Currently sequence messages are added to traces as they are
received. Therefore content of traces depends at the time, when
request is issued and data is reported. Sometimes not all sequence
traces are reported, because only first record was received yet and it
overwrote previous full sequence trace.
For example,
```
$GPGSV,3,1,12,01,35,282,,02,68,264,,03,02,243,,08,35,206,,1*68
$GPGSV,3,2,12,10,46,074,,14,16,312,,17,,,,22,16,327,,1*53
$GPGSV,3,3,12,23,07,068,,24,15,030,,27,14,181,,32,51,120,,1*60
$GLGSV,2,1,08,82,25,135,,76,11,029,,75,21,334,,69,11,201,,1*7A
```
one GSV sentence is missing and second later full sequence is
reported
```
$GPGSV,3,1,12,01,35,282,,02,68,264,,03,02,243,,08,35,206,,1*68
$GPGSV,3,2,12,10,46,074,,14,16,312,,17,,,,22,16,327,,1*53
$GPGSV,3,3,12,23,07,068,,24,15,030,,27,14,181,,32,51,120,,1*60
$GLGSV,2,1,08,82,25,135,,76,11,029,,75,21,334,,69,11,201,,1*7A
$GLGSV,2,2,08,84,40,306,,83,79,161,,67,56,056,,68,58,177,,1*7D
```
Therefore client software reports different number of visible
satellites.
Usually sequence sentences are reported consequently in NMEA
stream. Let's cache and join sequence sentences until new trace type
is received. In that case full sequence always overrides
previous full sequence.
Signed-off-by: Andrey Skvortsov <andrej.skvortzov@gmail.com>
The behavior of GRegex changed in 2.73.2 once it was ported from pcre1
to pcre2. In some cases it was made more strict, which is fine, in
other cases it exposed some change in how it behaves on certain
matches that is not extremely clear whether it's ok or not.
See https://gitlab.gnome.org/GNOME/glib/-/issues/2729
See https://gitlab.freedesktop.org/mobile-broadband/ModemManager/-/issues/601
See https://gitlab.freedesktop.org/mobile-broadband/ModemManager/-/issues/621
Either way, one thing that was assumed was that initializing all
GRegex/GMatchInfo variables to NULL and making sure they're NULL
before they're initialized by glib (especially the GMatchInfo) was a
good and safer approach.
So, whenever possible, g_autoptr() is used to cleanup the allocated
GMatchInfo/GRegex variables, and otherwise, g_clear_pointer() is used
to ensure that no free/unref is attempted unless the given variable is
not NULL, and also so that the variable is reseted to NULL after being
disposed.
Fixes the regexp in check_append_or_replace() to accept
1. more multi-sentence NMEA messages (ALM, GSV, RTE, SFI)
2. multi-sentence NMEA messages from all talkers, not just GPS
(i.e. GPGSV, GLGSV, GAGSV etc.)
The helper method returning a variant from a MMLocationGpsRaw would
return already a full variant reference instead of a floating one, so
we were really increasing the refcount when doing g_variant_ref_sink()
in the location interface.
Fix this by consolidating all helper methods in libmm-glib that return
variants from the different MMLocationXX objects, so that they all
return full variants instead of floating ones.
mm-location-gps-nmea.c: In function ‘mm_location_gps_nmea_get_string_variant’:
mm-location-gps-nmea.c:245:5: warning: ‘mm_location_gps_nmea_build_full’ is deprecated: Use 'mm_location_gps_nmea_get_traces' instead [-Wdeprecated-declarations]
245 | built = mm_location_gps_nmea_build_full (self);
| ^~~~~
There are potential memory leaks in MMLocationGpsNmea:
- When the `trace' string provided to location_gps_nmea_take_trace() isn't
added to the hash table, its ownership is still considered transferred.
It should thus be freed. Similarly, the `trace_type' string isn't
added the hash table and should thus be freed.
- mm_location_gps_nmea_add_trace() duplicates a given trace string and
then passes the trace copy to location_gps_nmea_take_trace(). When
location_gps_nmea_take_trace() returns FALSE, the ownership of the
copy isn't transferred. mm_location_gps_nmea_add_trace() should thus
free the copy.
This patch fixes the above memory leaks by having
location_gps_nmea_take_trace() always take the ownership of the `trace'
string and internally free `trace' and `trace_type' when necessary.
It's pointless to have libmm-common around, just merge it into libmm-glib and
make ModemManager depend on libmm-glib directly. At the end, the non-common
stuff in libmm-glib is really minimal.
Andrey Skvortsov
Aleksander Morgado
Aleksander Morgado
Teemu Ikonen
Ben Chan
Aleksander Morgado