
|
|
// Remember the set of sensors we had
|
|
sensor_mask |= 1ul << sensor;
|
|
}
|
|
}
|
|
|
|
// Ignore the sensors we have cylinders for
|
|
sensor_mask >>= dive->cylinders.nr;
|
|
|
|
// Do we need to add empty cylinders?
|
|
while (sensor_mask) {
|
|
add_empty_cylinder(&dive->cylinders);
|
|
sensor_mask >>= 1;
|
|
}
|
|
}
|
|
|
|
static void fixup_dive_dc(struct dive *dive, struct divecomputer *dc)
|
|
{
|
|
/* Fixup duration and mean depth */
|
|
fixup_dc_duration(dc);
|
|
|
|
/* Fix up sample depth data */
|
|
fixup_dc_depths(dive, dc);
|
|
|
|
/* Fix up first sample ndl data */
|
|
fixup_dc_ndl(dc);
|
|
|
|
/* Fix up dive temperatures based on dive computer samples */
|
|
fixup_dc_temp(dive, dc);
|
|
|
|
/* Fix up gas switch events */
|
|
fixup_dc_gasswitch(dive, dc);
|
|
|
|
/* Fix up cylinder ids in pressure sensors */
|
|
fixup_dc_sample_sensors(dive, dc);
|
|
|
|
/* Fix up cylinder pressures based on DC info */
|
|
fixup_dive_pressures(dive, dc);
|
|
|
|
fixup_dc_events(dc);
|
|
|
|
/* Fixup CCR / PSCR dives with o2sensor values, but without no_o2sensors */
|
|
fixup_no_o2sensors(dc);
|
|
|
|
/* If there are no samples, generate a fake profile based on depth and time */
|
|
if (!dc->samples)
|
|
fake_dc(dc);
|
|
}
|
|
|
|
extern "C" struct dive *fixup_dive(struct dive *dive)
|
|
{
|
|
int i;
|
|
struct divecomputer *dc;
|
|
|
|
sanitize_cylinder_info(dive);
|
|
dive->maxcns = dive->cns;
|
|
|
|
/*
|
|
* Use the dive's temperatures for minimum and maximum in case
|
|
* we do not have temperatures recorded by DC.
|
|
*/
|
|
|
|
update_min_max_temperatures(dive, dive->watertemp);
|
|
|
|
for_each_dc (dive, dc)
|
|
fixup_dive_dc(dive, dc);
|
|
|
|
fixup_water_salinity(dive);
|
|
if (!dive->surface_pressure.mbar)
|
|
fixup_surface_pressure(dive);
|
|
fixup_meandepth(dive);
|
|
fixup_duration(dive);
|
|
fixup_watertemp(dive);
|
|
fixup_airtemp(dive);
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
cylinder_t *cyl = get_cylinder(dive, i);
|
|
add_cylinder_description(&cyl->type);
|
|
if (same_rounded_pressure(cyl->sample_start, cyl->start))
|
|
cyl->start.mbar = 0;
|
|
if (same_rounded_pressure(cyl->sample_end, cyl->end))
|
|
cyl->end.mbar = 0;
|
|
}
|
|
update_cylinder_related_info(dive);
|
|
for (i = 0; i < dive->weightsystems.nr; i++) {
|
|
weightsystem_t *ws = &dive->weightsystems.weightsystems[i];
|
|
add_weightsystem_description(ws);
|
|
}
|
|
/* we should always have a uniq ID as that gets assigned during alloc_dive(),
|
|
* but we want to make sure... */
|
|
if (!dive->id)
|
|
dive->id = dive_getUniqID();
|
|
|
|
return dive;
|
|
}
|
|
|
|
/* Don't pick a zero for MERGE_MIN() */
|
|
#define MERGE_MAX(res, a, b, n) res->n = std::max(a->n, b->n)
|
|
#define MERGE_MIN(res, a, b, n) res->n = (a->n) ? (b->n) ? std::min(a->n, b->n) : (a->n) : (b->n)
|
|
#define MERGE_TXT(res, a, b, n, sep) res->n = merge_text(a->n, b->n, sep)
|
|
#define MERGE_NONZERO(res, a, b, n) res->n = a->n ? a->n : b->n
|
|
|
|
/*
|
|
* This is like add_sample(), but if the distance from the last sample
|
|
* is excessive, we add two surface samples in between.
|
|
*
|
|
* This is so that if you merge two non-overlapping dives, we make sure
|
|
* that the time in between the dives is at the surface, not some "last
|
|
* sample that happened to be at a depth of 1.2m".
|
|
*/
|
|
static void merge_one_sample(const struct sample *sample, int time, struct divecomputer *dc)
|
|
{
|
|
int last = dc->samples - 1;
|
|
if (last >= 0) {
|
|
struct sample *prev = dc->sample + last;
|
|
int last_time = prev->time.seconds;
|
|
int last_depth = prev->depth.mm;
|
|
|
|
/*
|
|
* Only do surface events if the samples are more than
|
|
* a minute apart, and shallower than 5m
|
|
*/
|
|
if (time > last_time + 60 && last_depth < 5000) {
|
|
struct sample surface;
|
|
|
|
/* Init a few values from prev sample to avoid useless info in XML */
|
|
surface.bearing.degrees = prev->bearing.degrees;
|
|
surface.ndl.seconds = prev->ndl.seconds;
|
|
|
|
add_sample(&surface, last_time + 20, dc);
|
|
add_sample(&surface, time - 20, dc);
|
|
}
|
|
}
|
|
add_sample(sample, time, dc);
|
|
}
|
|
|
|
static void renumber_last_sample(struct divecomputer *dc, const int mapping[]);
|
|
static void sample_renumber(struct sample *s, int i, const int mapping[]);
|
|
|
|
/*
|
|
* Merge samples. Dive 'a' is "offset" seconds before Dive 'b'
|
|
*/
|
|
static void merge_samples(struct divecomputer *res,
|
|
const struct divecomputer *a, const struct divecomputer *b,
|
|
const int *cylinders_map_a, const int *cylinders_map_b,
|
|
int offset)
|
|
{
|
|
int asamples = a->samples;
|
|
int bsamples = b->samples;
|
|
struct sample *as = a->sample;
|
|
struct sample *bs = b->sample;
|
|
|
|
/*
|
|
* We want a positive sample offset, so that sample
|
|
* times are always positive. So if the samples for
|
|
* 'b' are before the samples for 'a' (so the offset
|
|
* is negative), we switch a and b around, and use
|
|
* the reverse offset.
|
|
*/
|
|
if (offset < 0) {
|
|
const int *cylinders_map_tmp;
|
|
offset = -offset;
|
|
asamples = bsamples;
|
|
bsamples = a->samples;
|
|
as = bs;
|
|
bs = a->sample;
|
|
cylinders_map_tmp = cylinders_map_a;
|
|
cylinders_map_a = cylinders_map_b;
|
|
cylinders_map_b = cylinders_map_tmp;
|
|
}
|
|
|
|
for (;;) {
|
|
int j;
|
|
int at, bt;
|
|
struct sample sample;
|
|
|
|
if (!res)
|
|
return;
|
|
|
|
at = asamples ? as->time.seconds : -1;
|
|
bt = bsamples ? bs->time.seconds + offset : -1;
|
|
|
|
/* No samples? All done! */
|
|
if (at < 0 && bt < 0)
|
|
return;
|
|
|
|
/* Only samples from a? */
|
|
if (bt < 0) {
|
|
add_sample_a:
|
|
merge_one_sample(as, at, res);
|
|
renumber_last_sample(res, cylinders_map_a);
|
|
as++;
|
|
asamples--;
|
|
continue;
|
|
}
|
|
|
|
/* Only samples from b? */
|
|
if (at < 0) {
|
|
add_sample_b:
|
|
merge_one_sample(bs, bt, res);
|
|
renumber_last_sample(res, cylinders_map_b);
|
|
bs++;
|
|
bsamples--;
|
|
continue;
|
|
}
|
|
|
|
if (at < bt)
|
|
goto add_sample_a;
|
|
if (at > bt)
|
|
goto add_sample_b;
|
|
|
|
/* same-time sample: add a merged sample. Take the non-zero ones */
|
|
sample = *bs;
|
|
sample_renumber(&sample, 0, cylinders_map_b);
|
|
if (as->depth.mm)
|
|
sample.depth = as->depth;
|
|
if (as->temperature.mkelvin)
|
|
sample.temperature = as->temperature;
|
|
for (j = 0; j < MAX_SENSORS; ++j) {
|
|
int sensor_id;
|
|
|
|
sensor_id = cylinders_map_a[as->sensor[j]];
|
|
if (sensor_id < 0)
|
|
continue;
|
|
|
|
if (as->pressure[j].mbar)
|
|
sample.pressure[j] = as->pressure[j];
|
|
if (as->sensor[j])
|
|
sample.sensor[j] = sensor_id;
|
|
}
|
|
if (as->cns)
|
|
sample.cns = as->cns;
|
|
if (as->setpoint.mbar)
|
|
sample.setpoint = as->setpoint;
|
|
if (as->ndl.seconds)
|
|
sample.ndl = as->ndl;
|
|
if (as->stoptime.seconds)
|
|
sample.stoptime = as->stoptime;
|
|
if (as->stopdepth.mm)
|
|
sample.stopdepth = as->stopdepth;
|
|
if (as->in_deco)
|
|
sample.in_deco = true;
|
|
|
|
merge_one_sample(&sample, at, res);
|
|
|
|
as++;
|
|
bs++;
|
|
asamples--;
|
|
bsamples--;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Does the extradata key/value pair already exist in the
|
|
* supplied dive computer data?
|
|
*
|
|
* This is not hugely efficient (with the whole "do this for
|
|
* every value you merge" it's O(n**2)) but it's not like we
|
|
* have very many extra_data entries per dive computer anyway.
|
|
*/
|
|
static bool extra_data_exists(const struct extra_data *ed, const struct divecomputer *dc)
|
|
{
|
|
const struct extra_data *p;
|
|
|
|
for (p = dc->extra_data; p; p = p->next) {
|
|
if (strcmp(p->key, ed->key))
|
|
continue;
|
|
if (strcmp(p->value, ed->value))
|
|
continue;
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
/*
|
|
* Merge extra_data.
|
|
*
|
|
* The extra data from 'a' has already been copied into 'res'. So
|
|
* we really should just copy over the data from 'b' too.
|
|
*/
|
|
static void merge_extra_data(struct divecomputer *res,
|
|
const struct divecomputer *a, const struct divecomputer *b)
|
|
{
|
|
struct extra_data **ed, *src;
|
|
|
|
// Find the place to add things in the result
|
|
ed = &res->extra_data;
|
|
while (*ed)
|
|
ed = &(*ed)->next;
|
|
|
|
for (src = b->extra_data; src; src = src->next) {
|
|
if (extra_data_exists(src, a))
|
|
continue;
|
|
*ed = (extra_data *)malloc(sizeof(struct extra_data));
|
|
if (!*ed)
|
|
break;
|
|
copy_extra_data(src, *ed);
|
|
ed = &(*ed)->next;
|
|
}
|
|
|
|
// Terminate the result list
|
|
*ed = NULL;
|
|
}
|
|
|
|
static char *merge_text(const char *a, const char *b, const char *sep)
|
|
{
|
|
char *res;
|
|
if (!a && !b)
|
|
return NULL;
|
|
if (!a || !*a)
|
|
return copy_string(b);
|
|
if (!b || !*b)
|
|
return strdup(a);
|
|
if (!strcmp(a, b))
|
|
return copy_string(a);
|
|
res = (char *)malloc(strlen(a) + strlen(b) + 32);
|
|
if (!res)
|
|
return (char *)a;
|
|
sprintf(res, "%s%s%s", a, sep, b);
|
|
return res;
|
|
}
|
|
|
|
#define SORT(a, b) \
|
|
if (a != b) \
|
|
return a < b ? -1 : 1
|
|
#define SORT_FIELD(a, b, field) SORT(a->field, b->field)
|
|
|
|
static int sort_event(const struct event *a, const struct event *b, int time_a, int time_b)
|
|
{
|
|
SORT(time_a, time_b);
|
|
SORT_FIELD(a, b, type);
|
|
SORT_FIELD(a, b, flags);
|
|
SORT_FIELD(a, b, value);
|
|
return strcmp(a->name, b->name);
|
|
}
|
|
|
|
static int same_gas(const struct event *a, const struct event *b)
|
|
{
|
|
if (a->type == b->type && a->flags == b->flags && a->value == b->value && !strcmp(a->name, b->name) &&
|
|
same_gasmix(a->gas.mix, b->gas.mix)) {
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
static void event_renumber(struct event *ev, const int mapping[]);
|
|
static void add_initial_gaschange(struct dive *dive, struct divecomputer *dc, int offset, int idx);
|
|
|
|
static void merge_events(struct dive *d, struct divecomputer *res,
|
|
const struct divecomputer *src1, const struct divecomputer *src2,
|
|
const int *cylinders_map1, const int *cylinders_map2,
|
|
int offset)
|
|
{
|
|
const struct event *a, *b;
|
|
struct event **p = &res->events;
|
|
const struct event *last_gas = NULL;
|
|
|
|
/* Always use positive offsets */
|
|
if (offset < 0) {
|
|
const struct divecomputer *tmp;
|
|
const int *cylinders_map_tmp;
|
|
|
|
offset = -offset;
|
|
tmp = src1;
|
|
src1 = src2;
|
|
src2 = tmp;
|
|
|
|
cylinders_map_tmp = cylinders_map1;
|
|
cylinders_map1 = cylinders_map2;
|
|
cylinders_map2 = cylinders_map_tmp;
|
|
}
|
|
|
|
a = src1->events;
|
|
b = src2->events;
|
|
|
|
while (a || b) {
|
|
int s;
|
|
const struct event *pick;
|
|
const int *cylinders_map;
|
|
int event_offset;
|
|
|
|
if (!b)
|
|
goto pick_a;
|
|
|
|
if (!a)
|
|
goto pick_b;
|
|
|
|
s = sort_event(a, b, a->time.seconds, b->time.seconds + offset);
|
|
|
|
/* Identical events? Just skip one of them (we skip a) */
|
|
if (!s) {
|
|
a = a->next;
|
|
continue;
|
|
}
|
|
|
|
/* Otherwise, pick the one that sorts first */
|
|
if (s < 0) {
|
|
pick_a:
|
|
pick = a;
|
|
a = a->next;
|
|
event_offset = 0;
|
|
cylinders_map = cylinders_map1;
|
|
} else {
|
|
pick_b:
|
|
pick = b;
|
|
b = b->next;
|
|
event_offset = offset;
|
|
cylinders_map = cylinders_map2;
|
|
}
|
|
|
|
/*
|
|
* If that's a gas-change that matches the previous
|
|
* gas change, we'll just skip it
|
|
*/
|
|
if (event_is_gaschange(pick)) {
|
|
if (last_gas && same_gas(pick, last_gas))
|
|
continue;
|
|
last_gas = pick;
|
|
}
|
|
|
|
/* Add it to the target list */
|
|
*p = clone_event(pick);
|
|
(*p)->time.seconds += event_offset;
|
|
event_renumber(*p, cylinders_map);
|
|
p = &(*p)->next;
|
|
}
|
|
|
|
/* If the initial cylinder of a divecomputer was remapped, add a gas change event to that cylinder */
|
|
if (cylinders_map1[0] > 0)
|
|
add_initial_gaschange(d, res, 0, cylinders_map1[0]);
|
|
if (cylinders_map2[0] > 0)
|
|
add_initial_gaschange(d, res, offset, cylinders_map2[0]);
|
|
}
|
|
|
|
/* get_cylinder_idx_by_use(): Find the index of the first cylinder with a particular CCR use type.
|
|
* The index returned corresponds to that of the first cylinder with a cylinder_use that
|
|
* equals the appropriate enum value [oxygen, diluent, bailout] given by cylinder_use_type.
|
|
* A negative number returned indicates that a match could not be found.
|
|
* Call parameters: dive = the dive being processed
|
|
* cylinder_use_type = an enum, one of {oxygen, diluent, bailout} */
|
|
extern "C" int get_cylinder_idx_by_use(const struct dive *dive, enum cylinderuse cylinder_use_type)
|
|
{
|
|
int cylinder_index;
|
|
for (cylinder_index = 0; cylinder_index < dive->cylinders.nr; cylinder_index++) {
|
|
if (get_cylinder(dive, cylinder_index)->cylinder_use == cylinder_use_type)
|
|
return cylinder_index; // return the index of the cylinder with that cylinder use type
|
|
}
|
|
return -1; // negative number means cylinder_use_type not found in list of cylinders
|
|
}
|
|
|
|
/* Force an initial gaschange event to the (old) gas #0 */
|
|
static void add_initial_gaschange(struct dive *dive, struct divecomputer *dc, int offset, int idx)
|
|
{
|
|
/* if there is a gaschange event up to 30 sec after the initial event,
|
|
* refrain from adding the initial event */
|
|
const struct event *ev = dc->events;
|
|
while(ev && (ev = get_next_event(ev, "gaschange")) != NULL) {
|
|
if (ev->time.seconds > offset + 30)
|
|
break;
|
|
else if (ev->time.seconds > offset)
|
|
return;
|
|
ev = ev->next;
|
|
}
|
|
|
|
/* Old starting gas mix */
|
|
add_gas_switch_event(dive, dc, offset, idx);
|
|
}
|
|
|
|
static void sample_renumber(struct sample *s, int i, const int mapping[])
|
|
{
|
|
int j;
|
|
|
|
for (j = 0; j < MAX_SENSORS; j++) {
|
|
int sensor = -1;
|
|
|
|
if (s->sensor[j] != NO_SENSOR)
|
|
sensor = mapping[s->sensor[j]];
|
|
if (sensor == -1) {
|
|
// Remove sensor and gas pressure info
|
|
if (i == 0) {
|
|
s->sensor[j] = 0;
|
|
s->pressure[j].mbar = 0;
|
|
} else {
|
|
s->sensor[j] = s[-1].sensor[j];
|
|
s->pressure[j].mbar = s[-1].pressure[j].mbar;
|
|
}
|
|
} else {
|
|
s->sensor[j] = sensor;
|
|
}
|
|
}
|
|
}
|
|
|
|
static void renumber_last_sample(struct divecomputer *dc, const int mapping[])
|
|
{
|
|
int idx;
|
|
|
|
if (dc->samples <= 0)
|
|
return;
|
|
idx = dc->samples - 1;
|
|
sample_renumber(dc->sample + idx, idx, mapping);
|
|
}
|
|
|
|
static void event_renumber(struct event *ev, const int mapping[])
|
|
{
|
|
if (!event_is_gaschange(ev))
|
|
return;
|
|
if (ev->gas.index < 0)
|
|
return;
|
|
ev->gas.index = mapping[ev->gas.index];
|
|
}
|
|
|
|
static void dc_cylinder_renumber(struct dive *dive, struct divecomputer *dc, const int mapping[])
|
|
{
|
|
int i;
|
|
struct event *ev;
|
|
|
|
/* Remap or delete the sensor indices */
|
|
for (i = 0; i < dc->samples; i++)
|
|
sample_renumber(dc->sample + i, i, mapping);
|
|
|
|
/* Remap the gas change indices */
|
|
for (ev = dc->events; ev; ev = ev->next)
|
|
event_renumber(ev, mapping);
|
|
|
|
/* If the initial cylinder of a dive was remapped, add a gas change event to that cylinder */
|
|
if (mapping[0] > 0)
|
|
add_initial_gaschange(dive, dc, 0, mapping[0]);
|
|
}
|
|
|
|
/*
|
|
* If the cylinder indices change (due to merging dives or deleting
|
|
* cylinders in the middle), we need to change the indices in the
|
|
* dive computer data for this dive.
|
|
*
|
|
* Also note that we assume that the initial cylinder is cylinder 0,
|
|
* so if that got renamed, we need to create a fake gas change event
|
|
*/
|
|
extern "C" void cylinder_renumber(struct dive *dive, int mapping[])
|
|
{
|
|
struct divecomputer *dc;
|
|
for_each_dc (dive, dc)
|
|
dc_cylinder_renumber(dive, dc, mapping);
|
|
}
|
|
|
|
extern "C" int same_gasmix_cylinder(const cylinder_t *cyl, int cylid, const struct dive *dive, bool check_unused)
|
|
{
|
|
struct gasmix mygas = cyl->gasmix;
|
|
for (int i = 0; i < dive->cylinders.nr; i++) {
|
|
if (i == cylid)
|
|
continue;
|
|
struct gasmix gas2 = get_cylinder(dive, i)->gasmix;
|
|
if (gasmix_distance(mygas, gas2) == 0 && (is_cylinder_used(dive, i) || check_unused))
|
|
return i;
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
static int pdiff(pressure_t a, pressure_t b)
|
|
{
|
|
return a.mbar && b.mbar && a.mbar != b.mbar;
|
|
}
|
|
|
|
static int different_manual_pressures(const cylinder_t *a, const cylinder_t *b)
|
|
{
|
|
return pdiff(a->start, b->start) || pdiff(a->end, b->end);
|
|
}
|
|
|
|
/*
|
|
* Can we find an exact match for a cylinder in another dive?
|
|
* Take the "already matched" map into account, so that we
|
|
* don't match multiple similar cylinders to one target.
|
|
*
|
|
* To match, the cylinders have to have the same gasmix and the
|
|
* same cylinder use (ie OC/Diluent/Oxygen), and if pressures
|
|
* have been added manually they need to match.
|
|
*/
|
|
static int match_cylinder(const cylinder_t *cyl, const struct dive *dive, const bool try_match[])
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < dive->cylinders.nr; i++) {
|
|
const cylinder_t *target;
|
|
|
|
if (!try_match[i])
|
|
continue;
|
|
|
|
target = get_cylinder(dive, i);
|
|
if (!same_gasmix(cyl->gasmix, target->gasmix))
|
|
continue;
|
|
if (cyl->cylinder_use != target->cylinder_use)
|
|
continue;
|
|
if (different_manual_pressures(cyl, target))
|
|
continue;
|
|
|
|
/* open question: Should we check sizes too? */
|
|
return i;
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
/*
|
|
* Function used to merge manually set start or end pressures. This
|
|
* is used to merge cylinders when merging dives. We store up to two
|
|
* values for start _and_ end pressures: one derived from samples and
|
|
* one entered manually, whereby the latter takes precedence. It may
|
|
* happen that the user merges two dives where one has a manual,
|
|
* the other only a sample-derived pressure. In such a case we want to
|
|
* supplement the non-existing manual value by a sample derived one.
|
|
* Otherwise, the merged dive would end up with incomplete pressure
|
|
* information.
|
|
* The last argument to the function specifies whether the larger
|
|
* or smaller value of the two dives should be returned. Obviously,
|
|
* for the starting pressure we want the larger and for the ending
|
|
* pressure the smaller value.
|
|
*/
|
|
static pressure_t merge_pressures(pressure_t a, pressure_t sample_a, pressure_t b, pressure_t sample_b, bool take_min)
|
|
{
|
|
if (!a.mbar && !b.mbar)
|
|
return a;
|
|
if (!a.mbar)
|
|
a = sample_a;
|
|
if (!b.mbar)
|
|
b = sample_b;
|
|
if (!a.mbar)
|
|
a = b;
|
|
if (!b.mbar)
|
|
b = a;
|
|
if (take_min)
|
|
return a.mbar < b.mbar? a : b;
|
|
return a.mbar > b.mbar? a : b;
|
|
}
|
|
|
|
/*
|
|
* We matched things up so that they have the same gasmix and
|
|
* use, but we might want to fill in any missing cylinder details
|
|
* in 'a' if we had it from 'b'.
|
|
*/
|
|
static void merge_one_cylinder(cylinder_t *a, const cylinder_t *b)
|
|
{
|
|
if (!a->type.size.mliter)
|
|
a->type.size.mliter = b->type.size.mliter;
|
|
if (!a->type.workingpressure.mbar)
|
|
a->type.workingpressure.mbar = b->type.workingpressure.mbar;
|
|
if (empty_string(a->type.description))
|
|
a->type.description = copy_string(b->type.description);
|
|
|
|
/* If either cylinder has manually entered pressures, try to merge them.
|
|
* Use pressures from divecomputer samples if only one cylinder has such a value.
|
|
* Yes, this is an actual use case we encountered.
|
|
* Note that we don't merge the sample-derived pressure values, as this is
|
|
* perfomed after merging in fixup_dive() */
|
|
a->start = merge_pressures(a->start, a->sample_start, b->start, b->sample_start, false);
|
|
a->end = merge_pressures(a->end, a->sample_end, b->end, b->sample_end, true);
|
|
|
|
/* Really? */
|
|
a->gas_used.mliter += b->gas_used.mliter;
|
|
a->deco_gas_used.mliter += b->deco_gas_used.mliter;
|
|
a->bestmix_o2 = a->bestmix_o2 && b->bestmix_o2;
|
|
a->bestmix_he = a->bestmix_he && b->bestmix_he;
|
|
}
|
|
|
|
static bool cylinder_has_data(const cylinder_t *cyl)
|
|
{
|
|
return !cyl->type.size.mliter &&
|
|
!cyl->type.workingpressure.mbar &&
|
|
!cyl->type.description &&
|
|
!cyl->gasmix.o2.permille &&
|
|
!cyl->gasmix.he.permille &&
|
|
!cyl->start.mbar &&
|
|
!cyl->end.mbar &&
|
|
!cyl->sample_start.mbar &&
|
|
!cyl->sample_end.mbar &&
|
|
!cyl->gas_used.mliter &&
|
|
!cyl->deco_gas_used.mliter;
|
|
}
|
|
|
|
static bool cylinder_in_use(const struct dive *dive, int idx)
|
|
{
|
|
if (idx < 0 || idx >= dive->cylinders.nr)
|
|
return false;
|
|
|
|
/* This tests for gaschange events or pressure changes */
|
|
if (is_cylinder_used(dive, idx) || prefs.include_unused_tanks)
|
|
return true;
|
|
|
|
/* This tests for typenames or gas contents */
|
|
return cylinder_has_data(get_cylinder(dive, idx));
|
|
}
|
|
|
|
/*
|
|
* Merging cylinder information is non-trivial, because the two dive computers
|
|
* may have different ideas of what the different cylinder indexing is.
|
|
*
|
|
* Logic: take all the cylinder information from the preferred dive ('a'), and
|
|
* then try to match each of the cylinders in the other dive by the gasmix that
|
|
* is the best match and hasn't been used yet.
|
|
*
|
|
* For each dive, a cylinder-renumbering table is returned.
|
|
*/
|
|
static void merge_cylinders(struct dive *res, const struct dive *a, const struct dive *b,
|
|
int mapping_a[], int mapping_b[])
|
|
{
|
|
int i;
|
|
int max_cylinders = a->cylinders.nr + b->cylinders.nr;
|
|
auto used_in_a = std::make_unique<bool[]>(max_cylinders);
|
|
auto used_in_b = std::make_unique<bool[]>(max_cylinders);
|
|
auto try_to_match = std::make_unique<bool[]>(max_cylinders);
|
|
std::fill(try_to_match.get(), try_to_match.get() + max_cylinders, false);
|
|
|
|
/* First, clear all cylinders in destination */
|
|
clear_cylinder_table(&res->cylinders);
|
|
|
|
/* Clear all cylinder mappings */
|
|
std::fill(mapping_a, mapping_a + a->cylinders.nr, -1);
|
|
std::fill(mapping_b, mapping_b + b->cylinders.nr, -1);
|
|
|
|
/* Calculate usage map of cylinders, clear matching map */
|
|
for (i = 0; i < max_cylinders; i++) {
|
|
used_in_a[i] = cylinder_in_use(a, i);
|
|
used_in_b[i] = cylinder_in_use(b, i);
|
|
}
|
|
|
|
/*
|
|
* For each cylinder in 'a' that is used, copy it to 'res'.
|
|
* These are also potential matches for 'b' to use.
|
|
*/
|
|
for (i = 0; i < max_cylinders; i++) {
|
|
int res_nr = res->cylinders.nr;
|
|
if (!used_in_a[i])
|
|
continue;
|
|
mapping_a[i] = res_nr;
|
|
try_to_match[res_nr] = true;
|
|
add_cloned_cylinder(&res->cylinders, *get_cylinder(a, i));
|
|
}
|
|
|
|
/*
|
|
* For each cylinder in 'b' that is used, try to match it
|
|
* with an existing cylinder in 'res' from 'a'
|
|
*/
|
|
for (i = 0; i < b->cylinders.nr; i++) {
|
|
int j;
|
|
|
|
if (!used_in_b[i])
|
|
continue;
|
|
|
|
j = match_cylinder(get_cylinder(b, i), res, try_to_match.get());
|
|
|
|
/* No match? Add it to the result */
|
|
if (j < 0) {
|
|
int res_nr = res->cylinders.nr;
|
|
mapping_b[i] = res_nr;
|
|
add_cloned_cylinder(&res->cylinders, *get_cylinder(b, i));
|
|
continue;
|
|
}
|
|
|
|
/* Otherwise, merge the result to the one we found */
|
|
mapping_b[i] = j;
|
|
merge_one_cylinder(get_cylinder(res,j), get_cylinder(b, i));
|
|
|
|
/* Don't match the same target more than once */
|
|
try_to_match[j] = false;
|
|
}
|
|
}
|
|
|
|
/* Check whether a weightsystem table contains a given weightsystem */
|
|
static bool has_weightsystem(const struct weightsystem_table *t, const weightsystem_t w)
|
|
{
|
|
int i;
|
|
for (i = 0; i < t->nr; i++) {
|
|
if (same_weightsystem(w, t->weightsystems[i]))
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
static void merge_equipment(struct dive *res, const struct dive *a, const struct dive *b)
|
|
{
|
|
int i;
|
|
for (i = 0; i < a->weightsystems.nr; i++) {
|
|
if (!has_weightsystem(&res->weightsystems, a->weightsystems.weightsystems[i]))
|
|
add_cloned_weightsystem(&res->weightsystems, a->weightsystems.weightsystems[i]);
|
|
}
|
|
for (i = 0; i < b->weightsystems.nr; i++) {
|
|
if (!has_weightsystem(&res->weightsystems, b->weightsystems.weightsystems[i]))
|
|
add_cloned_weightsystem(&res->weightsystems, b->weightsystems.weightsystems[i]);
|
|
}
|
|
}
|
|
|
|
static void merge_temperatures(struct dive *res, const struct dive *a, const struct dive *b)
|
|
{
|
|
temperature_t airtemp_a = un_fixup_airtemp(a);
|
|
temperature_t airtemp_b = un_fixup_airtemp(b);
|
|
res->airtemp = airtemp_a.mkelvin ? airtemp_a : airtemp_b;
|
|
MERGE_NONZERO(res, a, b, watertemp.mkelvin);
|
|
}
|
|
|
|
/*
|
|
* Pick a trip for a dive
|
|
*/
|
|
static struct dive_trip *get_preferred_trip(const struct dive *a, const struct dive *b)
|
|
{
|
|
dive_trip_t *atrip, *btrip;
|
|
|
|
/* If only one dive has a trip, choose that */
|
|
atrip = a->divetrip;
|
|
btrip = b->divetrip;
|
|
if (!atrip)
|
|
return btrip;
|
|
if (!btrip)
|
|
return atrip;
|
|
|
|
/* Both dives have a trip - prefer the non-autogenerated one */
|
|
if (atrip->autogen && !btrip->autogen)
|
|
return btrip;
|
|
if (!atrip->autogen && btrip->autogen)
|
|
return atrip;
|
|
|
|
/* Otherwise, look at the trip data and pick the "better" one */
|
|
if (!atrip->location)
|
|
return btrip;
|
|
if (!btrip->location)
|
|
return atrip;
|
|
if (!atrip->notes)
|
|
return btrip;
|
|
if (!btrip->notes)
|
|
return atrip;
|
|
|
|
/*
|
|
* Ok, so both have location and notes.
|
|
* Pick the earlier one.
|
|
*/
|
|
if (a->when < b->when)
|
|
return atrip;
|
|
return btrip;
|
|
}
|
|
|
|
#if CURRENTLY_NOT_USED
|
|
/*
|
|
* Sample 's' is between samples 'a' and 'b'. It is 'offset' seconds before 'b'.
|
|
*
|
|
* If 's' and 'a' are at the same time, offset is 0, and b is NULL.
|
|
*/
|
|
static int compare_sample(struct sample *s, struct sample *a, struct sample *b, int offset)
|
|
{
|
|
unsigned int depth = a->depth.mm;
|
|
int diff;
|
|
|
|
if (offset) {
|
|
unsigned int interval = b->time.seconds - a->time.seconds;
|
|
unsigned int depth_a = a->depth.mm;
|
|
unsigned int depth_b = b->depth.mm;
|
|
|
|
if (offset > interval)
|
|
return -1;
|
|
|
|
/* pick the average depth, scaled by the offset from 'b' */
|
|
depth = (depth_a * offset) + (depth_b * (interval - offset));
|
|
depth /= interval;
|
|
}
|
|
diff = s->depth.mm - depth;
|
|
if (diff < 0)
|
|
diff = -diff;
|
|
/* cut off at one meter difference */
|
|
if (diff > 1000)
|
|
diff = 1000;
|
|
return diff * diff;
|
|
}
|
|
|
|
/*
|
|
* Calculate a "difference" in samples between the two dives, given
|
|
* the offset in seconds between them. Use this to find the best
|
|
* match of samples between two different dive computers.
|
|
*/
|
|
static unsigned long sample_difference(struct divecomputer *a, struct divecomputer *b, int offset)
|
|
{
|
|
int asamples = a->samples;
|
|
int bsamples = b->samples;
|
|
struct sample *as = a->sample;
|
|
struct sample *bs = b->sample;
|
|
unsigned long error = 0;
|
|
int start = -1;
|
|
|
|
if (!asamples || !bsamples)
|
|
return 0;
|
|
|
|
/*
|
|
* skip the first sample - this way we know can always look at
|
|
* as/bs[-1] to look at the samples around it in the loop.
|
|
*/
|
|
as++;
|
|
bs++;
|
|
asamples--;
|
|
bsamples--;
|
|
|
|
for (;;) {
|
|
int at, bt, diff;
|
|
|
|
|
|
/* If we run out of samples, punt */
|
|
if (!asamples)
|
|
return INT_MAX;
|
|
if (!bsamples)
|
|
return INT_MAX;
|
|
|
|
at = as->time.seconds;
|
|
bt = bs->time.seconds + offset;
|
|
|
|
/* b hasn't started yet? Ignore it */
|
|
if (bt < 0) {
|
|
bs++;
|
|
bsamples--;
|
|
continue;
|
|
}
|
|
|
|
if (at < bt) {
|
|
diff = compare_sample(as, bs - 1, bs, bt - at);
|
|
as++;
|
|
asamples--;
|
|
} else if (at > bt) {
|
|
diff = compare_sample(bs, as - 1, as, at - bt);
|
|
bs++;
|
|
bsamples--;
|
|
} else {
|
|
diff = compare_sample(as, bs, NULL, 0);
|
|
as++;
|
|
bs++;
|
|
asamples--;
|
|
bsamples--;
|
|
}
|
|
|
|
/* Invalid comparison point? */
|
|
if (diff < 0)
|
|
continue;
|
|
|
|
if (start < 0)
|
|
start = at;
|
|
|
|
error += diff;
|
|
|
|
if (at - start > 120)
|
|
break;
|
|
}
|
|
return error;
|
|
}
|
|
|
|
/*
|
|
* Dive 'a' is 'offset' seconds before dive 'b'
|
|
*
|
|
* This is *not* because the dive computers clocks aren't in sync,
|
|
* it is because the dive computers may "start" the dive at different
|
|
* points in the dive, so the sample at time X in dive 'a' is the
|
|
* same as the sample at time X+offset in dive 'b'.
|
|
*
|
|
* For example, some dive computers take longer to "wake up" when
|
|
* they sense that you are under water (ie Uemis Zurich if it was off
|
|
* when the dive started). And other dive computers have different
|
|
* depths that they activate at, etc etc.
|
|
*
|
|
* If we cannot find a shared offset, don't try to merge.
|
|
*/
|
|
static int find_sample_offset(struct divecomputer *a, struct divecomputer *b)
|
|
{
|
|
int offset, best;
|
|
unsigned long max;
|
|
|
|
/* No samples? Merge at any time (0 offset) */
|
|
if (!a->samples)
|
|
return 0;
|
|
if (!b->samples)
|
|
return 0;
|
|
|
|
/*
|
|
* Common special-case: merging a dive that came from
|
|
* the same dive computer, so the samples are identical.
|
|
* Check this first, without wasting time trying to find
|
|
* some minimal offset case.
|
|
*/
|
|
best = 0;
|
|
max = sample_difference(a, b, 0);
|
|
if (!max)
|
|
return 0;
|
|
|
|
/*
|
|
* Otherwise, look if we can find anything better within
|
|
* a thirty second window..
|
|
*/
|
|
for (offset = -30; offset <= 30; offset++) {
|
|
unsigned long diff;
|
|
|
|
diff = sample_difference(a, b, offset);
|
|
if (diff > max)
|
|
continue;
|
|
best = offset;
|
|
max = diff;
|
|
}
|
|
|
|
return best;
|
|
}
|
|
#endif
|
|
|
|
/*
|
|
* Are a and b "similar" values, when given a reasonable lower end expected
|
|
* difference?
|
|
*
|
|
* So for example, we'd expect different dive computers to give different
|
|
* max. depth readings. You might have them on different arms, and they
|
|
* have different pressure sensors and possibly different ideas about
|
|
* water salinity etc.
|
|
*
|
|
* So have an expected minimum difference, but also allow a larger relative
|
|
* error value.
|
|
*/
|
|
static int similar(unsigned long a, unsigned long b, unsigned long expected)
|
|
{
|
|
if (!a && !b)
|
|
return 1;
|
|
|
|
if (a && b) {
|
|
unsigned long min, max, diff;
|
|
|
|
min = a;
|
|
max = b;
|
|
if (a > b) {
|
|
min = b;
|
|
max = a;
|
|
}
|
|
diff = max - min;
|
|
|
|
/* Smaller than expected difference? */
|
|
if (diff < expected)
|
|
return 1;
|
|
/* Error less than 10% or the maximum */
|
|
if (diff * 10 < max)
|
|
return 1;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Match every dive computer against each other to see if
|
|
* we have a matching dive.
|
|
*
|
|
* Return values:
|
|
* -1 for "is definitely *NOT* the same dive"
|
|
* 0 for "don't know"
|
|
* 1 for "is definitely the same dive"
|
|
*/
|
|
static int match_dc_dive(const struct divecomputer *a, const struct divecomputer *b)
|
|
{
|
|
do {
|
|
const struct divecomputer *tmp = b;
|
|
do {
|
|
int match = match_one_dc(a, tmp);
|
|
if (match)
|
|
return match;
|
|
tmp = tmp->next;
|
|
} while (tmp);
|
|
a = a->next;
|
|
} while (a);
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Do we want to automatically try to merge two dives that
|
|
* look like they are the same dive?
|
|
*
|
|
* This happens quite commonly because you download a dive
|
|
* that you already had, or perhaps because you maintained
|
|
* multiple dive logs and want to load them all together
|
|
* (possibly one of them was imported from another dive log
|
|
* application entirely).
|
|
*
|
|
* NOTE! We mainly look at the dive time, but it can differ
|
|
* between two dives due to a few issues:
|
|
*
|
|
* - rounding the dive date to the nearest minute in other dive
|
|
* applications
|
|
*
|
|
* - dive computers with "relative datestamps" (ie the dive
|
|
* computer doesn't actually record an absolute date at all,
|
|
* but instead at download-time synchronizes its internal
|
|
* time with real-time on the downloading computer)
|
|
*
|
|
* - using multiple dive computers with different real time on
|
|
* the same dive
|
|
*
|
|
* We do not merge dives that look radically different, and if
|
|
* the dates are *too* far off the user will have to join two
|
|
* dives together manually. But this tries to handle the sane
|
|
* cases.
|
|
*/
|
|
static int likely_same_dive(const struct dive *a, const struct dive *b)
|
|
{
|
|
int match, fuzz = 20 * 60;
|
|
|
|
/* don't merge manually added dives with anything */
|
|
if (is_manually_added_dc(&a->dc) ||
|
|
is_manually_added_dc(&b->dc))
|
|
return 0;
|
|
|
|
/*
|
|
* Do some basic sanity testing of the values we
|
|
* have filled in during 'fixup_dive()'
|
|
*/
|
|
if (!similar(a->maxdepth.mm, b->maxdepth.mm, 1000) ||
|
|
(a->meandepth.mm && b->meandepth.mm && !similar(a->meandepth.mm, b->meandepth.mm, 1000)) ||
|
|
!a->duration.seconds || !b->duration.seconds ||
|
|
!similar(a->duration.seconds, b->duration.seconds, 5 * 60))
|
|
return 0;
|
|
|
|
/* See if we can get an exact match on the dive computer */
|
|
match = match_dc_dive(&a->dc, &b->dc);
|
|
if (match)
|
|
return match > 0;
|
|
|
|
/*
|
|
* Allow a time difference due to dive computer time
|
|
* setting etc. Check if they overlap.
|
|
*/
|
|
fuzz = std::max(a->duration.seconds, b->duration.seconds) / 2;
|
|
if (fuzz < 60)
|
|
fuzz = 60;
|
|
|
|
return (a->when <= b->when + fuzz) && (a->when >= b->when - fuzz);
|
|
}
|
|
|
|
/*
|
|
* This could do a lot more merging. Right now it really only
|
|
* merges almost exact duplicates - something that happens easily
|
|
* with overlapping dive downloads.
|
|
*
|
|
* If new dives are merged into the dive table, dive a is supposed to
|
|
* be the old dive and dive b is supposed to be the newly imported
|
|
* dive. If the flag "prefer_downloaded" is set, data of the latter
|
|
* will take priority over the former.
|
|
*
|
|
* Attn: The dive_site parameter of the dive will be set, but the caller
|
|
* still has to register the dive in the dive site!
|
|
*/
|
|
extern "C" struct dive *try_to_merge(struct dive *a, struct dive *b, bool prefer_downloaded)
|
|
{
|
|
struct dive *res;
|
|
struct dive_site *site;
|
|
|
|
if (!likely_same_dive(a, b))
|
|
return NULL;
|
|
|
|
res = merge_dives(a, b, 0, prefer_downloaded, NULL, &site);
|
|
res->dive_site = site; /* Caller has to call add_dive_to_dive_site()! */
|
|
return res;
|
|
}
|
|
|
|
static int same_sample(struct sample *a, struct sample *b)
|
|
{
|
|
if (a->time.seconds != b->time.seconds)
|
|
return 0;
|
|
if (a->depth.mm != b->depth.mm)
|
|
return 0;
|
|
if (a->temperature.mkelvin != b->temperature.mkelvin)
|
|
return 0;
|
|
if (a->pressure[0].mbar != b->pressure[0].mbar)
|
|
return 0;
|
|
return a->sensor[0] == b->sensor[0];
|
|
}
|
|
|
|
static int same_dc(struct divecomputer *a, struct divecomputer *b)
|
|
{
|
|
int i;
|
|
const struct event *eva, *evb;
|
|
|
|
i = match_one_dc(a, b);
|
|
if (i)
|
|
return i > 0;
|
|
|
|
if (a->when && b->when && a->when != b->when)
|
|
return 0;
|
|
if (a->samples != b->samples)
|
|
return 0;
|
|
for (i = 0; i < a->samples; i++)
|
|
if (!same_sample(a->sample + i, b->sample + i))
|
|
return 0;
|
|
eva = a->events;
|
|
evb = b->events;
|
|
while (eva && evb) {
|
|
if (!same_event(eva, evb))
|
|
return 0;
|
|
eva = eva->next;
|
|
evb = evb->next;
|
|
}
|
|
return eva == evb;
|
|
}
|
|
|
|
static int might_be_same_device(const struct divecomputer *a, const struct divecomputer *b)
|
|
{
|
|
/* No dive computer model? That matches anything */
|
|
if (!a->model || !b->model)
|
|
return 1;
|
|
|
|
/* Otherwise at least the model names have to match */
|
|
if (strcasecmp(a->model, b->model))
|
|
return 0;
|
|
|
|
/* No device ID? Match */
|
|
if (!a->deviceid || !b->deviceid)
|
|
return 1;
|
|
|
|
return a->deviceid == b->deviceid;
|
|
}
|
|
|
|
static void remove_redundant_dc(struct divecomputer *dc, int prefer_downloaded)
|
|
{
|
|
do {
|
|
struct divecomputer **p = &dc->next;
|
|
|
|
/* Check this dc against all the following ones.. */
|
|
while (*p) {
|
|
struct divecomputer *check = *p;
|
|
if (same_dc(dc, check) || (prefer_downloaded && might_be_same_device(dc, check))) {
|
|
*p = check->next;
|
|
check->next = NULL;
|
|
free_dc(check);
|
|
continue;
|
|
}
|
|
p = &check->next;
|
|
}
|
|
|
|
/* .. and then continue down the chain, but we */
|
|
prefer_downloaded = 0;
|
|
dc = dc->next;
|
|
} while (dc);
|
|
}
|
|
|
|
static const struct divecomputer *find_matching_computer(const struct divecomputer *match, const struct divecomputer *list)
|
|
{
|
|
const struct divecomputer *p;
|
|
|
|
while ((p = list) != NULL) {
|
|
list = list->next;
|
|
|
|
if (might_be_same_device(match, p))
|
|
break;
|
|
}
|
|
return p;
|
|
}
|
|
|
|
static void copy_dive_computer(struct divecomputer *res, const struct divecomputer *a)
|
|
{
|
|
*res = *a;
|
|
res->model = copy_string(a->model);
|
|
res->serial = copy_string(a->serial);
|
|
res->fw_version = copy_string(a->fw_version);
|
|
STRUCTURED_LIST_COPY(struct extra_data, a->extra_data, res->extra_data, copy_extra_data);
|
|
res->samples = res->alloc_samples = 0;
|
|
res->sample = NULL;
|
|
res->events = NULL;
|
|
res->next = NULL;
|
|
}
|
|
|
|
/*
|
|
* Join dive computers with a specific time offset between
|
|
* them.
|
|
*
|
|
* Use the dive computer ID's (or names, if ID's are missing)
|
|
* to match them up. If we find a matching dive computer, we
|
|
* merge them. If not, we just take the data from 'a'.
|
|
*/
|
|
static void interleave_dive_computers(struct dive *d, struct divecomputer *res,
|
|
const struct divecomputer *a, const struct divecomputer *b,
|
|
const int cylinders_map_a[], const int cylinders_map_b[],
|
|
int offset)
|
|
{
|
|
do {
|
|
const struct divecomputer *match;
|
|
|
|
copy_dive_computer(res, a);
|
|
|
|
match = find_matching_computer(a, b);
|
|
if (match) {
|
|
merge_events(d, res, a, match, cylinders_map_a, cylinders_map_b, offset);
|
|
merge_samples(res, a, match, cylinders_map_a, cylinders_map_b, offset);
|
|
merge_extra_data(res, a, match);
|
|
/* Use the diveid of the later dive! */
|
|
if (offset > 0)
|
|
res->diveid = match->diveid;
|
|
} else {
|
|
copy_dc_renumber(d, a, res, cylinders_map_a);
|
|
}
|
|
a = a->next;
|
|
if (!a)
|
|
break;
|
|
res->next = (divecomputer *)calloc(1, sizeof(struct divecomputer));
|
|
res = res->next;
|
|
} while (res);
|
|
}
|
|
|
|
|
|
/*
|
|
* Join dive computer information.
|
|
*
|
|
* If we have old-style dive computer information (no model
|
|
* name etc), we will prefer a new-style one and just throw
|
|
* away the old. We're assuming it's a re-download.
|
|
*
|
|
* Otherwise, we'll just try to keep all the information,
|
|
* unless the user has specified that they prefer the
|
|
* downloaded computer, in which case we'll aggressively
|
|
* try to throw out old information that *might* be from
|
|
* that one.
|
|
*/
|
|
static void join_dive_computers(struct dive *d, struct divecomputer *res,
|
|
const struct divecomputer *a, const struct divecomputer *b,
|
|
const int cylinders_map_a[], const int cylinders_map_b[],
|
|
int prefer_downloaded)
|
|
{
|
|
struct divecomputer *tmp;
|
|
|
|
if (a->model && !b->model) {
|
|
copy_dc_renumber(d, a, res, cylinders_map_a);
|
|
return;
|
|
}
|
|
if (b->model && !a->model) {
|
|
copy_dc_renumber(d, b, res, cylinders_map_b);
|
|
return;
|
|
}
|
|
|
|
copy_dc_renumber(d, a, res, cylinders_map_a);
|
|
tmp = res;
|
|
while (tmp->next)
|
|
tmp = tmp->next;
|
|
|
|
tmp->next = (divecomputer *)calloc(1, sizeof(*tmp));
|
|
copy_dc_renumber(d, b, tmp->next, cylinders_map_b);
|
|
|
|
remove_redundant_dc(res, prefer_downloaded);
|
|
}
|
|
|
|
static bool has_dc_type(const struct dive *dive, bool dc_is_planner)
|
|
{
|
|
const struct divecomputer *dc = &dive->dc;
|
|
|
|
while (dc) {
|
|
if (is_dc_planner(dc) == dc_is_planner)
|
|
return true;
|
|
dc = dc->next;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
// Does this dive have a dive computer for which is_dc_planner has value planned
|
|
extern "C" bool is_planned(const struct dive *dive)
|
|
{
|
|
return has_dc_type(dive, true);
|
|
}
|
|
|
|
extern "C" bool is_logged(const struct dive *dive)
|
|
{
|
|
return has_dc_type(dive, false);
|
|
}
|
|
|
|
/*
|
|
* Merging two dives can be subtle, because there's two different ways
|
|
* of merging:
|
|
*
|
|
* (a) two distinctly _different_ dives that have the same dive computer
|
|
* are merged into one longer dive, because the user asked for it
|
|
* in the divelist.
|
|
*
|
|
* Because this case is with the same dive computer, we *know* the
|
|
* two must have a different start time, and "offset" is the relative
|
|
* time difference between the two.
|
|
*
|
|
* (b) two different dive computers that we might want to merge into
|
|
* one single dive with multiple dive computers.
|
|
*
|
|
* This is the "try_to_merge()" case, which will have offset == 0,
|
|
* even if the dive times might be different.
|
|
*
|
|
* If new dives are merged into the dive table, dive a is supposed to
|
|
* be the old dive and dive b is supposed to be the newly imported
|
|
* dive. If the flag "prefer_downloaded" is set, data of the latter
|
|
* will take priority over the former.
|
|
*
|
|
* The trip the new dive should be associated with (if any) is returned
|
|
* in the "trip" output parameter.
|
|
*
|
|
* The dive site the new dive should be added to (if any) is returned
|
|
* in the "dive_site" output parameter.
|
|
*/
|
|
extern "C" struct dive *merge_dives(const struct dive *a, const struct dive *b, int offset, bool prefer_downloaded, struct dive_trip **trip, struct dive_site **site)
|
|
{
|
|
struct dive *res = alloc_dive();
|
|
|
|
if (offset) {
|
|
/*
|
|
* If "likely_same_dive()" returns true, that means that
|
|
* it is *not* the same dive computer, and we do not want
|
|
* to try to turn it into a single longer dive. So we'd
|
|
* join them as two separate dive computers at zero offset.
|
|
*/
|
|
if (likely_same_dive(a, b))
|
|
offset = 0;
|
|
}
|
|
|
|
if (is_dc_planner(&a->dc)) {
|
|
const struct dive *tmp = a;
|
|
a = b;
|
|
b = tmp;
|
|
}
|
|
res->when = prefer_downloaded ? b->when : a->when;
|
|
res->selected = a->selected || b->selected;
|
|
if (trip)
|
|
*trip = get_preferred_trip(a, b);
|
|
MERGE_TXT(res, a, b, notes, "\n--\n");
|
|
MERGE_TXT(res, a, b, buddy, ", ");
|
|
MERGE_TXT(res, a, b, diveguide, ", ");
|
|
MERGE_MAX(res, a, b, rating);
|
|
MERGE_TXT(res, a, b, suit, ", ");
|
|
MERGE_MAX(res, a, b, number);
|
|
MERGE_NONZERO(res, a, b, visibility);
|
|
MERGE_NONZERO(res, a, b, wavesize);
|
|
MERGE_NONZERO(res, a, b, current);
|
|
MERGE_NONZERO(res, a, b, surge);
|
|
MERGE_NONZERO(res, a, b, chill);
|
|
copy_pictures(a->pictures.nr ? &a->pictures : &b->pictures, &res->pictures);
|
|
taglist_merge(&res->tag_list, a->tag_list, b->tag_list);
|
|
/* if we get dives without any gas / cylinder information in an import, make sure
|
|
* that there is at leatst one entry in the cylinder map for that dive */
|
|
auto cylinders_map_a = std::make_unique<int[]>(std::max(1, a->cylinders.nr));
|
|
auto cylinders_map_b = std::make_unique<int[]>(std::max(1, b->cylinders.nr));
|
|
merge_cylinders(res, a, b, cylinders_map_a.get(), cylinders_map_b.get());
|
|
merge_equipment(res, a, b);
|
|
merge_temperatures(res, a, b);
|
|
if (prefer_downloaded) {
|
|
/* If we prefer downloaded, do those first, and get rid of "might be same" computers */
|
|
join_dive_computers(res, &res->dc, &b->dc, &a->dc, cylinders_map_b.get(), cylinders_map_a.get(), 1);
|
|
} else if (offset && might_be_same_device(&a->dc, &b->dc))
|
|
interleave_dive_computers(res, &res->dc, &a->dc, &b->dc, cylinders_map_a.get(), cylinders_map_b.get(), offset);
|
|
else
|
|
join_dive_computers(res, &res->dc, &a->dc, &b->dc, cylinders_map_a.get(), cylinders_map_b.get(), 0);
|
|
|
|
/* The CNS values will be recalculated from the sample in fixup_dive() */
|
|
res->cns = res->maxcns = 0;
|
|
|
|
/* we take the first dive site, unless it's empty */
|
|
*site = a->dive_site && !dive_site_is_empty(a->dive_site) ? a->dive_site : b->dive_site;
|
|
if (!dive_site_has_gps_location(*site) && dive_site_has_gps_location(b->dive_site)) {
|
|
/* we picked the first dive site and that didn't have GPS data, but the new dive has
|
|
* GPS data (that could be a download from a GPS enabled dive computer).
|
|
* Keep the dive site, but add the GPS data */
|
|
(*site)->location = b->dive_site->location;
|
|
}
|
|
fixup_dive(res);
|
|
return res;
|
|
}
|
|
|
|
// copy_dive(), but retaining the new ID for the copied dive
|
|
static struct dive *create_new_copy(const struct dive *from)
|
|
{
|
|
struct dive *to = alloc_dive();
|
|
int id;
|
|
|
|
// alloc_dive() gave us a new ID, we just need to
|
|
// make sure it's not overwritten.
|
|
id = to->id;
|
|
copy_dive(from, to);
|
|
to->id = id;
|
|
return to;
|
|
}
|
|
|
|
struct start_end_pressure {
|
|
pressure_t start;
|
|
pressure_t end;
|
|
};
|
|
|
|
static void force_fixup_dive(struct dive *d)
|
|
{
|
|
struct divecomputer *dc = &d->dc;
|
|
int old_temp = dc->watertemp.mkelvin;
|
|
int old_mintemp = d->mintemp.mkelvin;
|
|
int old_maxtemp = d->maxtemp.mkelvin;
|
|
duration_t old_duration = d->duration;
|
|
std::vector<start_end_pressure> old_pressures(d->cylinders.nr);
|
|
|
|
d->maxdepth.mm = 0;
|
|
dc->maxdepth.mm = 0;
|
|
d->watertemp.mkelvin = 0;
|
|
dc->watertemp.mkelvin = 0;
|
|
d->duration.seconds = 0;
|
|
d->maxtemp.mkelvin = 0;
|
|
d->mintemp.mkelvin = 0;
|
|
for (int i = 0; i < d->cylinders.nr; i++) {
|
|
cylinder_t *cyl = get_cylinder(d, i);
|
|
old_pressures[i].start = cyl->start;
|
|
old_pressures[i].end = cyl->end;
|
|
cyl->start.mbar = 0;
|
|
cyl->end.mbar = 0;
|
|
}
|
|
|
|
fixup_dive(d);
|
|
|
|
if (!d->watertemp.mkelvin)
|
|
d->watertemp.mkelvin = old_temp;
|
|
|
|
if (!dc->watertemp.mkelvin)
|
|
dc->watertemp.mkelvin = old_temp;
|
|
|
|
if (!d->maxtemp.mkelvin)
|
|
d->maxtemp.mkelvin = old_maxtemp;
|
|
|
|
if (!d->mintemp.mkelvin)
|
|
d->mintemp.mkelvin = old_mintemp;
|
|
|
|
if (!d->duration.seconds)
|
|
d->duration = old_duration;
|
|
for (int i = 0; i < d->cylinders.nr; i++) {
|
|
if (!get_cylinder(d, i)->start.mbar)
|
|
get_cylinder(d, i)->start = old_pressures[i].start;
|
|
if (!get_cylinder(d, i)->end.mbar)
|
|
get_cylinder(d, i)->end = old_pressures[i].end;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Split a dive that has a surface interval from samples 'a' to 'b'
|
|
* into two dives, but don't add them to the log yet.
|
|
* Returns the nr of the old dive or <0 on failure.
|
|
* Moreover, on failure both output dives are set to NULL.
|
|
* On success, the newly allocated dives are returned in out1 and out2.
|
|
*/
|
|
static int split_dive_at(const struct dive *dive, int a, int b, struct dive **out1, struct dive **out2)
|
|
{
|
|
int i, nr;
|
|
int32_t t;
|
|
struct dive *d1, *d2;
|
|
struct divecomputer *dc1, *dc2;
|
|
struct event *event, **evp;
|
|
|
|
/* if we can't find the dive in the dive list, don't bother */
|
|
if ((nr = get_divenr(dive)) < 0)
|
|
return -1;
|
|
|
|
/* Splitting should leave at least 3 samples per dive */
|
|
if (a < 3 || b > dive->dc.samples - 4)
|
|
return -1;
|
|
|
|
/* We're not trying to be efficient here.. */
|
|
d1 = create_new_copy(dive);
|
|
d2 = create_new_copy(dive);
|
|
d1->divetrip = d2->divetrip = 0;
|
|
|
|
/* now unselect the first first segment so we don't keep all
|
|
* dives selected by mistake. But do keep the second one selected
|
|
* so the algorithm keeps splitting the dive further */
|
|
d1->selected = false;
|
|
|
|
dc1 = &d1->dc;
|
|
dc2 = &d2->dc;
|
|
/*
|
|
* Cut off the samples of d1 at the beginning
|
|
* of the interval.
|
|
*/
|
|
dc1->samples = a;
|
|
|
|
/* And get rid of the 'b' first samples of d2 */
|
|
dc2->samples -= b;
|
|
memmove(dc2->sample, dc2->sample+b, dc2->samples * sizeof(struct sample));
|
|
|
|
/* Now the secondary dive computers */
|
|
t = dc2->sample[0].time.seconds;
|
|
while ((dc1 = dc1->next)) {
|
|
i = 0;
|
|
while (dc1->samples < i && dc1->sample[i].time.seconds <= t)
|
|
++i;
|
|
dc1->samples = i;
|
|
}
|
|
while ((dc2 = dc2->next)) {
|
|
i = 0;
|
|
while (dc2->samples < i && dc2->sample[i].time.seconds < t)
|
|
++i;
|
|
dc2->samples -= i;
|
|
memmove(dc2->sample, dc2->sample + i, dc2->samples * sizeof(struct sample));
|
|
}
|
|
dc1 = &d1->dc;
|
|
dc2 = &d2->dc;
|
|
/*
|
|
* This is where we cut off events from d1,
|
|
* and shift everything in d2
|
|
*/
|
|
d2->when += t;
|
|
while (dc1 && dc2) {
|
|
dc2->when += t;
|
|
for (i = 0; i < dc2->samples; i++)
|
|
dc2->sample[i].time.seconds -= t;
|
|
|
|
/* Remove the events past 't' from d1 */
|
|
evp = &dc1->events;
|
|
while ((event = *evp) != NULL && event->time.seconds < t)
|
|
evp = &event->next;
|
|
*evp = NULL;
|
|
while (event) {
|
|
struct event *next = event->next;
|
|
free(event);
|
|
event = next;
|
|
}
|
|
|
|
/* Remove the events before 't' from d2, and shift the rest */
|
|
evp = &dc2->events;
|
|
while ((event = *evp) != NULL) {
|
|
if (event->time.seconds < t) {
|
|
*evp = event->next;
|
|
free(event);
|
|
} else {
|
|
event->time.seconds -= t;
|
|
}
|
|
}
|
|
dc1 = dc1->next;
|
|
dc2 = dc2->next;
|
|
}
|
|
|
|
force_fixup_dive(d1);
|
|
force_fixup_dive(d2);
|
|
|
|
/*
|
|
* Was the dive numbered? If it was the last dive, then we'll
|
|
* increment the dive number for the tail part that we split off.
|
|
* Otherwise the tail is unnumbered.
|
|
*/
|
|
if (d2->number) {
|
|
if (divelog.dives->nr == nr + 1)
|
|
d2->number++;
|
|
else
|
|
d2->number = 0;
|
|
}
|
|
|
|
*out1 = d1;
|
|
*out2 = d2;
|
|
return nr;
|
|
}
|
|
|
|
/* in freedive mode we split for as little as 10 seconds on the surface,
|
|
* otherwise we use a minute */
|
|
static bool should_split(const struct divecomputer *dc, int t1, int t2)
|
|
{
|
|
int threshold = dc->divemode == FREEDIVE ? 10 : 60;
|
|
|
|
return t2 - t1 >= threshold;
|
|
}
|
|
|
|
/*
|
|
* Try to split a dive into multiple dives at a surface interval point.
|
|
*
|
|
* NOTE! We will split when there is at least one surface event that has
|
|
* non-surface events on both sides.
|
|
*
|
|
* The surface interval points are determined using the first dive computer.
|
|
*
|
|
* In other words, this is a (simplified) reversal of the dive merging.
|
|
*/
|
|
extern "C" int split_dive(const struct dive *dive, struct dive **new1, struct dive **new2)
|
|
{
|
|
int i;
|
|
int at_surface, surface_start;
|
|
const struct divecomputer *dc;
|
|
|
|
*new1 = *new2 = NULL;
|
|
if (!dive)
|
|
return -1;
|
|
|
|
dc = &dive->dc;
|
|
surface_start = 0;
|
|
at_surface = 1;
|
|
for (i = 1; i < dc->samples; i++) {
|
|
struct sample *sample = dc->sample+i;
|
|
int surface_sample = sample->depth.mm < SURFACE_THRESHOLD;
|
|
|
|
/*
|
|
* We care about the transition from and to depth 0,
|
|
* not about the depth staying similar.
|
|
*/
|
|
if (at_surface == surface_sample)
|
|
continue;
|
|
at_surface = surface_sample;
|
|
|
|
// Did it become surface after having been non-surface? We found the start
|
|
if (at_surface) {
|
|
surface_start = i;
|
|
continue;
|
|
}
|
|
|
|
// Going down again? We want at least a minute from
|
|
// the surface start.
|
|
if (!surface_start)
|
|
continue;
|
|
if (!should_split(dc, dc->sample[surface_start].time.seconds, sample[-1].time.seconds))
|
|
continue;
|
|
|
|
return split_dive_at(dive, surface_start, i-1, new1, new2);
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
extern "C" int split_dive_at_time(const struct dive *dive, duration_t time, struct dive **new1, struct dive **new2)
|
|
{
|
|
int i = 0;
|
|
|
|
if (!dive)
|
|
return -1;
|
|
|
|
struct sample *sample = dive->dc.sample;
|
|
*new1 = *new2 = NULL;
|
|
while(sample->time.seconds < time.seconds) {
|
|
++sample;
|
|
++i;
|
|
if (dive->dc.samples == i)
|
|
return -1;
|
|
}
|
|
return split_dive_at(dive, i, i - 1, new1, new2);
|
|
}
|
|
|
|
/*
|
|
* "dc_maxtime()" is how much total time this dive computer
|
|
* has for this dive. Note that it can differ from "duration"
|
|
* if there are surface events in the middle.
|
|
*
|
|
* Still, we do ignore all but the last surface samples from the
|
|
* end, because some divecomputers just generate lots of them.
|
|
*/
|
|
static inline int dc_totaltime(const struct divecomputer *dc)
|
|
{
|
|
int time = dc->duration.seconds;
|
|
int nr = dc->samples;
|
|
|
|
while (nr--) {
|
|
struct sample *s = dc->sample + nr;
|
|
time = s->time.seconds;
|
|
if (s->depth.mm >= SURFACE_THRESHOLD)
|
|
break;
|
|
}
|
|
return time;
|
|
}
|
|
|
|
/*
|
|
* The end of a dive is actually not trivial, because "duration"
|
|
* is not the duration until the end, but the time we spend under
|
|
* water, which can be very different if there are surface events
|
|
* during the dive.
|
|
*
|
|
* So walk the dive computers, looking for the longest actual
|
|
* time in the samples (and just default to the dive duration if
|
|
* there are no samples).
|
|
*/
|
|
static inline int dive_totaltime(const struct dive *dive)
|
|
{
|
|
int time = dive->duration.seconds;
|
|
const struct divecomputer *dc;
|
|
|
|
for_each_relevant_dc(dive, dc) {
|
|
int dc_time = dc_totaltime(dc);
|
|
if (dc_time > time)
|
|
time = dc_time;
|
|
}
|
|
return time;
|
|
}
|
|
|
|
extern "C" timestamp_t dive_endtime(const struct dive *dive)
|
|
{
|
|
return dive->when + dive_totaltime(dive);
|
|
}
|
|
|
|
extern "C" bool time_during_dive_with_offset(const struct dive *dive, timestamp_t when, timestamp_t offset)
|
|
{
|
|
timestamp_t start = dive->when;
|
|
timestamp_t end = dive_endtime(dive);
|
|
return start - offset <= when && when <= end + offset;
|
|
}
|
|
|
|
/* this sets a usually unused copy of the preferences with the units
|
|
* that were active the last time the dive list was saved to git storage
|
|
* (this isn't used in XML files); storing the unit preferences in the
|
|
* data file is usually pointless (that's a setting of the software,
|
|
* not a property of the data), but it's a great hint of what the user
|
|
* might expect to see when creating a backend service that visualizes
|
|
* the dive list without Subsurface running - so this is basically a
|
|
* functionality for the core library that Subsurface itself doesn't
|
|
* use but that another consumer of the library (like an HTML exporter)
|
|
* will need */
|
|
extern "C" void set_informational_units(const char *units)
|
|
{
|
|
if (strstr(units, "METRIC")) {
|
|
git_prefs.unit_system = METRIC;
|
|
} else if (strstr(units, "IMPERIAL")) {
|
|
git_prefs.unit_system = IMPERIAL;
|
|
} else if (strstr(units, "PERSONALIZE")) {
|
|
git_prefs.unit_system = PERSONALIZE;
|
|
if (strstr(units, "METERS"))
|
|
git_prefs.units.length = units::METERS;
|
|
if (strstr(units, "FEET"))
|
|
git_prefs.units.length = units::FEET;
|
|
if (strstr(units, "LITER"))
|
|
git_prefs.units.volume = units::LITER;
|
|
if (strstr(units, "CUFT"))
|
|
git_prefs.units.volume = units::CUFT;
|
|
if (strstr(units, "BAR"))
|
|
git_prefs.units.pressure = units::BAR;
|
|
if (strstr(units, "PSI"))
|
|
git_prefs.units.pressure = units::PSI;
|
|
if (strstr(units, "CELSIUS"))
|
|
git_prefs.units.temperature = units::CELSIUS;
|
|
if (strstr(units, "FAHRENHEIT"))
|
|
git_prefs.units.temperature = units::FAHRENHEIT;
|
|
if (strstr(units, "KG"))
|
|
git_prefs.units.weight = units::KG;
|
|
if (strstr(units, "LBS"))
|
|
git_prefs.units.weight = units::LBS;
|
|
if (strstr(units, "SECONDS"))
|
|
git_prefs.units.vertical_speed_time = units::SECONDS;
|
|
if (strstr(units, "MINUTES"))
|
|
git_prefs.units.vertical_speed_time = units::MINUTES;
|
|
}
|
|
|
|
}
|
|
|
|
extern "C" void set_git_prefs(const char *prefs)
|
|
{
|
|
if (strstr(prefs, "TANKBAR"))
|
|
git_prefs.tankbar = 1;
|
|
if (strstr(prefs, "SHOW_SETPOINT"))
|
|
git_prefs.show_ccr_setpoint = 1;
|
|
if (strstr(prefs, "SHOW_SENSORS"))
|
|
git_prefs.show_ccr_sensors = 1;
|
|
if (strstr(prefs, "PO2_GRAPH"))
|
|
git_prefs.pp_graphs.po2 = 1;
|
|
}
|
|
|
|
/* clones a dive and moves given dive computer to front */
|
|
extern "C" struct dive *make_first_dc(const struct dive *d, int dc_number)
|
|
{
|
|
struct dive *res;
|
|
struct divecomputer *dc, *newdc, *old_dc;
|
|
|
|
/* copy the dive */
|
|
res = alloc_dive();
|
|
copy_dive(d, res);
|
|
|
|
/* make a new unique id, since we still can't handle two equal ids */
|
|
res->id = dive_getUniqID();
|
|
|
|
if (dc_number == 0)
|
|
return res;
|
|
|
|
dc = &res->dc;
|
|
newdc = (divecomputer *)malloc(sizeof(*newdc));
|
|
old_dc = get_dive_dc(res, dc_number);
|
|
|
|
/* skip the current DC in the linked list */
|
|
for (dc = &res->dc; dc && dc->next != old_dc; dc = dc->next)
|
|
;
|
|
if (!dc) {
|
|
free(newdc);
|
|
report_info("data inconsistent: can't find the current DC");
|
|
return res;
|
|
}
|
|
dc->next = old_dc->next;
|
|
*newdc = res->dc;
|
|
res->dc = *old_dc;
|
|
res->dc.next = newdc;
|
|
free(old_dc);
|
|
|
|
return res;
|
|
}
|
|
|
|
static void delete_divecomputer(struct dive *d, int num)
|
|
{
|
|
int i;
|
|
|
|
/* Refuse to delete the last dive computer */
|
|
if (!d->dc.next)
|
|
return;
|
|
|
|
if (num == 0) {
|
|
/* remove the first one, so copy the second one in place of the first and free the second one
|
|
* be careful about freeing the no longer needed structures - since we copy things around we can't use free_dc()*/
|
|
struct divecomputer *fdc = d->dc.next;
|
|
free_dc_contents(&d->dc);
|
|
memcpy(&d->dc, fdc, sizeof(struct divecomputer));
|
|
free(fdc);
|
|
} else {
|
|
struct divecomputer *pdc = &d->dc;
|
|
for (i = 0; i < num - 1 && pdc; i++)
|
|
pdc = pdc->next;
|
|
if (pdc && pdc->next) {
|
|
struct divecomputer *dc = pdc->next;
|
|
pdc->next = dc->next;
|
|
free_dc(dc);
|
|
}
|
|
}
|
|
}
|
|
|
|
/* Clone a dive and delete goven dive computer */
|
|
extern "C" struct dive *clone_delete_divecomputer(const struct dive *d, int dc_number)
|
|
{
|
|
struct dive *res;
|
|
|
|
/* copy the dive */
|
|
res = alloc_dive();
|
|
copy_dive(d, res);
|
|
|
|
/* make a new unique id, since we still can't handle two equal ids */
|
|
res->id = dive_getUniqID();
|
|
|
|
delete_divecomputer(res, dc_number);
|
|
force_fixup_dive(res);
|
|
|
|
return res;
|
|
}
|
|
|
|
/*
|
|
* This splits the dive src by dive computer. The first output dive has all
|
|
* dive computers except num, the second only dive computer num.
|
|
* The dives will not be associated with a trip.
|
|
* On error, both output parameters are set to NULL.
|
|
*/
|
|
extern "C" void split_divecomputer(const struct dive *src, int num, struct dive **out1, struct dive **out2)
|
|
{
|
|
const struct divecomputer *srcdc = get_dive_dc_const(src, num);
|
|
|
|
if (src && srcdc) {
|
|
// Copy the dive, but only using the selected dive computer
|
|
*out2 = alloc_dive();
|
|
copy_dive_onedc(src, srcdc, *out2);
|
|
|
|
// This will also make fixup_dive() to allocate a new dive id...
|
|
(*out2)->id = 0;
|
|
fixup_dive(*out2);
|
|
|
|
// Copy the dive with all dive computers
|
|
*out1 = create_new_copy(src);
|
|
|
|
// .. and then delete the split-out dive computer
|
|
delete_divecomputer(*out1, num);
|
|
|
|
(*out1)->divetrip = (*out2)->divetrip = NULL;
|
|
} else {
|
|
*out1 = *out2 = NULL;
|
|
}
|
|
}
|
|
|
|
//Calculate O2 in best mix
|
|
extern "C" fraction_t best_o2(depth_t depth, const struct dive *dive, bool in_planner)
|
|
{
|
|
fraction_t fo2;
|
|
int po2 = in_planner ? prefs.bottompo2 : (int)(prefs.modpO2 * 1000.0);
|
|
|
|
fo2.permille = (po2 * 100 / depth_to_mbar(depth.mm, dive)) * 10; //use integer arithmetic to round down to nearest percent
|
|
// Don't permit >100% O2
|
|
if (fo2.permille > 1000)
|
|
fo2.permille = 1000;
|
|
return fo2;
|
|
}
|
|
|
|
//Calculate He in best mix. O2 is considered narcopic
|
|
extern "C" fraction_t best_he(depth_t depth, const struct dive *dive, bool o2narcotic, fraction_t fo2)
|
|
{
|
|
fraction_t fhe;
|
|
int pnarcotic, ambient;
|
|
pnarcotic = depth_to_mbar(prefs.bestmixend.mm, dive);
|
|
ambient = depth_to_mbar(depth.mm, dive);
|
|
if (o2narcotic) {
|
|
fhe.permille = (100 - 100 * pnarcotic / ambient) * 10; //use integer arithmetic to round up to nearest percent
|
|
} else {
|
|
fhe.permille = 1000 - fo2.permille - N2_IN_AIR * pnarcotic / ambient;
|
|
}
|
|
if (fhe.permille < 0)
|
|
fhe.permille = 0;
|
|
return fhe;
|
|
}
|
|
|
|
extern "C" void invalidate_dive_cache(struct dive *dive)
|
|
{
|
|
memset(dive->git_id, 0, 20);
|
|
}
|
|
|
|
extern "C" bool dive_cache_is_valid(const struct dive *dive)
|
|
{
|
|
static const unsigned char null_id[20] = { 0, };
|
|
return !!memcmp(dive->git_id, null_id, 20);
|
|
}
|
|
|
|
extern "C" int get_surface_pressure_in_mbar(const struct dive *dive, bool non_null)
|
|
{
|
|
int mbar = dive->surface_pressure.mbar;
|
|
if (!mbar && non_null)
|
|
mbar = SURFACE_PRESSURE;
|
|
return mbar;
|
|
}
|
|
|
|
/* This returns the conversion factor that you need to multiply
|
|
* a (relative) depth in mm to obtain a (relative) pressure in mbar.
|
|
* As everywhere in Subsurface, the expected unit of a salinity is
|
|
* g/10l such that sea water has a salinity of 10300
|
|
*/
|
|
static double salinity_to_specific_weight(int salinity)
|
|
{
|
|
return salinity * 0.981 / 100000.0;
|
|
}
|
|
|
|
/* Pa = N/m^2 - so we determine the weight (in N) of the mass of 10m
|
|
* of water (and use standard salt water at 1.03kg per liter if we don't know salinity)
|
|
* and add that to the surface pressure (or to 1013 if that's unknown) */
|
|
static double calculate_depth_to_mbarf(int depth, pressure_t surface_pressure, int salinity)
|
|
{
|
|
double specific_weight;
|
|
int mbar = surface_pressure.mbar;
|
|
|
|
if (!mbar)
|
|
mbar = SURFACE_PRESSURE;
|
|
if (!salinity)
|
|
salinity = SEAWATER_SALINITY;
|
|
if (salinity < 500)
|
|
salinity += FRESHWATER_SALINITY;
|
|
specific_weight = salinity_to_specific_weight(salinity);
|
|
return mbar + depth * specific_weight;
|
|
}
|
|
|
|
extern "C" int depth_to_mbar(int depth, const struct dive *dive)
|
|
{
|
|
return lrint(depth_to_mbarf(depth, dive));
|
|
}
|
|
|
|
extern "C" double depth_to_mbarf(int depth, const struct dive *dive)
|
|
{
|
|
// To downloaded and planned dives, use DC's values
|
|
int salinity = dive->dc.salinity;
|
|
pressure_t surface_pressure = dive->dc.surface_pressure;
|
|
|
|
if (is_manually_added_dc(&dive->dc)) { // To manual dives, salinity and pressure in another place...
|
|
surface_pressure = dive->surface_pressure;
|
|
salinity = dive->user_salinity;
|
|
}
|
|
return calculate_depth_to_mbarf(depth, surface_pressure, salinity);
|
|
}
|
|
|
|
extern "C" double depth_to_bar(int depth, const struct dive *dive)
|
|
{
|
|
return depth_to_mbar(depth, dive) / 1000.0;
|
|
}
|
|
|
|
extern "C" double depth_to_atm(int depth, const struct dive *dive)
|
|
{
|
|
return mbar_to_atm(depth_to_mbar(depth, dive));
|
|
}
|
|
|
|
/* for the inverse calculation we use just the relative pressure
|
|
* (that's the one that some dive computers like the Uemis Zurich
|
|
* provide - for the other models that do this libdivecomputer has to
|
|
* take care of this, but the Uemis we support natively */
|
|
extern "C" int rel_mbar_to_depth(int mbar, const struct dive *dive)
|
|
{
|
|
// To downloaded and planned dives, use DC's salinity. Manual dives, use user's salinity
|
|
int salinity = is_manually_added_dc(&dive->dc) ? dive->user_salinity : dive->dc.salinity;
|
|
if (!salinity)
|
|
salinity = SEAWATER_SALINITY;
|
|
|
|
/* whole mbar gives us cm precision */
|
|
double specific_weight = salinity_to_specific_weight(salinity);
|
|
return (int)lrint(mbar / specific_weight);
|
|
}
|
|
|
|
extern "C" int mbar_to_depth(int mbar, const struct dive *dive)
|
|
{
|
|
// To downloaded and planned dives, use DC's pressure. Manual dives, use user's pressure
|
|
pressure_t surface_pressure = is_manually_added_dc(&dive->dc)
|
|
? dive->surface_pressure
|
|
: dive->dc.surface_pressure;
|
|
|
|
if (!surface_pressure.mbar)
|
|
surface_pressure.mbar = SURFACE_PRESSURE;
|
|
|
|
return rel_mbar_to_depth(mbar - surface_pressure.mbar, dive);
|
|
}
|
|
|
|
/* MOD rounded to multiples of roundto mm */
|
|
extern "C" depth_t gas_mod(struct gasmix mix, pressure_t po2_limit, const struct dive *dive, int roundto)
|
|
{
|
|
depth_t rounded_depth;
|
|
|
|
double depth = (double) mbar_to_depth(po2_limit.mbar * 1000 / get_o2(mix), dive);
|
|
rounded_depth.mm = (int)lrint(depth / roundto) * roundto;
|
|
return rounded_depth;
|
|
}
|
|
|
|
/* Maximum narcotic depth rounded to multiples of roundto mm */
|
|
extern "C" depth_t gas_mnd(struct gasmix mix, depth_t end, const struct dive *dive, int roundto)
|
|
{
|
|
depth_t rounded_depth;
|
|
pressure_t ppo2n2;
|
|
ppo2n2.mbar = depth_to_mbar(end.mm, dive);
|
|
|
|
int maxambient = prefs.o2narcotic ?
|
|
(int)lrint(ppo2n2.mbar / (1 - get_he(mix) / 1000.0))
|
|
:
|
|
get_n2(mix) > 0 ?
|
|
(int)lrint(ppo2n2.mbar * N2_IN_AIR / get_n2(mix))
|
|
:
|
|
// Actually: Infinity
|
|
1000000;
|
|
rounded_depth.mm = (int)lrint(((double)mbar_to_depth(maxambient, dive)) / roundto) * roundto;
|
|
return rounded_depth;
|
|
}
|
|
|
|
extern "C" struct dive *get_dive(int nr)
|
|
{
|
|
if (nr >= divelog.dives->nr || nr < 0)
|
|
return NULL;
|
|
return divelog.dives->dives[nr];
|
|
}
|
|
|
|
extern "C" struct dive_site *get_dive_site_for_dive(const struct dive *dive)
|
|
{
|
|
return dive->dive_site;
|
|
}
|
|
|
|
extern "C" const char *get_dive_country(const struct dive *dive)
|
|
{
|
|
struct dive_site *ds = dive->dive_site;
|
|
return ds ? taxonomy_get_country(&ds->taxonomy) : NULL;
|
|
}
|
|
|
|
extern "C" const char *get_dive_location(const struct dive *dive)
|
|
{
|
|
const struct dive_site *ds = dive->dive_site;
|
|
if (ds && ds->name)
|
|
return ds->name;
|
|
return NULL;
|
|
}
|
|
|
|
extern "C" unsigned int number_of_computers(const struct dive *dive)
|
|
{
|
|
unsigned int total_number = 0;
|
|
const struct divecomputer *dc = &dive->dc;
|
|
|
|
if (!dive)
|
|
return 1;
|
|
|
|
do {
|
|
total_number++;
|
|
dc = dc->next;
|
|
} while (dc);
|
|
return total_number;
|
|
}
|
|
|
|
extern "C" struct divecomputer *get_dive_dc(struct dive *dive, int nr)
|
|
{
|
|
struct divecomputer *dc;
|
|
if (!dive)
|
|
return NULL;
|
|
dc = &dive->dc;
|
|
|
|
while (nr-- > 0) {
|
|
dc = dc->next;
|
|
if (!dc)
|
|
return &dive->dc;
|
|
}
|
|
return dc;
|
|
}
|
|
|
|
extern "C" const struct divecomputer *get_dive_dc_const(const struct dive *dive, int nr)
|
|
{
|
|
return get_dive_dc((struct dive *)dive, nr);
|
|
}
|
|
|
|
extern "C" struct dive *get_dive_by_uniq_id(int id)
|
|
{
|
|
int i;
|
|
struct dive *dive = NULL;
|
|
|
|
for_each_dive (i, dive) {
|
|
if (dive->id == id)
|
|
break;
|
|
}
|
|
#ifdef DEBUG
|
|
if (dive == NULL) {
|
|
report_info("Invalid id %x passed to get_dive_by_diveid, try to fix the code", id);
|
|
exit(1);
|
|
}
|
|
#endif
|
|
return dive;
|
|
}
|
|
|
|
extern "C" int get_idx_by_uniq_id(int id)
|
|
{
|
|
int i;
|
|
struct dive *dive = NULL;
|
|
|
|
for_each_dive (i, dive) {
|
|
if (dive->id == id)
|
|
break;
|
|
}
|
|
#ifdef DEBUG
|
|
if (dive == NULL) {
|
|
report_info("Invalid id %x passed to get_dive_by_diveid, try to fix the code", id);
|
|
exit(1);
|
|
}
|
|
#endif
|
|
return i;
|
|
}
|
|
|
|
extern "C" bool dive_site_has_gps_location(const struct dive_site *ds)
|
|
{
|
|
return ds && has_location(&ds->location);
|
|
}
|
|
|
|
extern "C" int dive_has_gps_location(const struct dive *dive)
|
|
{
|
|
if (!dive)
|
|
return false;
|
|
return dive_site_has_gps_location(dive->dive_site);
|
|
}
|
|
|
|
/* Extract GPS location of a dive computer stored in the GPS1
|
|
* or GPS2 extra data fields */
|
|
static location_t dc_get_gps_location(const struct divecomputer *dc)
|
|
{
|
|
location_t res = { };
|
|
|
|
for (struct extra_data *data = dc->extra_data; data; data = data->next) {
|
|
if (!strcmp(data->key, "GPS1")) {
|
|
parse_location(data->value, &res);
|
|
/* If we found a valid GPS1 field exit early since
|
|
* it has priority over GPS2 */
|
|
if (has_location(&res))
|
|
break;
|
|
} else if (!strcmp(data->key, "GPS2")) {
|
|
/* For GPS2 fields continue searching, as we might
|
|
* still find a GPS1 field */
|
|
parse_location(data->value, &res);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Get GPS location for a dive. Highest priority is given to the GPS1
|
|
* extra data written by libdivecomputer, as this comes from a real GPS
|
|
* device. If that doesn't exits, use the currently set dive site.
|
|
* This function is potentially slow, therefore only call sparingly
|
|
* and remember the result.
|
|
*/
|
|
extern "C" location_t dive_get_gps_location(const struct dive *d)
|
|
{
|
|
location_t res = { };
|
|
|
|
for (const struct divecomputer *dc = &d->dc; dc; dc = dc->next) {
|
|
res = dc_get_gps_location(dc);
|
|
if (has_location(&res))
|
|
return res;
|
|
}
|
|
|
|
/* No libdivecomputer generated GPS data found.
|
|
* Let's use the location of the current dive site.
|
|
*/
|
|
if (d->dive_site)
|
|
res = d->dive_site->location;
|
|
|
|
return res;
|
|
}
|
|
|
|
/* When evaluated at the time of a gasswitch, this returns the new gas */
|
|
extern "C" struct gasmix get_gasmix(const struct dive *dive, const struct divecomputer *dc, int time, const struct event **evp, struct gasmix gasmix)
|
|
{
|
|
const struct event *ev = *evp;
|
|
struct gasmix res;
|
|
|
|
/* if there is no cylinder, return air */
|
|
if (dive->cylinders.nr <= 0)
|
|
return gasmix_air;
|
|
|
|
if (!ev) {
|
|
/* on first invocation, get initial gas mix and first event (if any) */
|
|
int cyl = explicit_first_cylinder(dive, dc);
|
|
res = get_cylinder(dive, cyl)->gasmix;
|
|
ev = dc ? get_next_event(dc->events, "gaschange") : NULL;
|
|
} else {
|
|
res = gasmix;
|
|
}
|
|
|
|
while (ev && ev->time.seconds <= time) {
|
|
res = get_gasmix_from_event(dive, ev);
|
|
ev = get_next_event(ev->next, "gaschange");
|
|
}
|
|
*evp = ev;
|
|
return res;
|
|
}
|
|
|
|
/* get the gas at a certain time during the dive */
|
|
/* If there is a gasswitch at that time, it returns the new gasmix */
|
|
extern "C" struct gasmix get_gasmix_at_time(const struct dive *d, const struct divecomputer *dc, duration_t time)
|
|
{
|
|
const struct event *ev = NULL;
|
|
struct gasmix gasmix = gasmix_air;
|
|
return get_gasmix(d, dc, time.seconds, &ev, gasmix);
|
|
}
|
|
|
|
/* Does that cylinder have any pressure readings? */
|
|
extern "C" bool cylinder_with_sensor_sample(const struct dive *dive, int cylinder_id)
|
|
{
|
|
for (const struct divecomputer *dc = &dive->dc; dc; dc = dc->next) {
|
|
for (int i = 0; i < dc->samples; ++i) {
|
|
struct sample *sample = dc->sample + i;
|
|
for (int j = 0; j < MAX_SENSORS; ++j) {
|
|
if (!sample->pressure[j].mbar)
|
|
continue;
|
|
if (sample->sensor[j] == cylinder_id)
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
return false;
|
|
}
|