IPAddress::operator= needs 32-bit aligned pointer

[TD-er]

Basic Infos

• This issue complies with the issue POLICY doc.
• I have read the documentation at readthedocs and the issue is not addressed there.
• I have tested that the issue is present in current master branch (aka latest git).
• I have searched the issue tracker for a similar issue.
• If there is a stack dump, I have decoded it.
• I have filled out all fields below.

Platform

• Hardware: [ESP-12|ESP-01|ESP-07|ESP8285 device|other]
• Core Version: [Current master / 2.7.4]
• Development Env: [Platformio]
• Operating System: [Windows|Ubuntu]

Settings in IDE

• Module: [Generic ESP8266 Module]
• Flash Mode: [qio|dio|other]
• Flash Size: [4MB/1MB]
• lwip Variant: [v1.4|v2 Lower Memory|Higher Bandwidth]
• Reset Method: [ck|nodemcu]
• Flash Frequency: [40Mhz]
• CPU Frequency: [80Mhz|160MHz]
• Upload Using: [OTA|SERIAL]
• Upload Speed: [115200|other] (serial upload only)

Problem Description

LoadStoreAlignmentCause: Load or store to an unaligned address
  epc1=0x402a1358 in IPAddress::operator=(unsigned char const*) at ??:?
en deze stacktrace:
0x4029280d in String::operator=(char const*) at ??:?
0x401012bb in umm_free_core at umm_malloc.cpp:?
0x401015a2 in malloc at ??:?
0x402d1866 in wifi_get_opmode at ??:?
0x402a2568 in spiffs_cache_page_get_by_fd at ??:?
0x402a2568 in spiffs_cache_page_get_by_fd at ??:?
0x402a2568 in spiffs_cache_page_get_by_fd at ??:?
0x402a2568 in spiffs_cache_page_get_by_fd at ??:?
0x4024821f in setConnectionSpeed() at ??:?
0x40248dc2 in prepareWiFi() at ??:?
0x40241468 in WiFiEventData_t::markWiFiBegin() at ??:?
0x40248f29 in AttemptWiFiConnect() at ??:?
0x40205e24 in ESP8266WiFiSTAClass::status() at ??:?
0x40248084 in wifiAPmodeActivelyUsed() at ??:?
0x4024909c in WiFiConnectRelaxed() at ??:?
0x40245ce9 in NetworkConnectRelaxed() at ??:?
0x4024bf23 in ESPEasy_setup() at ??:?
0x40215084 in setup at ??:?
0x40293ddb in loop_wrapper() at core_esp8266_main.cpp:?
0x40100469 in cont_wrapper at ??:?

See the assignment operator and other functions where the const uint8_t* is cast to an const uint32_t*

Arduino/cores/esp8266/IPAddress.cpp

Lines 118 to 122 in f06710e

	IPAddress& IPAddress::operator=(const uint8_t *address) {
	setV4();
	v4() = *reinterpret_cast<const uint32_t*>(address);
	return *this;
	}

Arduino/cores/esp8266/IPAddress.cpp

Lines 130 to 132 in f06710e

	bool IPAddress::operator==(const uint8_t* addr) const {
	return isV4() && v4() == *reinterpret_cast<const uint32_t*>(addr);
	}

The call is simply like this:

  const IPAddress ip     = Settings.IP;
  const IPAddress gw     = Settings.Gateway;
  const IPAddress subnet = Settings.Subnet;
  const IPAddress dns    = Settings.DNS;

Where all these are part of a larger struct, which begins like this:

  unsigned long PID = 0;
  int           Version = 0;
  int16_t       Build = 0;
  uint8_t       IP[4] = {0};
  uint8_t       Gateway[4] = {0};
  uint8_t       Subnet[4] = {0};
  uint8_t       DNS[4] = {0};

As can be seen, these are indeed not 32-bit aligned.

I think the build toolchain may recently have been updated, as this struct and the code that's now failing has not been changed for at least 4 years.

The same code in IPAddress has not been changed for a while as the current master branch has the same code for these functions as the 2.7.4 we're using.

[mcspr]

s/Needs aligned/We need memcpy into u32 and then copy instead of reinterpret cast/?
If you compare with 2.7.4, things definitely have changed b/c of GCC 10.3

[TD-er]

There is already a proper constructor which does copy the bytes one-by-one, so why should there be a reinterpret_cast or memcpy?

By the way, I did a few more tests to let the compiler find all occurences of this operator= and found a few more.
I did make the assignment operator private.

But strangely enough the exact lines as above with const IPAddress ip = Settings.IP; etc. were not found.
This one was: WiFiEventData.dns0_cache = dns; which was one line below in my code. (dns0_cache is of type IPAddress)

So maybe the compiler already may strip this code in the optimizer step?

This is the complete function:
https://github.com/letscontrolit/ESPEasy/blob/0c0175ebed4938aecc21d60ea595af3f88fe9544/src/src/ESPEasyCore/ESPEasyWifi.cpp#L1448-L1468

[d-a-v]

v4() = *reinterpret_cast<const uint32_t*>(address);

Can you try

memcpy_P(&v4(), address, 4);

[TD-er]

OK.. also tried with IPAddress& operator=(uint32_t address); set to private to see what this may show up and then noticed some very strange thing in int ESP8266WiFiGenericClass::hostByName(const char* aHostname, IPAddress& aResult, uint32_t timeout_ms)

aResult = static_cast<uint32_t>(INADDR_NONE); where INADDR_NONE is already of type IPAddress, so casting to uint32_t is not needed.
And what makes it even more strange is that INADDR_NONE is 255.255.255.255 on ESP8266, but 0.0.0.0 on ESP32.

[d-a-v]

And what makes it even more strange is that INADDR_NONE is 255.255.255.255 on ESP8266, but 0.0.0.0 on ESP32.

isSet() is preferred with ESP8266 (which tests both)

[mcspr]

Let's just figure out assignments & ctors first. I'll prepare a PR.

[TD-er]

v4() = *reinterpret_cast<const uint32_t*>(address);

Can you try

memcpy_P(&v4(), address, 4);

This is a fix.
Just tested it and with this code it works just fine. reverted it and crashes very reproducible.

Just strange that the lines with the single used const IPAddress ip = Settings.IP; didn't fail to compile with the assignment operator set to private.