devlogs : C++ Adventures

I would like to both do more development work at work, and also do some more embedded development, and the answer to both is to get good at C++. To that end, here's my progress today:

    github.com/AlDanial/cloc v 1.90  T=0.01 s (712.3 files/s, 77556.5 lines/s)
    -------------------------------------------------------------------------------
    Language                     files          blank        comment           code
    -------------------------------------------------------------------------------
    C++                              5            168             17            618
    C/C++ Header                     4             51             21            105
    -------------------------------------------------------------------------------
    SUM:                             9            219             38            723
    -------------------------------------------------------------------------------

I mostly spent the time working on a dictionary system (key/value store) [1], but I managed to get some work done on the kimodem [2] too. This dictionary has the constraint that it needs to be able to be file-backed (at least, it needs to be able to be written to a file and read from a file). The ESP platform that I'm using for the modem uses SPIFFS, which I need to spend more time reading up on. Also, because the goal is for it to go on embedded systems, I'd like to have a constrained or fixed size for it so I understand how much memory it'll use.

Anyways, there's three layers this is built on, from bottom to top.

The arena, which is just a memory buffer with some basic management built in.
A TLV library built on the arena. Records can store at most 254 bytes (allowing for a trailing NUL byte).
A dictionary that uses pairs of TLV records to represent a key and value.

It has been a while since I wrote anything much, so I needed to dust off some of those systems programming skills. After a couple of segfaults and some brainos, I got it all working. Here's the dictionary test program running:

kyle@ono-sendai:~/code/klib$ ./dictionary_test
    TESTPROG: ./dictionary_test
    Arena @ 0x0x7ffeb8693a48
        Store is 128 bytes at address 0x0x7ffeb8693a48
        Type: mmap/file
    test Set foo->quux
        Set complete
    Dictionary KV pairs
        foo->quux
    test Set baz->quux
        Set complete
    Dictionary KV pairs
        foo->quux
        baz->quux
    test Set spam->xyzzx
        Set complete
    Dictionary KV pairs
        foo->quux
        baz->quux
        spam->xyzzx
    test overwriting key
    test Set baz->corvid
        Set complete
    Dictionary KV pairs
        foo->quux
        spam->xyzzx
        baz->corvid
        lookup
        compare records
        add new key to dictionary
    test Set quux->xyzzx
        Set complete
    Dictionary KV pairs
        foo->quux
        spam->xyzzx
        baz->corvid
        quux->xyzzx
    OK
    Dictionary KV pairs
        (NONE)

It needed a few minor tweaks to get it running on my ESP32 board (a MicroMod board [3] and carrier [4] from Sparkfun). Right now, the WiFi system is crashing, so that needs some investigation tomorrow.

Here's a few of the things I learned or worked through:

There's no guarantee that values will be set to 0. This was working on my Linux build, but completely broke on Windows. Mea culpa for using Windows, but it was a useful reminder.
I messed up my pointer math a few times.
The longest bug to debug (about 15 minutes) was an endless loop when deleting a record. Instead of tracking my progress in the arena, I kept trying to search from the beginning. I thought it was a memory corruption thing, but it turns out it was just changing a call from (eliding the error checking and whatnot):

cursor = FindTag(arena, NULL, record);

cursor = FindTag(arena, cursor, record);

The cursor variable controls how the library searches for things. If it's NULL, it searches from the beginning.

I was spoiled by having gdb; on the ESP32, I don't. So now I have to debug this:

Rebooting...
!�␘!�L␘B␇1�!�!���!y␌�J�MODEM BOOT
MODEM: TRY AUTOCONNECT
MODEM: CHECK HomeSSID_Guest
Guru Meditation Error: Core  1 panic'ed (LoadProhibited). Exception was unhandled.

Core  1 register dump:
PC      : 0x400d293c  PS      : 0x00060130  A0      : 0x800d2978  A1      : 0x3ffca3f0
A2      : 0x3ffc36b0  A3      : 0x00000002  A4      : 0x3ffca45c  A5      : 0x0000ff00
A6      : 0x00ff0000  A7      : 0xff000000  A8      : 0x00000000  A9      : 0x0000007f
A10     : 0x3ffca45c  A11     : 0x00000001  A12     : 0x00000200  A13     : 0x0000ff00
A14     : 0x00ff0000  A15     : 0xff000000  SAR     : 0x00000011  EXCCAUSE: 0x0000001c
EXCVADDR: 0x00000002  LBEG    : 0x40089881  LEND    : 0x40089891  LCOUNT  : 0xffffffff


Backtrace: 0x400d2939:0x3ffca3f0 0x400d2975:0x3ffca410 0x400d28f2:0x3ffca430 0x400d29e2:
0x3ffca450 0x400d2af9:0x3ffca580 0x400d2b8b:0x3ffca5b0 0x400d669a:0x3ffca5e0




ELF file SHA256: b472167c77126ae4

Rebooting...
%S9#�!��n�s����9␌�␌�
��MODEM BOOT

The line noise is because the code is using a baud rate of 9600 while the ESP firmware is sending at 115200. Something I can fix (without some hardware tweaks).

Anyways, it's late and I have work in the morning. I don't like leaving the board in a constant state of reboot, but I'm going to have to.

[1] https://git.wntrmute.dev/kyle/klib
[2] https://git.wntrmute.dev/kyle/kimodem
[3] https://www.sparkfun.com/products/16781
[4] https://www.sparkfun.com/products/16400

Tags: c++, esp32