std::string error

[JonB]

@mzimmers
I know nothing about embedded & debugging (you can't give us a stack trace, can you?), but in getIpAddr() can you check the value of s immediately before return s;? I guess you're going to say it's OK then, but not as it gets returned to the string s2(m_params->nvs->getIpAddr(IP_ADDRESS)); line?

[mzimmers]

Yeah it looks fine in the Nvs routine:

    ESP_LOGI(TAG, "getIpAddr(): returning \"%s\".", s.c_str());
    vTaskDelay(10);
    return s;

Produces: I (26142) Nvs: getIpAddr(): returning "10.10.0.157".

[JonB]

@mzimmers
So if you have debug like that, try to debug out the result from m_params->nvs->getIpAddr(IP_ADDRESS)); in the caller? You're saying doing that raises exception? (BTW, do we get any details about the exception, I don't think you've said?)

[mzimmers]

I can't output the result -- the error occurs as I'm trying to execute that line.

The error is known in the ESP32 world as a "Guru Meditation Error," and about all I get with my current debugging resources is this:

Guru Meditation Error: Core  0 panic'ed (LoadProhibited). Exception was unhandled.
Core 0 register dump:
PC      : 0x400e1bae  PS      : 0x00060230  A0      : 0x800dd03d  A1      : 0x3ffebf80
A2      : 0x3fff17c4  A3      : 0x3fff0ef0  A4      : 0x3fff17c4  A5      : 0x3fff17c8
A6      : 0x00000000  A7      : 0x3fff0ef4  A8      : 0x00000000  A9      : 0x3ffebf30
A10     : 0x73d2d9f0  A11     : 0x73d2d9f0  A12     : 0x00000000  A13     : 0x0000001f
A14     : 0x00000001  A15     : 0x00000005  SAR     : 0x00000004  EXCCAUSE: 0x0000001c
EXCVADDR: 0x00000010  LBEG    : 0x400014fd  LEND    : 0x4000150d  LCOUNT  : 0xfffffffd

Backtrace: 0x400e1bae:0x3ffebf80 0x400dd03a:0x3ffec040 0x400dd48d:0x3ffec290 0x400de12b:0x3ffec730 0x400de197:0x3ffec780

This is very likely what the Linux world refers to as a segmentation fault. Sorry I can't give better information; the debugging tools are the weak point of ESP32 development (IMO).

[hskoglund]

Hi, if you change to a static s, does it crash in the same way?
I.e change to:

...
string Nvs::getIpAddr(IpIdentifier addr)
{
    static string s;
    char *pAddr;
...

[mzimmers]

@hskoglund yes it does. I eliminated the pointer in favor of a static char array, too; same result.

[hskoglund]

Hmm maybe some of the pointers when calling are bad, if you log them, say something like:

void Message::buildSilenceAck()
{
    string s;

    s.clear();
    ESP_LOGI(TAG, "params = %d",(int) m_params);
    ESP_LOGI(TAG, "m_params->nvs = %d",(int) m_params->nvs);
    s.append(m_params->nvs->getIpAddr(IP_ADDRESS));

[mzimmers]

@hskoglund I've done that. The pointers seem fine, and the getIpAddr() routine executes successfully.

Weird problem, I know.

[kshegunov]

@mzimmers said in std::string error:

This is very likely what the Linux world refers to as a segmentation fault.

More like kernel panic, looking at the dump. Ordinary segfaults are handled by the kernel and don't usually dump the CPU registers. Are you sure you have enough memory on that device? One'd observe a similar thing on desktop if swapping is disabled and there's a syscall that can't free up memory (or the memory is corrupt at the point of the system call). Funnily I currently get similar dumps, but that's because my CPU is buggy.

[kshegunov]

PS.
A quick look here (see error code 28) leads me to believe you have dereferencing of an invalid pointer (or a call to a function through an invalid address) due to EXCVADDR holding nonsense; so I'm back to my original assumption. It's going to be hard without debug info to trace this down, but could you try to build this application in debug mode so at least you can get a more human(e) backtrace?

[mzimmers]

@kshegunov The app is already built in debug. The reason the trace is so human-unfriendly is that I can't run monitor (a big part of my testing is connecting/disconnecting line power to the device, which entails removal of the USB cable that the monitor would run on), so I'm just logging what I can to the 2nd UART on the device. But, I've used xtensa-esp32-elf-addr2line to determine the line of source code, and it's definitely at the creation/assignment of the string in the message object.

[kshegunov]

Do you at least have logging?
I'd love to see the output of something along those lines (or equivalent):

std::cerr << uintptr_t(m_params);
std::cerr << uintptr_t(m_params->nvs);
char * p  = m_params->nvs->getIpAddr(IP_ADDRESS)
std::cerr << uintptr_t(p);

[mzimmers]

Yes, I can get those. I'll do it first thing Wednesday when I return to the office.

[mzimmers]

Well...this is kind of embarrassing (and a little odd, too). I put in the telltales that kshegunov suggested, and...m_params had a value of 0. That's the embarrassing part, that I didn't check that myself.

Here's the odd part: the reason the value was 0 was because when I created the Message object, I passed the incorrect object type to the c'tor. A Message object is supposed to be created like this:

Message::Message(Tasks *params)
{
    m_params = params;
}

But I was creating it like this:

MsgType Worker::processMsg(Message &msg)
{
    MsgType mt = msg.getType();
    switch (mt)
    {
    ...
    case MSG_SILENCE_BUZZER:
        msgOut = new Message(msg);
        msgOut->buildSilenceAck();
        break;
...

Sheer sloppiness on my part, but...why didn't the compiler yell at me?

Anyway, thanks to everyone who looked at this.

[kshegunov]

@mzimmers said in std::string error:

Sheer sloppiness on my part, but...why didn't the compiler yell at me?

C++ (and naturally C) is notorious for its implicit conversions and many compilers happily give you just enough rope to hang yourself; A reference is almost the same as a pointer to some object, same for an integer and by extension an enum. My advice is to (almost) always declare the constructor explicit so you don't get into that kind of trouble.
E.g.:

class Message
{
public
    explicit Message(Tasks *);
};

[mzimmers]

@kshegunov that's good advice, but this was more than just a reference/pointer mismatch: they were referring to/pointing to 2 different object types. I'd expect C++ to have been stricter about that...

[kshegunov]

they were referring to/pointing to 2 different object types. I'd expect C++ to have been stricter about that...

It'd depend on the compiler really, but from the machine's point of view it's all the same. Everything is/is converted to an address to some region in memory. Type-safety is something people invented to have at least some idea about what we are working with, and I agree, the compiler should've warned you at least. But, well, as I said almost everything decays to void * and from there the step into the abyss is just tiny ... ;)