subsurface/dive.h

536 lines
15 KiB
C
Raw Normal View History

#ifndef DIVE_H
#define DIVE_H
#include <stdlib.h>
#include <stdint.h>
#include <time.h>
#include <glib.h>
#include <glib/gstdio.h>
#include <libxml/tree.h>
#include <openssl/sha.h>
/*
* Some silly typedefs to make our units very explicit.
*
* Also, the units are chosen so that values can be expressible as
* integers, so that we never have FP rounding issues. And they
* are small enough that converting to/from imperial units doesn't
* really matter.
*
* We also strive to make '0' a meaningless number saying "not
* initialized", since many values are things that may not have
* been reported (eg cylinder pressure or temperature from dive
* computers that don't support them). But sometimes -1 is an even
* more explicit way of saying "not there".
*
* Thus "millibar" for pressure, for example, or "millikelvin" for
* temperatures. Doing temperatures in celsius or fahrenheit would
* make for loss of precision when converting from one to the other,
* and using millikelvin is SI-like but also means that a temperature
* of '0' is clearly just a missing temperature or cylinder pressure.
*
* Also strive to use units that can not possibly be mistaken for a
* valid value in a "normal" system without conversion. If the max
* depth of a dive is '20000', you probably didn't convert from mm on
* output, or if the max depth gets reported as "0.2ft" it was either
* a really boring dive, or there was some missing input conversion,
* and a 60-ft dive got recorded as 60mm.
*
* Doing these as "structs containing value" means that we always
* have to explicitly write out those units in order to get at the
* actual value. So there is hopefully little fear of using a value
* in millikelvin as Fahrenheit by mistake.
*
* We don't actually use these all yet, so maybe they'll change, but
* I made a number of types as guidelines.
*/
typedef gint64 timestamp_t;
typedef struct {
int seconds;
} duration_t;
typedef struct {
int mm;
} depth_t;
typedef struct {
int mbar;
} pressure_t;
typedef struct {
int mkelvin;
} temperature_t;
typedef struct {
int mliter;
} volume_t;
typedef struct {
int permille;
} fraction_t;
typedef struct {
int grams;
} weight_t;
struct gasmix {
fraction_t o2;
fraction_t he;
};
typedef struct {
volume_t size;
pressure_t workingpressure;
const char *description; /* "LP85", "AL72", "AL80", "HP100+" or whatever */
} cylinder_type_t;
typedef struct {
cylinder_type_t type;
struct gasmix gasmix;
pressure_t start, end, sample_start, sample_end;
} cylinder_t;
typedef struct {
weight_t weight;
const char *description; /* "integrated", "belt", "ankle" */
} weightsystem_t;
Fix missing save of (almost empty) cylinder information If we have no explicit cylinder info at all (it's normal air, no size or working pressure information, and no beginning/end pressure information), we don't save the cylinders in question because that would be redundant. Such non-saved cylinders may still show up in the equipment list because there may be implicit mention of them elsewhere, notably due to sample data, so not saving them is the right thing to do - there is nothing to save. However, we missed one case: if there were other cylinders that *did* have explicit information in it following such an uninteresting cylinder, we do need to save the cylinder information for the useless case - if only in order to be able to save the non-useless information for subsequent cylinders. This patch does that. Now, if you had an air-filled cylinder with no information as your first cylinder, and a 51% nitrox as your second one, it will save that information as <cylinder /> <cylinder o2='51.0%' /> rather than dropping the cylinder information entirely. This bug has been there for a long time, and was hidden by the fact that normally you'd fill in cylinder descriptions etc after importing new dives. It also used to be that we saved the cylinder beginning/end pressure even if that was generated from the sample data, so if you imported from a air-integrated computer and had samples for that cylinder, we used to save it even though it was technically redundant. We stopped saving redundant air sample information in commit 0089dd8819b7 ("Don't save cylinder start/end pressures unless set by hand"). Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Removed start and end in save_cylinder_info(). These two variables are no longer used. Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-09-21 21:06:57 +00:00
extern gboolean cylinder_nodata(cylinder_t *cyl);
extern gboolean cylinder_nosamples(cylinder_t *cyl);
extern gboolean cylinder_none(void *_data);
extern gboolean no_cylinders(cylinder_t *cyl);
extern gboolean no_weightsystems(weightsystem_t *ws);
extern gboolean weightsystems_equal(weightsystem_t *ws1, weightsystem_t *ws2);
extern int get_pressure_units(unsigned int mb, const char **units);
extern double get_depth_units(unsigned int mm, int *frac, const char **units);
extern double get_volume_units(unsigned int ml, int *frac, const char **units);
extern double get_temp_units(unsigned int mk, const char **units);
extern double get_weight_units(unsigned int grams, int *frac, const char **units);
static inline double grams_to_lbs(int grams)
{
return grams / 453.6;
}
static inline int lbs_to_grams(double lbs)
{
return lbs * 453.6 + 0.5;
}
static inline double ml_to_cuft(int ml)
{
return ml / 28316.8466;
}
static inline double cuft_to_l(double cuft)
{
return cuft * 28.3168466;
}
static inline double mm_to_feet(int mm)
{
return mm * 0.00328084;
}
static inline unsigned long feet_to_mm(double feet)
{
return feet * 304.8 + 0.5;
}
static inline int to_feet(depth_t depth)
{
return mm_to_feet(depth.mm) + 0.5;
}
static inline double mkelvin_to_C(int mkelvin)
{
return (mkelvin - 273150) / 1000.0;
}
static inline double mkelvin_to_F(int mkelvin)
{
return mkelvin * 9 / 5000.0 - 459.670;
}
static inline unsigned long F_to_mkelvin(double f)
{
return (f-32) * 1000 / 1.8 + 273150.5;
}
static inline unsigned long C_to_mkelvin(double c)
{
return c * 1000 + 273150.5;
}
static inline int to_C(temperature_t temp)
{
if (!temp.mkelvin)
return 0;
return mkelvin_to_C(temp.mkelvin) + 0.5;
}
static inline int to_F(temperature_t temp)
{
if (!temp.mkelvin)
return 0;
return mkelvin_to_F(temp.mkelvin) + 0.5;
}
static inline int to_K(temperature_t temp)
{
if (!temp.mkelvin)
return 0;
return (temp.mkelvin + 499)/1000;
}
static inline double psi_to_bar(double psi)
{
return psi / 14.5037738;
}
static inline unsigned long psi_to_mbar(double psi)
{
return psi_to_bar(psi)*1000 + 0.5;
}
static inline int to_PSI(pressure_t pressure)
{
return pressure.mbar * 0.0145037738 + 0.5;
}
static inline double bar_to_atm(double bar)
{
return bar / 1.01325;
}
static inline double to_ATM(pressure_t pressure)
{
return pressure.mbar / 1013.25;
}
static inline int mbar_to_PSI(int mbar)
{
pressure_t p = {mbar};
return to_PSI(p);
}
struct sample {
duration_t time;
depth_t depth;
temperature_t temperature;
pressure_t cylinderpressure;
int cylinderindex;
duration_t ndl;
duration_t stoptime;
depth_t stopdepth;
};
/*
* Events are currently pretty meaningless. This is
* just based on the random data that libdivecomputer
* gives us. I'm not sure what a real "architected"
* event model would actually look like, but right
* now you can associate a list of events with a dive,
* and we'll do something about it.
*/
struct event {
struct event *next;
duration_t time;
int type, flags, value;
gboolean deleted;
char name[];
};
/*
* NOTE! The deviceid and diveid are model-specific *hashes* of
* whatever device identification that model may have. Different
* dive computers will have different identifying data, it could
* be a firmware number or a serial ID (in either string or in
* numeric format), and we do not care.
*
* The only thing we care about is that subsurface will hash
* that information the same way. So then you can check the ID
* of a dive computer by comparing the hashes for equality.
*
* A deviceid or diveid of zero is assumed to be "no ID".
*/
struct divecomputer {
timestamp_t when;
const char *model;
uint32_t deviceid, diveid;
int samples, alloc_samples;
struct sample *sample;
struct event *events;
struct divecomputer *next;
};
#define MAX_CYLINDERS (8)
#define MAX_WEIGHTSYSTEMS (4)
#define W_IDX_PRIMARY 0
#define W_IDX_SECONDARY 1
typedef enum { TF_NONE, NO_TRIP, IN_TRIP, ASSIGNED_TRIP, NUM_TRIPFLAGS } tripflag_t;
First cut of explicit trip tracking This code establishes the explicit trip data structures and loads and saves them in the XML data. No attempts are made to edit / modify the trips, yet. Loading XML files without trip data creates the trips based on timing as before. Saving out the same, unmodified data will create 'trip' entries in the XML file with a 'number' that reflects the number of dives in that trip. The trip tag also stores the beginning time of the first dive in the trip and the location of the trip (which we display in the summary entries in the UI). The logic allows for dives that aren't part of a dive trip. All other dives simply belong to the "previous" dive trip - i.e. the dive trip with the latest start time that is earlier or equal to the start time of this dive. This logic significantly simplifies the tracking of trips compared to other approaches that I have tried. The automatic grouping into trips now is an option that defaults to off (as it makes changes to the XML file - and people who don't want this feature shouldn't have trips added to their XML files that they then need to manually remove). For now you have to select this option, then exit the program and start it again. Still to do is to trigger the trip generation at run time. We also need a way to mark dives as not part of trips and to allow options to combine trips, split trips, edit trip location data, etc. The code has only had some limited testing when opening multiple files. The code is known to fail if a location name contains unquoted special characters like an "'". This commit also fixes a visual inconsistency in the preferences dialog where the font selector button didn't have a frame around it that told you what this option was about. Inspired-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-08-22 05:04:24 +00:00
typedef struct dive_trip {
timestamp_t when;
char *location;
char *notes;
struct dive *dives;
2012-11-10 18:51:03 +00:00
int nrdives;
unsigned expanded:1, selected:1, autogen:1;
struct dive_trip *next;
} dive_trip_t;
/* List of dive trips (sorted by date) */
extern dive_trip_t *dive_trip_list;
struct dive {
int number;
First cut of explicit trip tracking This code establishes the explicit trip data structures and loads and saves them in the XML data. No attempts are made to edit / modify the trips, yet. Loading XML files without trip data creates the trips based on timing as before. Saving out the same, unmodified data will create 'trip' entries in the XML file with a 'number' that reflects the number of dives in that trip. The trip tag also stores the beginning time of the first dive in the trip and the location of the trip (which we display in the summary entries in the UI). The logic allows for dives that aren't part of a dive trip. All other dives simply belong to the "previous" dive trip - i.e. the dive trip with the latest start time that is earlier or equal to the start time of this dive. This logic significantly simplifies the tracking of trips compared to other approaches that I have tried. The automatic grouping into trips now is an option that defaults to off (as it makes changes to the XML file - and people who don't want this feature shouldn't have trips added to their XML files that they then need to manually remove). For now you have to select this option, then exit the program and start it again. Still to do is to trigger the trip generation at run time. We also need a way to mark dives as not part of trips and to allow options to combine trips, split trips, edit trip location data, etc. The code has only had some limited testing when opening multiple files. The code is known to fail if a location name contains unquoted special characters like an "'". This commit also fixes a visual inconsistency in the preferences dialog where the font selector button didn't have a frame around it that told you what this option was about. Inspired-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-08-22 05:04:24 +00:00
tripflag_t tripflag;
dive_trip_t *divetrip;
struct dive *next, **pprev;
int selected;
gboolean downloaded;
int start, end;
timestamp_t when;
char *location;
char *notes;
char *divemaster, *buddy;
int rating;
double latitude, longitude;
depth_t maxdepth, meandepth;
int salinity; // kg per 10000 l
duration_t duration, surfacetime;
int visibility; /* 0 - 5 star rating */
temperature_t airtemp, watertemp;
pressure_t surface_pressure;
cylinder_t cylinder[MAX_CYLINDERS];
weightsystem_t weightsystem[MAX_WEIGHTSYSTEMS];
char *suit;
int sac, otu;
/* Eventually we'll do multiple dive computers */
struct divecomputer dc;
};
/* 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 inline int depth_to_mbar(int depth, struct dive *dive)
{
double specific_weight = 1.03 * 0.981;
int surface_pressure = 1013;
if (dive->salinity)
specific_weight = dive->salinity / 10000.0 * 0.981;
if (dive->surface_pressure.mbar)
surface_pressure = dive->surface_pressure.mbar;
return depth / 10.0 * specific_weight + surface_pressure + 0.5;
}
/* 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 */
static inline int rel_mbar_to_depth(int mbar, struct dive *dive)
{
int cm;
double specific_weight = 1.03 * 0.981;
if (dive->salinity)
specific_weight = dive->salinity / 10000.0 * 0.981;
/* whole mbar gives us cm precision */
cm = mbar / specific_weight + 0.5;
return cm * 10;
}
#define SURFACE_THRESHOLD 750 /* somewhat arbitrary: only below 75cm is it really diving */
/* this is a global spot for a temporary dive structure that we use to
* be able to edit a dive without unintended side effects */
extern struct dive edit_dive;
First cut of explicit trip tracking This code establishes the explicit trip data structures and loads and saves them in the XML data. No attempts are made to edit / modify the trips, yet. Loading XML files without trip data creates the trips based on timing as before. Saving out the same, unmodified data will create 'trip' entries in the XML file with a 'number' that reflects the number of dives in that trip. The trip tag also stores the beginning time of the first dive in the trip and the location of the trip (which we display in the summary entries in the UI). The logic allows for dives that aren't part of a dive trip. All other dives simply belong to the "previous" dive trip - i.e. the dive trip with the latest start time that is earlier or equal to the start time of this dive. This logic significantly simplifies the tracking of trips compared to other approaches that I have tried. The automatic grouping into trips now is an option that defaults to off (as it makes changes to the XML file - and people who don't want this feature shouldn't have trips added to their XML files that they then need to manually remove). For now you have to select this option, then exit the program and start it again. Still to do is to trigger the trip generation at run time. We also need a way to mark dives as not part of trips and to allow options to combine trips, split trips, edit trip location data, etc. The code has only had some limited testing when opening multiple files. The code is known to fail if a location name contains unquoted special characters like an "'". This commit also fixes a visual inconsistency in the preferences dialog where the font selector button didn't have a frame around it that told you what this option was about. Inspired-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-08-22 05:04:24 +00:00
extern gboolean autogroup;
/* random threashold: three days without diving -> new trip
* this works very well for people who usually dive as part of a trip and don't
* regularly dive at a local facility; this is why trips are an optional feature */
#define TRIP_THRESHOLD 3600*24*3
#define UNGROUPED_DIVE(_dive) ((_dive)->tripflag == NO_TRIP)
#define DIVE_IN_TRIP(_dive) ((_dive)->tripflag == IN_TRIP || (_dive)->tripflag == ASSIGNED_TRIP)
#define DIVE_NEEDS_TRIP(_dive) ((_dive)->tripflag == TF_NONE)
First cut of explicit trip tracking This code establishes the explicit trip data structures and loads and saves them in the XML data. No attempts are made to edit / modify the trips, yet. Loading XML files without trip data creates the trips based on timing as before. Saving out the same, unmodified data will create 'trip' entries in the XML file with a 'number' that reflects the number of dives in that trip. The trip tag also stores the beginning time of the first dive in the trip and the location of the trip (which we display in the summary entries in the UI). The logic allows for dives that aren't part of a dive trip. All other dives simply belong to the "previous" dive trip - i.e. the dive trip with the latest start time that is earlier or equal to the start time of this dive. This logic significantly simplifies the tracking of trips compared to other approaches that I have tried. The automatic grouping into trips now is an option that defaults to off (as it makes changes to the XML file - and people who don't want this feature shouldn't have trips added to their XML files that they then need to manually remove). For now you have to select this option, then exit the program and start it again. Still to do is to trigger the trip generation at run time. We also need a way to mark dives as not part of trips and to allow options to combine trips, split trips, edit trip location data, etc. The code has only had some limited testing when opening multiple files. The code is known to fail if a location name contains unquoted special characters like an "'". This commit also fixes a visual inconsistency in the preferences dialog where the font selector button didn't have a frame around it that told you what this option was about. Inspired-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-08-22 05:04:24 +00:00
2012-11-10 18:51:03 +00:00
extern void add_dive_to_trip(struct dive *, dive_trip_t *);
extern void remove_dive_from_trip(struct dive *);
extern void delete_single_dive(int idx);
extern void add_single_dive(int idx, struct dive *dive);
extern void insert_trip(dive_trip_t **trip);
First cut of explicit trip tracking This code establishes the explicit trip data structures and loads and saves them in the XML data. No attempts are made to edit / modify the trips, yet. Loading XML files without trip data creates the trips based on timing as before. Saving out the same, unmodified data will create 'trip' entries in the XML file with a 'number' that reflects the number of dives in that trip. The trip tag also stores the beginning time of the first dive in the trip and the location of the trip (which we display in the summary entries in the UI). The logic allows for dives that aren't part of a dive trip. All other dives simply belong to the "previous" dive trip - i.e. the dive trip with the latest start time that is earlier or equal to the start time of this dive. This logic significantly simplifies the tracking of trips compared to other approaches that I have tried. The automatic grouping into trips now is an option that defaults to off (as it makes changes to the XML file - and people who don't want this feature shouldn't have trips added to their XML files that they then need to manually remove). For now you have to select this option, then exit the program and start it again. Still to do is to trigger the trip generation at run time. We also need a way to mark dives as not part of trips and to allow options to combine trips, split trips, edit trip location data, etc. The code has only had some limited testing when opening multiple files. The code is known to fail if a location name contains unquoted special characters like an "'". This commit also fixes a visual inconsistency in the preferences dialog where the font selector button didn't have a frame around it that told you what this option was about. Inspired-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
2012-08-22 05:04:24 +00:00
/*
* We keep our internal data in well-specified units, but
* the input and output may come in some random format. This
* keeps track of those units.
*/
/* turns out in Win32 PASCAL is defined as a calling convention */
#ifdef WIN32
#undef PASCAL
#endif
struct units {
enum { METERS, FEET } length;
enum { LITER, CUFT } volume;
enum { BAR, PSI, PASCAL } pressure;
enum { CELSIUS, FAHRENHEIT, KELVIN } temperature;
enum { KG, LBS } weight;
};
extern const struct units SI_units, IMPERIAL_units;
extern struct units input_units, output_units;
extern int verbose;
struct dive_table {
int nr, allocated, preexisting;
struct dive **dives;
};
extern struct dive_table dive_table;
extern int selected_dive;
#define current_dive (get_dive(selected_dive))
static inline struct dive *get_dive(int nr)
{
if (nr >= dive_table.nr || nr < 0)
return NULL;
return dive_table.dives[nr];
}
static inline struct dive *get_dive_by_diveid(int diveid, int deviceid)
{
int i;
for (i = 0; i < dive_table.nr; i++)
if (dive_table.dives[i]->dc.diveid == diveid &&
dive_table.dives[i]->dc.deviceid == deviceid)
return dive_table.dives[i];
return NULL;
}
/*
* Iterate over each dive, with the first parameter being the index
* iterator variable, and the second one being the dive one.
*
* I don't think anybody really wants the index, and we could make
* it local to the for-loop, but that would make us requires C99.
*/
#define for_each_dive(_i,_x) \
for ((_i) = 0; ((_x) = get_dive(_i)) != NULL; (_i)++)
extern void parse_xml_init(void);
extern void parse_xml_buffer(const char *url, const char *buf, int size, GError **error);
extern void parse_xml_exit(void);
extern void set_filename(const char *filename, gboolean force);
extern void parse_file(const char *filename, GError **error, gboolean possible_default_filename);
#ifdef XSLT
extern xmlDoc *test_xslt_transforms(xmlDoc *doc);
#endif
extern void show_dive_info(struct dive *);
extern void show_dive_equipment(struct dive *, int w_idx);
extern void clear_equipment_widgets(void);
extern void show_dive_stats(struct dive *);
extern void clear_stats_widgets(void);
extern void show_yearly_stats(void);
extern void update_dive(struct dive *new_dive);
extern void save_dives(const char *filename);
extern timestamp_t utc_mktime(struct tm *tm);
extern void utc_mkdate(timestamp_t, struct tm *tm);
extern struct dive *alloc_dive(void);
extern void record_dive(struct dive *dive);
extern void delete_dive(struct dive *dive);
extern struct sample *prepare_sample(struct divecomputer *dc);
extern void finish_sample(struct divecomputer *dc);
extern void report_dives(gboolean imported, gboolean prefer_imported);
extern struct dive *fixup_dive(struct dive *dive);
extern struct dive *merge_dives(struct dive *a, struct dive *b, int offset, gboolean prefer_downloaded);
extern struct dive *try_to_merge(struct dive *a, struct dive *b, gboolean prefer_downloaded);
extern void renumber_dives(int nr);
extern void add_event(struct divecomputer *dc, int time, int type, int flags, int value, const char *name);
/* UI related protopypes */
extern void init_ui(int *argcp, char ***argvp);
extern void run_ui(void);
extern void exit_ui(void);
extern void report_error(GError* error);
extern void add_cylinder_description(cylinder_type_t *);
extern void add_weightsystem_description(weightsystem_t *);
extern void add_people(const char *string);
extern void add_location(const char *string);
extern void add_suit(const char *string);
extern void remember_event(const char *eventname);
extern void evn_foreach(void (*callback)(const char *, int *, void *), void *data);
extern int add_new_dive(struct dive *dive);
extern gboolean edit_trip(dive_trip_t *trip);
extern int edit_dive_info(struct dive *dive, gboolean newdive);
extern int edit_multi_dive_info(struct dive *single_dive);
extern void dive_list_update_dives(void);
extern void flush_divelist(struct dive *dive);
#define DIVE_ERROR_PARSE 1
const char *weekday(int wday);
const char *monthname(int mon);
#define UTF8_DEGREE "\xc2\xb0"
#define UTF8_SUBSCRIPT_2 "\xe2\x82\x82"
#define UTF8_WHITESTAR "\xe2\x98\x86"
#define UTF8_BLACKSTAR "\xe2\x98\x85"
#define ZERO_STARS UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR
#define ONE_STARS UTF8_BLACKSTAR UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR
#define TWO_STARS UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_WHITESTAR UTF8_WHITESTAR UTF8_WHITESTAR
#define THREE_STARS UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_WHITESTAR UTF8_WHITESTAR
#define FOUR_STARS UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_WHITESTAR
#define FIVE_STARS UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR UTF8_BLACKSTAR
extern const char *star_strings[];
/* enum holding list of OS features */
typedef enum {
UTF8_FONT_WITH_STARS
} os_feature_t;
extern const char *default_filename;
extern const char *existing_filename;
extern const char *subsurface_default_filename(void);
extern const char *subsurface_gettext_domainpath(char *);
extern gboolean subsurface_os_feature_available(os_feature_t);
extern void subsurface_command_line_init(gint *, gchar ***);
extern void subsurface_command_line_exit(gint *, gchar ***);
#define AIR_PERMILLE 209
#define FRACTION(n,x) ((unsigned)(n)/(x)),((unsigned)(n)%(x))
#ifdef DEBUGFILE
extern char *debugfilename;
extern FILE *debugfile;
#endif
#endif /* DIVE_H */