Fix ESP8266 logging to not require global Serial object

[bdraco]

Fixes #379

PR #371 switched ESP8266 logging from ets_printf to Serial.printf_P, while nice and convenient for PROGMEM strings on ESP8266, it requires the global Serial object. This breaks builds using NO_GLOBAL_SERIAL to free up UART0 pins for other purposes.

This change restores using ets_printf with PROGMEM format strings and %c for the log level (char literals don't consume RAM).

Note: ESPHome may additionally use ASYNCWEBSERVER_LOG_CUSTOM to integrate with its own logging system, but that's a separate decision. This PR fixes the regression for all other users.

Untested - written while traveling and only verified it compiles ok for now. Will test on a real device when I reach a stable work environment.

Tested on an esp8266

[bdraco]

tested on a real esp8266 with ets_printf (may route to nothing), and ets_uart_printf (always uart) and no crashes

❯ SSE Crash Reproducer Starting...                                                                                                                             
  E async_ws 52: Test error log: 42                                                                                                                            
  W async_ws 53: Test warning log: 43                                                                                                                          
  I async_ws 54: Test info log: 44                                                                                                                             
  After logging macros   

[willmmiles]

Very nice! This is strictly superior to the previous version.

Two non-blocking questions:

• Is there any value in moving the hardcoded buffer size constant 64 to a separate #define, in case it needs to be adjusted? Or - given that the macro is instantiated at each call site - could the buffer size be set to sizeof(__fmt), so it'll float to the correct size for each format string?
• As another way to save code space, could the fixed portion of the format string ("%c async_ws %d: ") be stored in a common location for all calls instead of replicated for every call?

[bdraco]

Is there any value in moving the hardcoded buffer size constant 64 to a separate #define, in case it needs to be adjusted? Or - given that the macro is instantiated at each call site - could the buffer size be set to sizeof(__fmt), so it'll float to the correct size for each format string?

I didn't add a #define since there's only one use site and it's immediately visible. For the sizeof(__fmt) suggestion, that would size the buffer to the format string length, but the actual output can be larger once %d, %s, etc. are expanded with real values. A fixed buffer handles that safely.

As another way to save code space, could the fixed portion of the format string ("%c async_ws %d: ") be stored in a common location for all calls instead of replicated for every call?

Moving the prefix to a shared location would require two strcpy_P/strncpy_P operations at runtime instead of one. The extra complexity didn't justify the small flash savings. For RAM savings, that trade-off would be worthwhile.

[mathieucarbou]

I will merge and issue a fix release tomorrow.

[wmiles-sgl]

For the sizeof(__fmt) suggestion, that would size the buffer to the format string length, but the actual output can be larger once %d, %s, etc. are expanded with real values. A fixed buffer handles that safely.

That's true, but we aren't expanding text in to that buffer: it exists only to hold the format string.

[bdraco]

For the sizeof(__fmt) suggestion, that would size the buffer to the format string length, but the actual output can be larger once %d, %s, etc. are expanded with real values. A fixed buffer handles that safely.

That's true, but we aren't expanding text in to that buffer: it exists only to hold the format string.

You are right, I was thinking about the earlier iteration. Let me fix that.

[bdraco]

SSE connect, heap: 34688
E async_ws 234: Event message queue overflow: discard message
E async_ws 234: Event message queue overflow: discard message
E async_ws 234: Event message queue overflow: discard message
SSE disconnect, heap: 35008
E async_ws 234: Event message queue overflow: discard message
E async_ws 234: Event message queue overflow: discard message
E async_ws 234: Event message queue overflow: discard message
SSE connect, heap: 33504
E async_ws 234: Event message queue overflow: discard message
E async_ws 234: Event message queue overflow: discard message
SSE disconnect, heap: 31296
SSE disconnect, heap: 33712
SSE connect, heap: 35776
SSE connect, heap: 35600

still works fine

[bdraco]

Thanks for the quick release and review