If the client or server crashes during the integration test run
the next write to its fifo file will be blocking and get stuck forever.
This is especially annoying for github actions because they just run 10
hours and then get terminated by github without ever reaching the end
of the integration test script that then prints the details about the
crash.
ALso exclude coverage.map from twmap-checker due to using tele
checkpoints without matching tele checkpoints to test physics.
Disable exact checks on valgrind. Only require at least one finish.
Fix valgrind errors not being detected as failures when the tests otherwise finish, as the valgrind results are printed to stderr and not to the logfiles. Additionally, valgrind failures were not being detected because the error summary is not printed to stderr immediately but only after some time when valgrind has processed the summary.
Find random unused port to bind to by opening a socket on port 0 and closing it immediately. This gives a port that will very likely be unused because the system cycles through all ports before reusing old ones obtained this way.
Add `wait_for_launch` function to reduce duplicate code. Wait for server to launch before starting clients.
Improve log messages.
Spam like the following is now hidden by the default config file:
```
[2022-06-16 01:55:22][console]: moderator access for 'left' is now enabled
[2022-06-16 01:55:22][console]: helper access for 'left' is now enabled
[2022-06-16 01:55:22][console]: user access for 'left' is now disabled
```
Also, the following were already converted to debug messages by #5904.
```
[2022-06-16 01:57:05][datafile]: loading data index=88 size=6159 uncompressed=875088
[2022-06-16 01:57:05][datafile]: loading data index=89 size=1481 uncompressed=972400
[2022-06-16 01:57:05][datafile]: loading data index=90 size=7199 uncompressed=972400
```
Remove square brackets to reduce the amount of space used.
Add log level indicator. The position between timestamp and system was
chosen because it is at a fixed position (unlike after the system) but
the log still remains naively sortable (which wouldn't happen if we were
to place it in front of the timestamp.
Before:
```
[2022-04-29 15:25:37][engine]: running on unix-linux-amd64
[2022-04-29 15:25:37][engine]: arch is little endian
[2022-04-29 15:25:37][storage]: added path '$USERDIR' ('/path/to/home/.teeworlds')
[2022-04-29 15:25:37][storage]: added path '$DATADIR' ('data')
[2022-04-29 15:25:37][storage]: added path '$CURRENTDIR' ('/path/to/ddnet')
[2022-04-29 15:25:37][host_lookup]: host='localhost' port=0 1
[2022-04-29 15:25:37][host_lookup]: host='localhost' port=0 2
[2022-04-29 15:25:37][console]: executing 'autoexec_server.cfg'
```
After:
```
2022-04-29 15:25:37 I engine: running on unix-linux-amd64
2022-04-29 15:25:37 I engine: arch is little endian
2022-04-29 15:25:37 I storage: added path '$USERDIR' ('/path/to/home/.teeworlds')
2022-04-29 15:25:37 I storage: added path '$DATADIR' ('data')
2022-04-29 15:25:37 I storage: added path '$CURRENTDIR' ('/path/to/ddnet')
2022-04-29 15:25:37 I host_lookup: host='localhost' port=0 1
2022-04-29 15:25:37 I host_lookup: host='localhost' port=0 2
2022-04-29 15:25:37 I console: executing 'autoexec_server.cfg'
```
When the client or server crash the fifo will be dead
then the cleanup command hangs at
echo "quit" > client1.fifo
because the fifo is not responding and thus it is stuck there.
which causes the user to probably ctrl+c at some point
which then avoids client2 to be quit which might not have crashed
So this commit checks if a fail_client1.txt file exits and if it crashed
already it will skip the call to the broken fifo
See also Jupeyy's explanation in
https://github.com/ddnet/ddnet/issues/4294 for why we want to ignore
this:
> sadly components have no OnDestruct on anything like that
> so aslong it doesn't actually leak it's just non destructed memory
So it's something we allocate once for the entire client lifetime from
my understanding.
=================================================================
==421461==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 411760 byte(s) in 10 object(s) allocated from:
#0 0x55eaf0c08a29 in malloc (/media/ddnet/integration_test/DDNet+0x736a29)
#1 0x55eaf0de9f72 in CCommandProcessorFragment_OpenGL2::Cmd_CreateBufferObject(CCommandBuffer::SCommand_CreateBufferObject const*) /media/ddnet/src/engine/client/backend/opengl/backend_opengl.cpp:1898:25
#2 0x55eaf0dd82f5 in CCommandProcessorFragment_OpenGL::RunCommand(CCommandBuffer::SCommand const*) /media/ddnet/src/engine/client/backend/opengl/backend_opengl.cpp
#3 0x55eaf0f73229 in CCommandProcessor_SDL_GL::RunBuffer(CCommandBuffer*) /media/ddnet/src/engine/client/backend_sdl.cpp:246:20
#4 0x55eaf0f70568 in CGraphicsBackend_Threaded::ThreadFunc(void*) /media/ddnet/src/engine/client/backend_sdl.cpp:75:25
#5 0x55eaf0d679d7 in thread_run(void*) /media/ddnet/src/base/system.cpp:686:2
#6 0x7f08b788c54c (/usr/lib/libc.so.6+0x8c54c)
SUMMARY: AddressSanitizer: 411760 byte(s) leaked in 10 allocation(s).
As described in #5295, the integration test script waits for the client
to close before launching the next one. I don't know what causes this,
but disabling full-screen mode works around the issue.
I don't know if this is okay. @ChillerDragon?
Fixes#5295.
ChillerDragon
