The return value of subsurface_get_conf is a const void * - so we shouldn't
just assign it to a variable where we know it will be changed. Instead we
duplicate the string and free the original one. A little less efficient but
cleaner.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
When updating the "You have usnaved change..." string I didn't pay
attention and missed the fact that we used it's hard coded length above to
allocate a large enough buffer.
Thanks you, Valgrind.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This is the same issue as 882cb159a4 (although now we
don't have "Import XML Files" within a dialog). It applies when
in the "Dive info" dialog.
There is some sort of a GTK bug on Ubuntu 12.04 with GTK 2.24.10 that
prevents us from using the gtk_window_set_accept_focus() and similar
API to make the window behind fully inactive.
The proposed portable solution is to completely disable the background
window (NOTE: unless its the main window), disabling child controls
(gtk_widget_set_sensitive) and making the top window "transient for" or
putting it on top (gtk_window_set_transient_for).
Still we do not want to hide the background window titlebar with
gtk_window_set_decorated(), which makes it still clickable.
Make this change to older code in gtk-gui.c as well.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Debian and Ubuntu's distributed Gtk decide to make the window
tiny (completely hiding the Notes section). To prevent that we "put"
a default size of 400x300px to the window, which is proportionally OK
compared to the main window's default size.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
After calling dive_list_update_dives() in delete_selected_dives_cb(),
if the selection length is zero, we can clear the display widgets
not to show information of a deleted dive.
[Dirk Hohndel: please watch your whitespace - you once again added a bunch
of empty lines that really didn't help the code...
I removed them]
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
otherwise the filedescriptor keeps open which prevents a
smooth unmounting as long as subsurface is open
Signed-off-by: Martin Gysel <me@bearsh.org>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The existing code (both my first single dive delete and then Lubomir's
multi dive delete code) had way too many issues and was just painfully
inefficient.
This new code takes a radically different approach and mostly ignores the
Gtk tree model (as that gets recreated after a delete, anyway) and instead
is linear time on the number of dives in the list. It does do its best to
maintain the existing selection and the expand state of tree model (the
latter isn't possible if we have switched to the list model).
Many thanks to "Lubomir I. Ivanov" <neolit123@gmail.com> for his work on
this - this commit actually contains a few lines out of one of the patches
that he wrote.
Reported-by: "Lubomir I. Ivanov" <neolit123@gmail.com>
Tested-by: "Lubomir I. Ivanov" <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Running under Valgrind showed a couple of silly bugs.
Worse, intentionally running into various error scenarios showed that we
could get the buffer handling in the raw parsing code to break down - we
would fail to process the correctly downloaded files.
To make it easier to get this right I restructured the code to collect the
XML buffer in a different way - this works much better and has stood up
well under testing so far.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Instead of trying to figure out in the GUI code whether to call the
downloader again, the logic was moved into the downloader itself. It now
attempts to deal cleverly with running out of space on the dive computer
filesystem - and in return is able to process the maximum number of dives
(instead of just ten or so at a time).
Even on partial reads before a failure we are able to collect the data
that was completely transferred and report those dives.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This brings in the code to download dive information directly from a Uemis
Zurich dive computer.
The implementation contains a major hack that hooks the uemis code into
the same data structures used to setup libdivecomputer. This gives the
best result for the user, but is not something that I like as a long term
solution as it relies on internal libdivecomputer data structures.
I guess no one has ever tried to import Uemis dive data under Windows.
The glib-2 libraries for Windows (at least the ones that are part of the
mingw package, but my guess is this is true for all of them), force the
whole program to be compiled with Windows packing rules for structures.
That broke the structure we use for decoding Uemis binary data.
This commit changes the data structure to no longer use unaligned 16bit
values but instead two 8bit values and assemble them in the actual code.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This includes one major hack that uses a private data structure from
libdivecomputer to allow us to show the Uemis Zurich as one computer the
user can import from.
Once the user has chosen the Uemis we don't use libdivecomputer but our
own downloader. Just like in the libdicecomputer case this runs in its own
thread and updates the import dialog with progress information.
The code also keeps track of the last dive that has been downloaded from a
Uemis computer so we only import new dives on subsequent downloads. And
since the Uemis Zurich gives us its device id, we make this a "per
divecomputer" property for people who dive with multiple Uemis Zurich
computers.
This uses the debugfile infrastructure to allow easily collecting
debugging output - especially on Windows where by default console output
is lost.
Known limitations: when the Uemis runs out of space (it uses its
filesystem for communication with the host computer) we have no graceful
way to reset things. This is why the code doesn't try to download ALL
dives on the computer but instead download them in increments of ten
dives. This clearly needs to be addressed once I understand how to reset
the device.
The Cancel button of the import dialog isn't correctly hooked up, yet.
I still need to figure out how to gracefully shut down a download without
potentially hanging the device.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Especially when asking non-developers for help debugging a problem it can
be extremely useful to have debugging output not go to the console but to
a log file instead.
This just adds the infrastructure to create (and close) such a file. No
changes to the debug output are made.
All this is of course #ifdef'ed out.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This brings in Lubomir's latest version of the code. This allows deleting
multiple selected dives from the divelist.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The "cylinders_equal()/copy_cylinders()" functions were buggered, and
only checked (and copied) the cylinder type. That was on purpose, since
you do want to be able to change the type of a cylinder without changing
the gasmix of the cylinder.
HOWEVER, the reverse is also true: you may want to change the gasmix of
a cylinder without changing the type.
So it's not that the type of the cylinder is special - it's that the
type and the gasmix should be considered separately.
Do that properly for the equipment editing case.
Reported-by: Ďoďo <dodo.sk@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Both gtk_tree_selection_selected_foreach() and
gtk_tree_selection_get_selected_rows() are problematic.
gtk_tree_selection_get_selected_rows is not compatible with older GTK,
while gtk_tree_selection_selected_foreach() should not be used to
modify the tree. A workaround to is allocate memory and store what
is returned from the gtk_tree_selection_selected_foreach() callback
function as a GtkTreeIter array. Once done iterate trought the array
and pass the values to delete_single_dive().
A bit excesive, but it is not certain how safe is modifying the tree
while in the "_foreach" loop, even if it only shows a warning.
On the other hand the GTK source shows gtk_tree_selection_get_selected_rows()
to be a rather complicated and slow method.
Inside delete_single_dive(), once a dive is no longer part of "dive_table"
and if the dive was part of a trip, remove the dive from the tree
(gtk_tree_store_remove()) and call update_trip_timestamp().
The struct type "tree_selected_st" and tree_selected_foreach() are
reusable.
Reported-by: Dirk Hohndel <dirk@hohndel.org>
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
gtk-gui.c:about_dialog():
Before assigning a pixel buffer to "logo" check if "image"
has a GTK_IMAGE_PIXBUF storage type.
Prevents the following message if the icon file is missing:
Gtk-CRITICAL **: gtk_image_get_pixbuf: assertion
The message that the file is missing is still shown.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
When calling gtk-gui.c:set_default_dive_computer_device(), first
free memory at default_dive_computer_device and then use strdup().
This fixed a case, where after selecting a device for the first time
in the "Download" dialog, then pressing "OK" followed by "Cancel"
(after a device error) the device text is garbled the next time the
dialog is opened.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Moved portion of the code from delete_dive_cb() to a function called
delete_single_dive(), that directly accepts a GtkTreeIter pointer.
Added the function delete_selected_dives_cb(), which is called
when calling "Delete dives" from the combo box for the selected dives.
The above function iterates trought the selection calling
delete_selected_foreach(), which on its own calls delete_single_dive().
The "for-each" API in this case looks much prettier C code wise,
however we do potentially create an extra jump and also
do not have anything but the redirection:
delete_selected_foreach() -> delete_single_dive()
Probably slighly slower than using gtk_tree_selection_get_selected_rows(),
performance wise, but less C code.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
When clicking multiple dives in the list, check if more than one
are selected and if so show the text "Delete dives".
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
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 0089dd8819
("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>
Renamed the function pick_import_files() to import_files()
and added a declaration in "display-gtk.h". The declaration is "extern"
for consistency with download_dialog().
Since the new dialog is a real child off "main_window", we no longer need
some of the calls to gtk_widget_set_sensitive(), gtk_window_set_decorated()
to disable the parent, which previously was the old "Import" combined dialog.
Once the file list is retrieved, attempt to import each file and update
the list via report_dives().
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Minor changes to file select box title
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Renamed "import_dialog" to "download_dialog" and also focused the download
dialog strictly on dive computer functionality. Removed the file button
and "filenames" checks from "download_dialog".
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
New entries with icons and shortcut keys in the menus.
"Import XML File" is now using CTRL + I.
Reserve CTRL + E for "Export XML File" if possible.
"Download From Dive Computer" is using CTRL + D and
a GTK_STOCK_GO_DOWN icon.
Added separator below this entry in the "Log" menu.
Note: menu entry callbacks are NULL at this stage.
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
There are possible bugs when targeting GTK 2.24.10.
When opening the "import XML files" file selector from the "import" dialog
on Ubuntu 12.04, even with gtk_window_set_accept_focus() setting FALSE
to the parent ("import"), it is still possible to click the title bar
and buttons, causing actions. This can trigger undefined behaviour -
e.g. opening multiple "import XML files" dialogs.
While on Windows the situation is more acceptable, on Ubuntu it is also
possible to click the [x] button on the title bar of the parent window,
which will close the parent dialog _with_ the current forward dialog.
Multiple proposed methods in the GTK documentation seem not to function
as expected, such as: gtk_window_set_modal() and gtk_window_set_accept_focus().
Passing the correct parent on dialog creation doesn't seems to help either.
The same problem occurs when selecting a default file from the preferences
dialog.
The proposed portable solution in this patch is to completely disable
the background windows by hiding the titlebar (gtk_window_set_decorated),
disabling child controls (gtk_widget_set_sensitive) and making the window
"transient for" or putting it on top (gtk_window_set_transient_for).
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Renamed the XML button and titles slightly.
Generalized the dialog topmost text to "Choose what to import:".
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
After deleting a dive the dive list is recreated. If there are still dives
selected we should select the last dive as well. If there isn't any dive
selected, then the last dive is as good a default as any, I guess.
Reported-by: "Lubomir I. Ivanov" <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
When switching the dive trips to be stored in a different data structure I
forgot to update the code in save_trip() - and since we were passing the
pointer around via a gpointer the compiler didn't catch this. Oops.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
There's no guarantee that gtk is thread-safe (apparently it can be
broken at least on Windows, even though it should be fine on top of X).
So don't update the progress bar directly from the dive computer import
code, instead just update the progress information in static variables,
and let the GUI thread update it while it does the idle loop polling
anyway.
Reported-by: Jef Driesen <jefdriesen@telenet.be>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
One of the "+1900" cases got incorrectly dropped.
Signed-off-by: Miika Turkia <miika.turkia@gmail.com>
[ I really wanted the utc_mkdate/mktime functions to use proper years,
rather than the idiotic "offset by 1900". However, we still mix
things with "localtime()" in one place, so for now we have to be
compatible with the insane tm_year semantics. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Merge the dive trip rewrite by Dirk Hohndel.
This just merges the dive trip changes with the timestamp handling
changes. There were multiple small data conflicts, along with some
newly added 'time_t' cases in the dive trip handling that needed to be
converted to 'timestamp_t' along the way.
* 'divetrip-rewrite' of git://github.com/torvalds/subsurface:
Convert FIND_TRIP into function
Partial rewrite of the dive trip code
Check if trip is NULL before calling DIVE_TRIP
Merge the 64-bit timestamp_t time function branch.
This makes subsurface not only safe against the 2038-year problem, but
also avoids the use of thread-unsafe gmtime() etc.
We still use the system time_t for initializing the calendar widget for
adding a new dive, but that's cosmetic rather than anything fundamental.
* time-function:
FIND_TRIP: don't cast a timestamp to a pointer
dive-time widget: fix incorrect use of timestamp_t
Fix the incorrect data type for DIVE_DATE accesses
Use a 64-bit 'timestamp_t' for all timestamps, rather than 'time_t'
The pointer size may not be large enough to contain a timestamp, so make
FIND_TRIP() just pass the pointer to the timestamp instead.
And use an inline function instead of macros with casts. That gets us
proper type safety while at it, so that we get a warning if somebody
doesn't pass the expected "timestamp_t *". Plus the code actually looks
simpler and way more straightforward.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I did a global search-and-replace to make all "time_t" users use the
internal subsurface 64-bit "timestamp_t" type instead, but we have one
case that still uses the system time functions: the use of "localtime()"
in the dive_time_widget().
Everywhere else we always just use UTC for all our time handling, and we
don't really ever care about the local timezone etc. However, for the
dive time widget, we initialize the calendar widget to the current time,
which obviously does want to take the local timezone into account, so
there we end up using the whole system time handling code.
So that one should continue to use time_t, even if it might have the
year-2038 problem. We also don't care about the fact that it's not
thread-safe, since this is just initializing the widget which definitely
doesn't happen threaded.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This helps us deal with the issue that the g_list convenience functions
don't allow us to easily compare 64bit values on 32bit architectures. And
since these convenience functions are truly trivial in nature, it seemed
easier to simply implement our own logic here.
In the process I moved all the dive_trip_list helper functions into the
same spot in divelist.c
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
This is the same bugfix that Lubomir did in the master branch, but now
on top of the new 64-bit timestamp_t model. So now we also remove the
comment about the year 2038 problem, because it's not true any more. We
do all the date handling in a 64-bit integer.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This introduces a new data structure for dive trips - reuseing the struct
dive just got way too messy.
The dive_trip_t datastructure now allows the code to remember if the trip
was auto generated or if its time stamp changed when dives where added to
the trip during auto generation.
The algorithm also distinguishes between dives that were intentionally
added to a trip (either in an XML file or by adding them to trip in the
UI) and dives that were added to trips via autogen. Saving dives that were
added to trips via autogen makes that assignment "intentional".
With this partial rewrite several of the oddities of the old code should
be resolved - especially turning autogen on and off again should get the
divelist back to the previous stage.
Also, when dives are merged during file open or import we now try to pick
the correct tripflag (instead of just ignoring the tripflag completely and
resetting it to TF_NONE by mistake).
Finally, the dive trip debugging code got more verbose and is trying
harder to detect issues at the earliest time possible.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The type of the DIVE_DATE field is G_TYPE_LONG, so when accessing it we
have to pass gtk_tree_model_get() the proper "long *" pointer, rather
than an int. On 64-bit platforms, passing a pointer to an int would
otherwise result in randomly overwriting another four bytes of stack.
Sadly, there are not much safety checks to warn if our passed variable
to the vararg list is of the correct type, which is unlike
gtk_tree_model_get_value(), which will report typed errors.
This solves a bug that may potentially result in undefined behaviour
on some x64 OS (e.g Ubuntu 12.04 x64).
Reported-by: Miika Turkia <miika.turkia@gmail.com>
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This makes the time type unambiguous, and we can use G_TYPE_INT64 for it
in the divelist too.
It also implements a portable (and thread-safe) "utc_mkdate()" function
that acts kind of like gmtime_r(), but using the 64-bit timestamp_t. It
matches our original "utc_mktime()".
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Other places have this check, but for this particular one a crash
can be reproduced:
./subsurface ./dives/*
log -> autogroup
log -> autogroup
Against d14932058f
Signed-off-by: Lubomir I. Ivanov <neolit123@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Thanks to Christian for running the static code analysis tool against
subsurface...
There were some false positives, a few style issues that I'll ignore for
now, and two actual potential bugs.
First: Don't check unsigned variables for < 0
This has been around for a while and we are lucky that while technically a
bug it still works as expected. Passing a negative idx simply turns it
into a very large unsigned integer which then fails the > dive_table.nr
test. So it still gets a NULL returned. A bug? Yes. Critical? No.
Mismatched allocation and free
This is an actual bug that potentially could cause issues. We allocate
memory with malloc and free it with g_free. Not good.
Reported-by: Cristian Ionescu-Idbohrn <cristian.ionescu-idbohrn@axis.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
