The intention is that JFreegman would run it locally on his machine for
every Toxic release and publish the build artifacts on GitHub for
download, similarly to how the qTox team makes release builds. This
should mitigate the loss of the Jenkins build server to some extent,
which used to provide Toxic builds before.
When a user provides variable=value as an argument to make, all
assignments to that variable are ignored within the Makefile because the
user has explicitly overrode variable to be "value". This made the
ENABLE_ASAN assignment to be ignored, resulting in Toxic always enabling
ASAN unless you run `make ENABLE_ASAN=disabled`, which is not
documented and not how it's intended to work. This can be fixed by
prefixing the assignment with "override", but to be in line with other
argument assignments we just change the variable name.
See more at:
https://www.gnu.org/software/make/manual/html_node/Overriding.html
- Give window tab and statusbar a make over
- Place window tab above input field
- Reduce input field to one square in height
- Refactor window tab so that it's now a subwin of its parent ToxWindow
- Fix bug causing notification counter to sometimes increment by 2
- No longer scroll on output when output is not at bottom of screen
- Show a small indicator on far left of window tab when output is
not at bottom of screen
- Reduce ncurses/UI thread sleep time by half
- Handle nanosleep errors better
A line's line count now increments without messing up the formatting in cases where
the noread flag wraps to the next line.
Additionally, the way noread flags are handled has been refactored and made
more efficient.
When renaming a log file, if the new name already exists we just
start appending to it and delete the old file. There's no need
to create a backup.
This fixes an issue where leaving and rejoining the same group
multiple times will eventually lead to the logger not working
due to trying to use the same file name over and over again.
The release build uses -O2 and flto, and has no debug symbols.
-Wmissing-field-initializer was removed due to false positives
on newer versions of clang
Instead of trying (and failing) to correctly calculate the number of lines on the
screen a given message will need space for, we dynamically set the value both
on initiation of a new line, and whenever that line is printed to the screen
using the print_wrap() function.
This should fix the screen occasionally not scrolling down the correct number of
lines when a new message is printed.
A counter now increments in the bottom bar for unfocused windows showing how
many unread messages are pending. Tabs with no pending messages show [*]
instead of their index (showing the index is useless and somewhat confusing)
- Fix race condition in draw_peer()
- Handle realloc_peer_list() error
- Remove dead code in cmd_conference()
- Reduce scope of a few variable variables
- Fix possible buffer truncation in api.c
- Bottom tab now only shows indices of active chat windows unless focused
- Always focus Home screen on startup instead of the last loaded conference
- Conference tab names are no longer prefixed with the conference number
- Home and Contact tab names are now capitalized
- Conference logging now behaves the same as 1-on-1 chats: Instead
of creating a new log file every time we restat the client
we use the unique conference ID to keep track of path names.
This also allows us to load history for saved groups on client startup
- Added a log init function / general code refactor.
- Fixed a bug that caused log files to be created even when logging
is disabled.
- Fix bug causing failed avatar transfers to be sent as normal file transfers
when a friend goes offline and comes back online
- Remove some unused members of the FileTransfer struct
- Rename filenum -> filenumber and friendnum -> friendnumber
- It's no longer possible for messages to be sent out of order
- Check if logger is enabled before doing pointless API calls
- Fix linked list bug (a part of the code almost never executes)
We now attempt to send all queued messages per call to cqueue_try_send() instead
of just the oldest message in the queue. This speeds things up substantially.
Also fixed a very unlikely bug where the read receipt might wrap around to zero
which we used as a reserved value for an unsent message
jfreegman
Maxim Biro
Lily
zugz (tox)