Fix: KNX empty UDP packets sent to the network

[morgenroth]

Description:

The flush() call after read() causes empty packets being sent as response
for all UDP packets received on the KNX UDP port. In my tests with sonoff devices, the zero length packets mentioned in #8332 vanished after removing that flush() call.

Since I do not have any ESP32 device, it would be nice if someone else could test PR.

Related issue (if applicable): fixes #8332

Checklist:

• The pull request is done against the latest dev branch
• Only relevant files were touched
• Only one feature/fix was added per PR.
• The code change is tested and works on core ESP8266 V.2.7.1
• The code change is tested and works on core ESP32 V.1.12.0
• I accept the CLA.

NOTE: The code change must pass CI tests. Your PR cannot be merged unless tests pass

[ascillato]

Nice.

Please, let me do some tests over this.

On ESP32 that issue didn't happen because the problem on empty response with flush command is just arduino core related and arduino core for ESP32 don't have that issue.

[arendst]

@ascillato Adrian let me know if to merge.

[ascillato]

Testing. Please, give me some time.

[ascillato]

This Bug is only in the Arduino Core for ESP8266. It have happened before, it was fixed in the core but some changes in the core have reintroduce it.

The FLUSH command is needed in the ESP KNX Library. It can't be deleted. If deleted, in a busy network, your device will start to be unresponsive or slow to respond due to full UDP RX buffer on both ESP8266 and ESP32.

Checking the source code of the Cores, you can easily see that the flush command is not performing as it should, but only in ESP8266.

On Arduino ESP8266 Core - File /libraries/ESP8266WiFi/src/WiFiUdp.cpp:

void WiFiUDP::flush()
{
    endPacket();
}

int WiFiUDP::endPacket()
{
    if (!_ctx)
        return 0;

    return (_ctx->send()) ? 1 : 0;
}

On Arduino ESP32 Core - File /libraries/WiFi/src/WiFiUdp.cpp:

void WiFiUDP::flush(){
  if(!rx_buffer) return;
  cbuf *b = rx_buffer;
  rx_buffer = 0;
  delete b;
}

Therefore, I'm creating an issue in Arduino ESP8266 repository and closing this PR. Anyway, your research was very useful to find faster the bug in the core.

---

EDIT: Seems that the bug was introduced here esp8266/Arduino#3967 esp8266/Arduino@26980b3#diff-9480bf2495731fa559bb9c3019650538

[ascillato]

As explained in:

• Fix Flush implementations to match corrected code in latest Arduino libraries esp8266/Arduino#1002,
• follow arduino API: pure virtual Stream::flush() moved to empty virtual Print::flush() esp8266/Arduino#4029,
• fix #1002 ::Flush() wait for empty send buffer esp8266/Arduino#3967,
• flush() is a TX related function esp8266/Arduino#4018,
• and in line 105 of print.h of the core,
  https://github.com/esp8266/Arduino/blob/3e4d7c76c42f95933695065737e674c9234ec464/cores/esp8266/Print.h#L105

they move the flush command to match the actual Arduino framework, being now a TX command instead of clearing buffer commands.

So, what was happening was intended from the core side.

Your change was correct and other optimizations should be implemented on the driver.

For now, it is better to merge your changes as it is and do the optimizations later on.

Thanks for sharing.