Are multiple calls to AsyncClient::_discard_cb intentional?

[mastersign]

I am using AsyncTCP as part of ESPAsyncWebServer and I occasional get a heap corruption.

The reason is, that in some scenarios AsyncClient::_close() and AsyncClient::_fin() are called in succession. They call _discard_cb with the following call sequence:

• AsyncClient::_discard_cb = Lambda from AsyncWebServerRequest constructor
• AsyncWebServerRequest::_onDisconnect()
• AsyncWebServer::_handleDisconnect()
• delete request → double free!

So my question is: Is it intentional, that the onDisconnect callback of an AsyncClient is called multiple times?

• If yes, then the bug is the ESPAsyncWebServer library. Somehow the freeing of the request instance must be reworked.
• If no, then all calls to _discard_cb in AsyncClient should be followed by a _discard_cb = NULL.

The phenomenon occures more often when connecting from a mobile Chrome browser on Android, then from a Chrome on PC. My guess is, that the connection handling differs somewhat.

[mathieucarbou]

I already saw this issue from time to times yes and I merged a PR from @willmmiles 4 hours ago that is likely to fix this long standing issue (#79).

Can you try with the new release ?

The callback should obviously be called once and there should not be any double free.

[mastersign]

The heap corruption does not occur with AsyncTCP 3.4.6. I conclude, that the fixed race condition between close and fin resolved the double call of _discard_cb. Thank you for the quick reply! And thanks to @willmmiles for the fix!