summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeLines
* Call GetProfileNameForm::init() in the constructorFrédéric Brière2020-03-31-0/+1
| | | | | | | init() is responsible for setting a validator for profileLineEdit; its invocation was probably lost in the Qt4/5 transition. Closes #189
* Merge pull request #152 from fbriere/issue/104-bcg729-apiLuboš Doležel2020-03-29-0/+68
|\ | | | | Add support for the new bcg729 API, introduced in version 1.0.2
| * Test building with bcg729 on Travis CI, trying both APIsFrédéric Brière2020-03-29-0/+9
| |
| * Add support for the new bcg729 API, introduced in version 1.0.2Frédéric Brière2020-03-29-0/+59
|/ | | | | | | | | | | | | | | | | | | | | | | | | Starting with version 1.0.2, bcg729 has changed its API to add support for G.729B, thus requiring us to adjust our function calls depending on which version is installed. When dealing with the new API, we merely need to add a few parameters to disable all G.729B features, namely: * On the decoder side: When `SIDFrameFlag` is not set, the decoder will behave just like before, decoding the payload as a standard G.729A voice frame (or concealing an erased frame). The other parameters, `rfc3389PayloadFlag` and `bitStreamLength`, are only of use when dealing with a SID frame sent as per RFC 3389, and are ignored if `SIDFrameFlag` is not set. * On the encoder side: When `enableVAD` is disabled, the encoder will behave just like before, producing only standard G.729A voice frames. The only API difference is the introduction of `*bitStreamLength`, to return the length of the encoded frame (0, 2 or 10 bytes). In our case, this will always be 10 bytes just like before; an assert() was added to guarantee this. Closes #104
* Merge pull request #194 from glixx/masterLuboš Doležel2020-02-24-175/+68
|\ | | | | update Russian translation
| * update Russian translationglixx2020-02-24-175/+68
|/
* Merge pull request #193 from jose1711/masterLuboš Doležel2020-02-23-23/+23
|\ | | | | Add more markdown in README.md
| * Add more markdown in README.mdJose Riha2020-02-23-23/+23
|/ | | | | Mostly cosmetic changes. This also solves <user> string being invisible on github page.
* Merge pull request #180 from fbriere/issue/165-deadlocks-complexLuboš Doležel2020-02-23-29/+243
|\ | | | | Introduce read-write-update locks and guards to prevent deadlocks
| * Fix "REFER declined" deadlockFrédéric Brière2019-12-27-1/+1
| | | | | | | | | | | | | | | | | | This occurs when the user declines an incoming call transfer request, due to the following path: - recvd_refer_permission() acquires a read lock over lines_mtx - move_releasing_lines_to_background() is called - A write lock over lines_mtx is acquired
| * Fix "Transfer with consultation without Replaces" deadlockFrédéric Brière2019-12-27-1/+1
| | | | | | | | | | | | | | | | | | | | | | This occurs when performing a consult transfer if the target does not support the Replaces extension, due to the following path: - refer() acquires a read lock over lines_mtx - refer_consultation() is called - A write lock over lines_mtx is acquired Closes #118
| * Fix "NOTIFY received when line is idle" deadlockFrédéric Brière2019-12-27-1/+1
| | | | | | | | | | | | | | | | | | | | This can easily be triggered by starting a blind transfer, and hanging up before the other end has accepted/rejected it, due to the following path: - recvd_notify() acquires a read lock on lines_mtx - cleanup_dead_lines() is called - A write lock on lines_mtx is acquired
| * Fix "INVITE with Replaces header" deadlockFrédéric Brière2019-12-27-1/+1
| | | | | | | | | | | | | | | | | | This occurs on reception of an INVITE with a Replaces header, due to the following path: - recvd_invite() acquires a read lock on lines_mtx - recvd_initial_invite() is called - A write lock over lines_mtx is acquired
| * Introduce read-write-update locks and guards to prevent deadlocksFrédéric Brière2019-12-27-13/+199
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This converts t_rwmutex and t_rwmutex_guard into a read-write-update[*] lock and guard, in an attempt to circumvent the various deadlocks that were introduced with the addition of lines_mtx in 38bb6b7. [*] For more details, see https://stackoverflow.com/a/18785300 and http://lkml.iu.edu/hypermail/linux/kernel/0004.3/0117.html. Note that this is not a real fix; this would require analyzing and refactoring phone.cpp, which is well beyond my abilities. This is at best a workaround that appears to conveniently dodge all the deadlocks I've encountered so far. (It would have been more proper to introduce a separate class for this purpose, but this would have required modifying over 80 lines just to change one type for another. As phone_users_mtx is the only other instance of this class, the impact of subverting t_rwmutex directly is minimal.)
| * Create t_rwmutex_guard base classFrédéric Brière2019-12-26-7/+19
| | | | | | | | (Doing this ahead of time to simplify the next commit a bit.)
| * Move t_rwmutex_reader/writer definitions into mutex.cppFrédéric Brière2019-10-02-16/+32
| | | | | | | | | | These classes are about to get more complex, so let's move them ahead of time into mutex.cpp.
* | Merge pull request #163 from fbriere/issue/162-multiple-auth-headersLuboš Doležel2020-02-23-0/+13
|\ \ | | | | | | Support multiple WWW-Authenticate/Proxy-Authenticate headers
| * | Support multiple WWW-Authenticate/Proxy-Authenticate headersFrédéric Brière2019-09-21-0/+13
| | | | | | | | | | | | | | | | | | | | | | | | | | | The server may send multiple WWW-Authenticate/Proxy-Authenticate headers, with different digest algorithms, in decreasing order of preference. We must therefore avoid overwriting any supported challenge once we've got a hold of one. Closes #162
* | | Merge pull request #154 from fbriere/feature/inhibit-idle-sessionLuboš Doležel2020-02-23-0/+531
|\ \ \ | | | | | | | | Add an option to prevent an idle session while a call is in progress
| * | | Properly display and log warning messages in IdleSessionInhibitorFrédéric Brière2019-07-13-13/+27
| | | |
| * | | Add an option to prevent an idle session while a call is in progressFrédéric Brière2019-07-13-0/+517
| |/ / | | | | | | | | | | | | | | | | | | | | | | | | Having the session marked as idle while a call is in progress could trigger certain actions (such as locking the screen, logging out automatically, or suspending the system) which could be undesirable in this situation. Closes #123
* | | Merge pull request #148 from fbriere/feature/disable-incoming-call-popupLuboš Doležel2020-02-23-2/+42
|\ \ \ | | | | | | | | Add configuration option to disable incoming call popup window
| * | | Add configuration option to disable incoming call popup windowFrédéric Brière2019-06-30-2/+42
| |/ / | | | | | | | | | Closes #126
* | | Merge pull request #147 from fbriere/fix/mutex-on-boolLuboš Doležel2020-02-23-2/+2
|\ \ \ | | | | | | | | Use a mutex in all t_sys_settings getters/setters, even bool
| * | | Use a mutex in all t_sys_settings getters/setters, even boolFrédéric Brière2019-07-03-2/+2
| |/ / | | | | | | | | | | | | | | | | | | | | | | | | | | | Despite what one might intuitively expect, the C++ standard does not make any guarantee about fundamental types being atomic, not even bool. (In fact, it explicitly mentions the existence of std::atomic<bool> for that purpose.) See https://stackoverflow.com/a/35226186 for more details about this subject.
* | | Merge pull request #116 from fbriere/misc/old-autotools-stuffLuboš Doležel2020-02-23-2/+62
|\ \ \ | | | | | | | | Add missing tests lost in the autotools to CMake transition
| * | | CMake: Check if libilbc links without 'extern "C"' (ILBC_CPP)Frédéric Brière2019-12-29-0/+23
| | | | | | | | | | | | | | | | This variable was originally set manually in configure.in.
| * | | CMake: Check for endianness (WORDS_BIGENDIAN)Frédéric Brière2019-12-29-0/+4
| | | | | | | | | | | | | | | | | | | | This check (AC_C_BIGENDIAN) was originally present in configure.in, and was lost in the transition to CMake.
| * | | CMake: Check for res_init() (HAVE_RES_INIT)Frédéric Brière2019-12-29-0/+15
| | | | | | | | | | | | | | | | | | | | This check (AC_CHECK_RES_INIT) was originally defined in acinclude.m4, and was lost in the transition to CMake.
| * | | CMake: Check for strerror_r() (HAVE_STRERROR_R and STRERROR_R_CHAR_P)Frédéric Brière2019-12-29-2/+20
| |/ / | | | | | | | | | | | | This check (AC_FUNC_STRERROR_R) was originally present in configure.in, and was lost in the transition to CMake.
* | | Merge pull request #149 from fbriere/issue/121-toggle-window-menu-entryLuboš Doležel2020-02-23-6/+33
|\ \ \ | | | | | | | | Systray icon: Always toggle visibility when clicking & Add "Show/Hide" menu entry
| * | | Add a "Show/Hide window" entry to the systray icon menuFrédéric Brière2019-06-30-1/+33
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | It's customary for applications embedding themselves in the system tray to offer a "Show/Hide window" menu entry on right-click, even for those applications which offer the same functionality via a single left-click. (Thanks to qBittorrent for the idea of using QMenu::aboutToShow to update the label of this menu entry.)
| * | | Always toggle window visibility when clicking on the systray iconFrédéric Brière2019-06-30-6/+1
| |/ / | | | | | | | | | | | | | | | | | | | | | | | | Left-clicking on the system tray icon should always result in toggling the visibility of the main window; if the icon is visible and clickble, then the window can always be hidden via --hide, or on startup via the "Startup hidden in system tray" option. (In the latter case, this previously resulted in a hidden and inaccessible window, as reported in issue #121.)
* | | Merge pull request #164 from fbriere/fix/qml-binding-loopLuboš Doležel2020-02-23-1/+7
|\ \ \ | | | | | | | | Fix QML binding loop in TextImageButton (used in incoming call popup)
| * | | Fix QML binding loop in TextImageButton (used in incoming call popup)Frédéric Brière2019-09-25-1/+7
| |/ /
* | | Merge pull request #144 from fbriere/issue/143-readline-blocks-cmd-quitLuboš Doležel2020-02-23-17/+93
|\ \ \ | | | | | | | | Respond immediately to a "quit" command while in CLI mode
| * | | Make Readline non-blocking: Use a self-pipe to break the Readline loopFrédéric Brière2019-06-29-0/+32
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Now that we are no longer blocking on Readline calls, we can set up a self-pipe that will let us break out of the Readline loop upon receiving a "quit" command on our local socket, thus (finally) fixing issue #143. (Thanks to https://stackoverflow.com/a/27662212 for the tip!) Fixes #143
| * | | Make Readline non-blocking: Wrap select(2) around Readline callFrédéric Brière2019-06-29-1/+23
| | | | | | | | | | | | | | | | | | | | By only invoking rl_callback_read_char() when there is actual input on stdin, we can now avoid blocking on Readline calls.
| * | | Make Readline non-blocking: Relay SIGWINCH to ReadlineFrédéric Brière2019-06-29-0/+16
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When using Readline's callback interface, signals are (by default) only captured within the duration of the rl_callback_read_char() call. It is therefore expected of the application to capture SIGWINCH on its own, and notify Readline of the fact. (While it's possible to change this with rl_persistent_signal_handlers, some signals, such as SIGHUP, will still only be *processed* during that call, making it a somewhat unappealing solution. This all could be alleviated by calling rl_check_signals() periodically, but that function was only introduced in Readline 8.0, which is far too recent.) Note that we are using signal(2) and not sigaction(2), depite the various warnings that come with it, mostly because it's what is already present in the codebase.
| * | | Make Readline non-blocking: Use Readline's callback interfaceFrédéric Brière2019-06-29-19/+25
| |/ / | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When Twinkle is running in CLI mode and is sent a "quit" command to its local socket, it will currently not respond immediately, but rather wait until the next line has been read from its standard input (issue #143). This is due to the blocking nature of readline(), which only returns once a complete line has been read. Switching to Readline's "alternate" callback interface is the first step in addressing this issue. (As a bonus, this also fixes a bug where the line pointer returned by readline() was not freed correctly.)
* | | Merge pull request #158 from 4-FLOSS-Free-Libre-Open-Source-Software/patch-1Luboš Doležel2020-02-23-2/+21
|\ \ \ | | | | | | | | Add twinkle-console option --sip-port --rtp-port
| * | | Add twinkle-console option --sip-port --rtp-port4-FLOSS-Free-Libre-Open-Source-Software2019-07-19-2/+21
| |/ /
* | | Merge pull request #150 from fbriere/issue/88-mutex-lock-failedLuboš Doležel2020-02-23-2/+8
|\ \ \ | | | | | | | | Prevent recursive locking of phone_users_mtx in add_phone_user()
| * | | Prevent recursive locking of phone_users_mtx in add_phone_user()Frédéric Brière2019-07-03-2/+8
| |/ / | | | | | | | | | | | | | | | | | | | | | When encountering two users with the same contact name, attempting to differentiate them using USER_HOST(), a.k.a. get_ip_sip(), would result in EDEADLK since phone_users_mtx was already locked by add_phone_user(). Closes #88
* | | Merge pull request #160 from 4-FLOSS-Free-Libre-Open-Source-Software/patch-2Luboš Doležel2020-02-23-1/+1
|\ \ \ | | | | | | | | resolve nat_public_ip hostname for dyndns to work
| * | | resolve nat_public_ip hostname for dyndns to work4-FLOSS-Free-Libre-Open-Source-Software2019-08-01-1/+1
| |/ /
* | | Merge pull request #166 from fbriere/issue/165-deadlocks-simpleLuboš Doležel2020-02-23-3/+1
|\ \ \ | | |/ | |/| Simple fixes for two deadlocks
| * | Make `t_phone::mutex_3way` recursive to avoid a deadlockFrédéric Brière2019-10-02-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | `t_phone::mutex_3way` can be locked twice when hanging up a conference call: - `t_phone::cleanup_3way_state()` acquires a lock - `t_audio_session::stop_3way()` is called - `t_audio_session::get_peer_3way()` is called - `t_phone::get_3way_peer_line()` is called - which acquires another lock Making that mutex recursive is a simple way to work around this issue.
| * | Remove duplicate mutex lock in `t_phone::end_call()`Frédéric Brière2019-10-02-2/+0
| |/ | | | | | | | | A write lock on `lines_mtx` has already been acquired at the beginning of `end_call()`.
* | Merge pull request #182 from fbriere/misc/travis-bionicLuboš Doležel2020-02-23-9/+14
|\ \ | | | | | | Switch Travis CI distribution to bionic
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-22 00:32:36 +0000