This was the umpteenth inefficient reinvention of a trivial
map. Replace by a hash-map (std::unordered_map). Might just
as well use a balanced binary tree or a sorted array. In the
end, it probably doesn't matter at all.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
uemis_get_divenr() returns maxdiveid and passes mindiveid as a
global variable.
Make this more reasonable by returning a min, max pair.
The way mindiveid is an unsigned int and then reinterpreted as
int is very sketchy. This commit attempts to not change that
behavior.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This is only initialized and used in one loop. Very mysterious
why this should be a global variable.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This global variable is used in two function independently of
each other. I don't see how there should be transport of this
value from one function to the other. Ominous.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
uemis_get_answer() would put the raw response into a global variable.
This could be anywhere in the call stack and thus you never knew
when the existing buffer was removed under your feet.
Instead, return the buffer explicitly from uemis_get_answer().
I'm nit perfectly happy about the new interface: an error is
indicated by an empty buffer, which is awkward to test for.
If an empty buffer turns out to be a valid response, this
should be replaced by an std::optional<> or std::expected<>.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The string code of uemis-downloader.cpp was broken in more ways
than can be listed here. Notably, it brazenly refused to free any
memory allocated for the parameters buffer.
Using std::string and std::string_view should plug all those
memory holes. That made it necessary to do some major refactoring.
This was done blind and therefore will break.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
newmax was an integer variable kept as a string. Very ominous.
Moreover, memory management seems to be broken:
1) The string is never freed.
2) The string is passed as value from do_uemis_import() to
get_matching_dives(), which passes it as reference to
process_raw_buffer(), which may reallocate it, which means
that do_uemis_import() now possesses a pointer to a free()d
string.
Simplify all that by making newmax an integer variable and
passing it as a reference from do_uemis_import() to
get_matching_dives().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
There were a number of fprintf()s that escaped the conversion
to report_info(), because they used "debugfile" instead of
"stderr" as target. However, debugfile was just #defined to
be stderr, so we might just use report_info() for consistency.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The uemis code is wild. It simply doesn't deallocate memory
and uses global variables. To get this under control, create
a "struct uemis" and make the functions exported by "uemis.h"
members of "struct uemis". Thus, we don't have to carry around
a parameter for the state of the importing process.
Turn a linked list of "helper" structures (one per imported dive)
into a std::unordered_map, to fix leaking of the helper structures.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
This was very obscure: the function that parses into a struct
dive was passed a void-pointer instead of a struct dive-pointer.
Why? Just pass the correct type.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The SHA1() helper function was only used when calculating a
SHA1 hash and taking the first four bytes of it as uint32.
Make that explicit by renaming the function into SHA1_uint32()
and directly returning an uint32_t.
Note that the usage in cochran.cpp is sketchy: it generates
a four-byte hash out of two-byte data. Why!?
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The strtod_flags() function allowed for fine control of how to
parse strings. However, only two different modes were actually
used: ascii mode ("C" locale) and permissive mode (accept ","
and "." as decimal separator).
The former had already its own function name (ascii_strtod).
Make the latter a separatge function as well (permissive_strtod)
and remove all the flags rigmarole.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Owing to bitrot, this union only contained a float and therefore
is pointless. Let's remove it.
That makes the function name "integer_or_float()" non-sensical.
Call it "parse_float()" instead.
Moreover, change the output-arguments of "parse_float()" from
pointers to references, as null-pointers are not supported.
Finally, remove the "errno" check after "ascii_strtod()". As far as
I can tell, errno is not set in "ascii_strtod()" and using a global
variable for error-reporting it is an incredibly silly interface
anyway.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Use the version information detected in the manage-version action, as
the actual build is happening in the speculative post-merge branch,
resulting in a bogus 'commit distance' number.
Signed-off-by: Michael Keller <github@ike.ch>
Make the version information used throughout the application and build
process compliant with the SemVer specifications, so that it can be
processed with libraries.
Signed-off-by: Michael Keller <github@ike.ch>
Remove pointers to the developer mailing list from the README and
CONTRIBUTING pages, as it is no longer possible to join it due to
excessive brute force breakin attempts.
Add pointers to the user forum instead.
Signed-off-by: Michael Keller <mikeller@042.ch>
Add parameters to the docker build script for Android, to make it
possible to inject custom configuration when calling it from other
scripts.
Signed-off-by: Michael Keller <github@ike.ch>
Update the AppImage workflow to be based on ubuntu 20.04 in order to
support the updated requirement of GitHub actions to run on node 20.
Signed-off-by: Michael Keller <github@ike.ch>
Some improvements:
- for constraints where the condition is applied to any one of a list of
items, show the condition as 'any X';
- when filtering for tags, do not include divemode as a 'tag' - this is
hard to impossible for users to understand, and we do have a dedicated
condition for the dive mode.
Signed-off-by: Michael Keller <github@ike.ch>
Ensure that any bluetooth scan started (by changing the device name) in the
download-from-dive-computer dialog (desktop version) is stopped before the
download process is started up. Because the QT bluetooth discovery agent uses
a QTimer internally, it must be stopped from the same thread as it was started
from. The download process uses a different thread, so ends up crashing when
it tries to dispose of the timer from a different thread.
Reported-by: Steve Buie <sbuie321@gmail.com>
Signed-off-by: Steve Buie <sbuie321@gmail.com>
Commit 185b4678ff changed the parser-test to use sorted dive
lists. However, for the "new Seabear" data format test, the
sorting was done after comparison. Which is obviously silly.
Fix it.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The dive list will be changed to an always-sorted list where one can
use binary search.
However, this makes some tests fail, because they only use parse_dive(),
which doesn't do any sorting.
To fix this future problem, sort the tables before performing the tests.
This provides a more realistic setup, as in the actual application,
the dive list will always be sorted on import.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
As noted in a comment introduced in fe074ccad1, the profile test
should probably best be run using the default preferences. This
wasn't done back then, because the reference data assumes a (bogus)
setting of modO2 of 0.
This commit runs the test using the default preferences and updates
the reference data accordingly.
This is in preparation of changes to the preference system, where
the preference structure initializes itself to the default values.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Fix loading of the negation of filter conditions. Unlike other
conditions that are persisted as `<key>="<value>"`, this is persisted
to git as `negate`.
This fix remediates this for all cases where the condition has already
been saved to the cloud storage.
Saving to XML takes a different approach and indicates negated
conditions with `negate="1"`, making it identical to all other
attributes. The question is if this approach should be implemented in
addition to the above fix, in order to unify the storage format.
Fixes#4246.
Signed-off-by: Michael Keller <github@ike.ch>
- add correct setting of the water type drop down for the dive shown
initially after program start;
- change salinity to have 3 decimals in planner, to make it consistency
with the log.
Fixes#4240.
Reported-by: @ccsieh
Signed-off-by: Michael Keller <github@ike.ch>
