28 KiB
Go on very small hardware (Part 2)
At the end of the first part of this article I promised to write something about interfaces . I don’t want to write here a complete or even brief lecture about the interfaces. Instead, I’ll show a simple example how to define and use an interface, and then, how to take advantage of ubiquitous io.Writer interface. There will also be a few words about reflection and semihosting .
Interfaces are a crucial part of Go language. If you want to learn more about them, I suggest to read Effective Go and Russ Cox article.
Concurrent Blinky – revisited
When you read the code of previous examples you probably noticed a counterintuitive way to turn the LED on or off. The Set method was used to turn the LED off and the Clear method was used to turn the LED on. This is due to driving the LEDs in open-drain configuration. What we can do to make the code less confusing? Let’s define the LED type with On and Off methods:
type LED struct {
pin gpio.Pin
}
func (led LED) On() {
led.pin.Clear()
}
func (led LED) Off() {
led.pin.Set()
}
Now we can simply call led.On() and led.Off() which no longer raises any doubts.
In all previous examples I tried to use the same open-drain configuration to don’t complicate the code. But in the last example, it would be easier for me to connect the third LED between GND and PA3 pins and configure PA3 in push-pull mode. The next example will use a LED connected this way.
But our new LED type doesn’t support the push-pull configuration. In fact, we should call it OpenDrainLED and define another PushPullLED type:
type PushPullLED struct {
pin gpio.Pin
}
func (led PushPullLED) On() {
led.pin.Set()
}
func (led PushPullLED) Off() {
led.pin.Clear()
}
Note, that both types have the same methods that work the same. It would be nice if the code that operates on LEDs could use both types, without paying attention to which one it uses at the moment. The interface type comes to help:
package main
import (
"delay"
"stm32/hal/gpio"
"stm32/hal/system"
"stm32/hal/system/timer/systick"
)
type LED interface {
On()
Off()
}
type PushPullLED struct{ pin gpio.Pin }
func (led PushPullLED) On() {
led.pin.Set()
}
func (led PushPullLED) Off() {
led.pin.Clear()
}
func MakePushPullLED(pin gpio.Pin) PushPullLED {
pin.Setup(&gpio.Config{Mode: gpio.Out, Driver: gpio.PushPull})
return PushPullLED{pin}
}
type OpenDrainLED struct{ pin gpio.Pin }
func (led OpenDrainLED) On() {
led.pin.Clear()
}
func (led OpenDrainLED) Off() {
led.pin.Set()
}
func MakeOpenDrainLED(pin gpio.Pin) OpenDrainLED {
pin.Setup(&gpio.Config{Mode: gpio.Out, Driver: gpio.OpenDrain})
return OpenDrainLED{pin}
}
var led1, led2 LED
func init() {
system.SetupPLL(8, 1, 48/8)
systick.Setup(2e6)
gpio.A.EnableClock(false)
led1 = MakeOpenDrainLED(gpio.A.Pin(4))
led2 = MakePushPullLED(gpio.A.Pin(3))
}
func blinky(led LED, period int) {
for {
led.On()
delay.Millisec(100)
led.Off()
delay.Millisec(period - 100)
}
}
func main() {
go blinky(led1, 500)
blinky(led2, 1000)
}
We’ve defined LED interface that has two methods: On and Off . The PushPullLED and OpenDrainLED types represent two ways of driving LEDs. We also defined two Make *LED functions which act as constructors. Both types implement the LED interface, so the values of these types can be assigned to the variables of type LED :
led1 = MakeOpenDrainLED(gpio.A.Pin(4))
led2 = MakePushPullLED(gpio.A.Pin(3))
In this case the assignability is checked at compile time. After the assignment the led1 variable contains OpenDrainLED{gpio.A.Pin(4)} and a pointer to the method set of the OpenDrainLED type. The led1.On() call roughly corresponds to the following C code:
led1.methods->On(led1.value)
As you can see, this is quite inexpensive abstraction if only consider the function call overhead.
But any assigment to an interface causes to include a lot of information about the assigned type. There can be a lot information in case of complex type which consists of many other types:
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
10356 196 212 10764 2a0c cortexm0.elf
If we don’t use reflection we can save some bytes by avoid to include the names of types and struct fields:
$ egc -nf -nt
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
10312 196 212 10720 29e0 cortexm0.elf
The resulted binary still contains some necessary information about types and full information about all exported methods (with names). This information is need for checking assignability at runtime, mainly when you assign one value stored in the interface variable to any other variable.
We can also remove type and field names from imported packages by recompiling them all:
$ cd $HOME/emgo
$ ./clean.sh
$ cd $HOME/firstemgo
$ egc -nf -nt
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
10272 196 212 10680 29b8 cortexm0.elf
Let’s load this program to see does it work as expected. This time we’ll use the st-flash command:
$ arm-none-eabi-objcopy -O binary cortexm0.elf cortexm0.bin
$ st-flash write cortexm0.bin 0x8000000
st-flash 1.4.0-33-gd76e3c7
2018-04-10T22:04:34 INFO usb.c: -- exit_dfu_mode
2018-04-10T22:04:34 INFO common.c: Loading device parameters....
2018-04-10T22:04:34 INFO common.c: Device connected is: F0 small device, id 0x10006444
2018-04-10T22:04:34 INFO common.c: SRAM size: 0x1000 bytes (4 KiB), Flash: 0x4000 bytes (16 KiB) in pages of 1024 bytes
2018-04-10T22:04:34 INFO common.c: Attempting to write 10468 (0x28e4) bytes to stm32 address: 134217728 (0x8000000)
Flash page at addr: 0x08002800 erased
2018-04-10T22:04:34 INFO common.c: Finished erasing 11 pages of 1024 (0x400) bytes
2018-04-10T22:04:34 INFO common.c: Starting Flash write for VL/F0/F3/F1_XL core id
2018-04-10T22:04:34 INFO flash_loader.c: Successfully loaded flash loader in sram
11/11 pages written
2018-04-10T22:04:35 INFO common.c: Starting verification of write complete
2018-04-10T22:04:35 INFO common.c: Flash written and verified! jolly good!
I didn’t connected the NRST signal to the programmer so the —reset option can’t be used and the reset button have to be pressed to run the program.
It seems that the st-flash works a bit unreliably with this board (often requires reseting the ST-LINK dongle). Additionally, the current version doesn’t issue the reset command over SWD (uses only NRST signal). The software reset isn’t realiable however it usually works and lack of it introduces inconvenience. For this board-programmer pair the OpenOCD works much better.
UART
UART (Universal Aynchronous Receiver-Transmitter) is still one of the most important peripherals of today’s microcontrollers. Its advantage is unique combination of the following properties:
-
relatively high speed,
-
only two signal lines (even one in case of half-duplex communication),
-
symmetry of roles,
-
synchronous in-band signaling about new data (start bit),
-
accurate timing inside transmitted word.
This causes that UART, originally intedned to transmit asynchronous messages consisting of 7-9 bit words, is also used to efficiently implement various other phisical protocols such as used by WS28xx LEDs or 1-wire devices.
However, we will use the UART in its usual role: to printing text messages from our program.
package main
import (
"io"
"rtos"
"stm32/hal/dma"
"stm32/hal/gpio"
"stm32/hal/irq"
"stm32/hal/system"
"stm32/hal/system/timer/systick"
"stm32/hal/usart"
)
var tts *usart.Driver
func init() {
system.SetupPLL(8, 1, 48/8)
systick.Setup(2e6)
gpio.A.EnableClock(true)
tx := gpio.A.Pin(9)
tx.Setup(&gpio.Config{Mode: gpio.Alt})
tx.SetAltFunc(gpio.USART1_AF1)
d := dma.DMA1
d.EnableClock(true)
tts = usart.NewDriver(usart.USART1, d.Channel(2, 0), nil, nil)
tts.Periph().EnableClock(true)
tts.Periph().SetBaudRate(115200)
tts.Periph().Enable()
tts.EnableTx()
rtos.IRQ(irq.USART1).Enable()
rtos.IRQ(irq.DMA1_Channel2_3).Enable()
}
func main() {
io.WriteString(tts, "Hello, World!\r\n")
}
func ttsISR() {
tts.ISR()
}
func ttsDMAISR() {
tts.TxDMAISR()
}
//c:__attribute__((section(".ISRs")))
var ISRs = [...]func(){
irq.USART1: ttsISR,
irq.DMA1_Channel2_3: ttsDMAISR,
}
You can find this code slightly complicated but for now there is no simpler UART driver in STM32 HAL (simple polling driver will be probably useful in some cases). The usart.Driver is efficient driver that uses DMA and interrupts to ofload the CPU.
STM32 USART peripheral provides traditional UART and its synchronous version. To use it as output we have to connect its Tx signal to the right GPIO pin:
tx.Setup(&gpio.Config{Mode: gpio.Alt})
tx.SetAltFunc(gpio.USART1_AF1)
The usart.Driver is configured in Tx-only mode (rxdma and rxbuf are set to nil):
tts = usart.NewDriver(usart.USART1, d.Channel(2, 0), nil, nil)
We use its WriteString method to print the famous sentence. Let’s clean everything and compile this program:
$ cd $HOME/emgo
$ ./clean.sh
$ cd $HOME/firstemgo
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
12728 236 176 13140 3354 cortexm0.elf
To see something you need an UART peripheral in your PC.
Do not use RS232 port or USB to RS232 converter!
The STM32 family uses 3.3 V logic but RS232 can produce from -15 V to +15 V which will probably demage your MCU. You need USB to UART converter that uses 3.3 V logic. Popular converters are based on FT232 or CP2102 chips.
You also need some terminal emulator program (I prefer picocom). Flash the new image, run the terminal emulator and press the reset button a few times:
$ openocd -d0 -f interface/stlink.cfg -f target/stm32f0x.cfg -c 'init; program cortexm0.elf; reset run; exit'
Open On-Chip Debugger 0.10.0+dev-00319-g8f1f912a (2018-03-07-19:20)
Licensed under GNU GPL v2
For bug reports, read
http://openocd.org/doc/doxygen/bugs.html
debug_level: 0
adapter speed: 1000 kHz
adapter_nsrst_delay: 100
none separate
adapter speed: 950 kHz
target halted due to debug-request, current mode: Thread
xPSR: 0xc1000000 pc: 0x080016f4 msp: 0x20000a20
adapter speed: 4000 kHz
** Programming Started **
auto erase enabled
target halted due to breakpoint, current mode: Thread
xPSR: 0x61000000 pc: 0x2000003a msp: 0x20000a20
wrote 13312 bytes from file cortexm0.elf in 1.020185s (12.743 KiB/s)
** Programming Finished **
adapter speed: 950 kHz
$
$ picocom -b 115200 /dev/ttyUSB0
picocom v3.1
port is : /dev/ttyUSB0
flowcontrol : none
baudrate is : 115200
parity is : none
databits are : 8
stopbits are : 1
escape is : C-a
local echo is : no
noinit is : no
noreset is : no
hangup is : no
nolock is : no
send_cmd is : sz -vv
receive_cmd is : rz -vv -E
imap is :
omap is :
emap is : crcrlf,delbs,
logfile is : none
initstring : none
exit_after is : not set
exit is : no
Type [C-a] [C-h] to see available commands
Terminal ready
Hello, World!
Hello, World!
Hello, World!
Every press of the reset button produces new “Hello, World!” line. Everything works as expected.
To see bi-directional UART code for this MCU check out this example.
io.Writer
The io.Writer interface is probably the second most commonly used interface type in Go, right after the error interface. Its definition looks like this:
type Writer interface {
Write(p []byte) (n int, err error)
}
usart.Driver implements io.Writer so we can replace:
tts.WriteString("Hello, World!\r\n")
with
io.WriteString(tts, "Hello, World!\r\n")
Additionally you need to add the io package to the import section.
The declaration of io.WriteString function looks as follows:
func WriteString(w Writer, s string) (n int, err error)
As you can see, the io.WriteString allows to write strings using any type that implements io.Writer interface. Internally it check does the underlying type has WriteString method and uses it instead of Write if available.
Let’s compile the modified program:
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
15456 320 248 16024 3e98 cortexm0.elf
As you can see, io.WriteString causes a significant increase in the size of the binary: 15776 - 12964 = 2812 bytes. There isn’t too much space left on the Flash. What caused such a drastic increase in size?
Using the command:
arm-none-eabi-nm --print-size --size-sort --radix=d cortexm0.elf
we can print all symbols ordered by its size for both cases. By filtering and analyzing the obtained data (awk, diff) we can find about 80 new symbols. The ten largest are:
> 00000062 T stm32$hal$usart$Driver$DisableRx
> 00000072 T stm32$hal$usart$Driver$RxDMAISR
> 00000076 T internal$Type$Implements
> 00000080 T stm32$hal$usart$Driver$EnableRx
> 00000084 t errors$New
> 00000096 R $8$stm32$hal$usart$Driver$$
> 00000100 T stm32$hal$usart$Error$Error
> 00000360 T io$WriteString
> 00000660 T stm32$hal$usart$Driver$Read
So, even though we don’t use the usart.Driver.Read method it was compiled in, same as DisableRx , RxDMAISR , EnableRx and other not mentioned above. Unfortunately, if you assign something to the interface, its full method set is required (with all dependences). This isn’t a problem for a large programs that use most of the methods anyway. But for our simple one it’s a huge burden.
We’re already close to the limits of our MCU but let’s try to print some numbers (you need to replace io package with strconv in import section):
func main() {
a := 12
b := -123
tts.WriteString("a = ")
strconv.WriteInt(tts, a, 10, 0, 0)
tts.WriteString("\r\n")
tts.WriteString("b = ")
strconv.WriteInt(tts, b, 10, 0, 0)
tts.WriteString("\r\n")
tts.WriteString("hex(a) = ")
strconv.WriteInt(tts, a, 16, 0, 0)
tts.WriteString("\r\n")
tts.WriteString("hex(b) = ")
strconv.WriteInt(tts, b, 16, 0, 0)
tts.WriteString("\r\n")
}
As in the case of io.WriteString function, the first argument of the strconv.WriteInt is of type io.Writer .
$ egc
/usr/local/arm/bin/arm-none-eabi-ld: /home/michal/firstemgo/cortexm0.elf section `.rodata' will not fit in region `Flash'
/usr/local/arm/bin/arm-none-eabi-ld: region `Flash' overflowed by 692 bytes
exit status 1
This time we’ve run out of space. Let’s try to slim down the information about types:
$ cd $HOME/emgo
$ ./clean.sh
$ cd $HOME/firstemgo
$ egc -nf -nt
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
15876 316 320 16512 4080 cortexm0.elf
It was close, but we fit. Let’s load and run this code:
a = 12
b = -123
hex(a) = c
hex(b) = -7b
The strconv package in Emgo is quite different from its archetype in Go. It is intended for direct use to write formatted numbers and in many cases can replace heavy fmt package. That’s why the function names start with Write instead of Format and have additional two parameters. Below is an example of their use:
func main() {
b := -123
strconv.WriteInt(tts, b, 10, 0, 0)
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, 6, ' ')
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, 6, '0')
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, 6, '.')
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, -6, ' ')
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, -6, '0')
tts.WriteString("\r\n")
strconv.WriteInt(tts, b, 10, -6, '.')
tts.WriteString("\r\n")
}
There is its output:
-123
-123
-00123
..-123
-123
-123
-123..
Unix streams and Morse code
Thanks to the fact that most of the functions that write something use io.Writer instead of concrete type (eg. FILE in C) we get a functionality similar to Unix streams . In Unix we can easily combine simple commands to perform larger tasks. For example, we can write text to the file this way:
echo "Hello, World!" > file.txt
The > operator writes the output stream of the preceding command to the file. There is also |operator that connects output and input streams of adjacent commands.
Thanks to the streams we can easily convert/filter output of any command. For example, to convert all letters to uppercase we can filter the echo’s output through tr command:
echo "Hello, World!" | tr a-z A-Z > file.txt
To show the analogy between io.Writer and Unix streams let’s write our:
io.WriteString(tts, "Hello, World!\r\n")
in the following pseudo-unix form:
io.WriteString "Hello, World!" | usart.Driver usart.USART1
The next example will show how to do this:
io.WriteString "Hello, World!" | MorseWriter | usart.Driver usart.USART1
Let’s create a simple encoder that encodes the text written to it using Morse coding:
type MorseWriter struct {
W io.Writer
}
func (w *MorseWriter) Write(s []byte) (int, error) {
var buf [8]byte
for n, c := range s {
switch {
case c == '\n':
c = ' ' // Replace new lines with spaces.
case 'a' <= c && c <= 'z':
c -= 'a' - 'A' // Convert to upper case.
}
if c < ' ' || 'Z' < c {
continue // c is outside ASCII [' ', 'Z']
}
var symbol morseSymbol
if c == ' ' {
symbol.length = 1
buf[0] = ' '
} else {
symbol = morseSymbols[c-'!']
for i := uint(0); i < uint(symbol.length); i++ {
if (symbol.code>>i)&1 != 0 {
buf[i] = '-'
} else {
buf[i] = '.'
}
}
}
buf[symbol.length] = ' '
if _, err := w.W.Write(buf[:symbol.length+1]); err != nil {
return n, err
}
}
return len(s), nil
}
type morseSymbol struct {
code, length byte
}
//emgo:const
var morseSymbols = [...]morseSymbol{
{1<<0 | 1<<1 | 1<<2, 4}, // ! ---.
{1<<1 | 1<<4, 6}, // " .-..-.
{}, // #
{1<<3 | 1<<6, 7}, // $ ...-..-
// Some code omitted...
{1<<0 | 1<<3, 4}, // X -..-
{1<<0 | 1<<2 | 1<<3, 4}, // Y -.--
{1<<0 | 1<<1, 4}, // Z --..
}
You can find the full morseSymbols array here. The //emgo:const directive ensures that morseSymbols array won’t be copied to the RAM.
Now we can print our sentence in two ways:
func main() {
s := "Hello, World!\r\n"
mw := &MorseWriter{tts}
io.WriteString(tts, s)
io.WriteString(mw, s)
}
We use the pointer to the MorseWriter &MorseWriter{tts} instead os simple MorseWriter{tts} value beacuse the MorseWriter is to big to fit into an interface variable.
Emgo, unlike Go, doesn’t dynamically allocate memory for value stored in interface variable. The interface type has limited size, equal to the size of three pointers (to fit slice ) or two float64 (to fit complex128 ), what is bigger. It can directly store values of all basic types and small structs/arrays but for bigger values you must use pointers.
Let’s compile this code and see its output:
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
15152 324 248 15724 3d6c cortexm0.elf
Hello, World!
.... . .-.. .-.. --- --..-- .-- --- .-. .-.. -.. ---.
The Ultimate Blinky
The Blinky is hardware equivalent of Hello, World! program. Once we have a Morse encoder we can easly combine both to obtain the Ultimate Blinky program:
package main
import (
"delay"
"io"
"stm32/hal/gpio"
"stm32/hal/system"
"stm32/hal/system/timer/systick"
)
var led gpio.Pin
func init() {
system.SetupPLL(8, 1, 48/8)
systick.Setup(2e6)
gpio.A.EnableClock(false)
led = gpio.A.Pin(4)
cfg := gpio.Config{Mode: gpio.Out, Driver: gpio.OpenDrain, Speed: gpio.Low}
led.Setup(&cfg)
}
type Telegraph struct {
Pin gpio.Pin
Dotms int // Dot length [ms]
}
func (t Telegraph) Write(s []byte) (int, error) {
for _, c := range s {
switch c {
case '.':
t.Pin.Clear()
delay.Millisec(t.Dotms)
t.Pin.Set()
delay.Millisec(t.Dotms)
case '-':
t.Pin.Clear()
delay.Millisec(3 * t.Dotms)
t.Pin.Set()
delay.Millisec(t.Dotms)
case ' ':
delay.Millisec(3 * t.Dotms)
}
}
return len(s), nil
}
func main() {
telegraph := &MorseWriter{Telegraph{led, 100}}
for {
io.WriteString(telegraph, "Hello, World! ")
}
}
// Some code omitted...
In the above example I omitted the definition of MorseWriter type because it was shown earlier. The full version is available here. Let’s compile it and run:
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
11772 244 244 12260 2fe4 cortexm0.elf
Reflection
Yes, Emgo supports reflection. The reflect package isn’t complete yet but that what is done is enough to implement fmt.Print family of functions. Let’s see what can we do on our small MCU.
To reduce memory usage we will use semihosting as standard output. For convenience, we also write simple println function which to some extent mimics fmt.Println .
package main
import (
"debug/semihosting"
"reflect"
"strconv"
"stm32/hal/system"
"stm32/hal/system/timer/systick"
)
var stdout semihosting.File
func init() {
system.SetupPLL(8, 1, 48/8)
systick.Setup(2e6)
var err error
stdout, err = semihosting.OpenFile(":tt", semihosting.W)
for err != nil {
}
}
type stringer interface {
String() string
}
func println(args ...interface{}) {
for i, a := range args {
if i > 0 {
stdout.WriteString(" ")
}
switch v := a.(type) {
case string:
stdout.WriteString(v)
case int:
strconv.WriteInt(stdout, v, 10, 0, 0)
case bool:
strconv.WriteBool(stdout, v, 't', 0, 0)
case stringer:
stdout.WriteString(v.String())
default:
stdout.WriteString("%unknown")
}
}
stdout.WriteString("\r\n")
}
type S struct {
A int
B bool
}
func main() {
p := &S{-123, true}
v := reflect.ValueOf(p)
println("kind(p) =", v.Kind())
println("kind(*p) =", v.Elem().Kind())
println("type(*p) =", v.Elem().Type())
v = v.Elem()
println("*p = {")
for i := 0; i < v.NumField(); i++ {
ft := v.Type().Field(i)
fv := v.Field(i)
println(" ", ft.Name(), ":", fv.Interface())
}
println("}")
}
The semihosting.OpenFile function allows to open/create file on the host side. The special path :tt corresponds to host’s standard output.
The println function accepts arbitrary number of arguments, each of arbitrary type:
func println(args ...interface{})
It’s possible because any type implements the empty interface interface{} . The println uses type switch to print strings, integers and booleans:
switch v := a.(type) {
case string:
stdout.WriteString(v)
case int:
strconv.WriteInt(stdout, v, 10, 0, 0)
case bool:
strconv.WriteBool(stdout, v, 't', 0, 0)
case stringer:
stdout.WriteString(v.String())
default:
stdout.WriteString("%unknown")
}
Additionally it supports any type that implements stringer interface, that is, any type that has String() method. In any case clause the v variable has the right type, same as listed after case keyword.
The reflect.ValueOf(p) returns p in the form that allows to analyze its type and content programmatically. As you can see, we can even dereference pointers using v.Elem() and print all struct fields with their names.
Let’s try to compile this code. For now let’s see what will come out if compiled without type and field names:
$ egc -nt -nf
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
16028 216 312 16556 40ac cortexm0.elf
Only 140 free bytes left on the Flash. Let’s load it using OpenOCD with semihosting enabled:
$ openocd -d0 -f interface/stlink.cfg -f target/stm32f0x.cfg -c 'init; program cortexm0.elf; arm semihosting enable; reset run'
Open On-Chip Debugger 0.10.0+dev-00319-g8f1f912a (2018-03-07-19:20)
Licensed under GNU GPL v2
For bug reports, read
http://openocd.org/doc/doxygen/bugs.html
debug_level: 0
adapter speed: 1000 kHz
adapter_nsrst_delay: 100
none separate
adapter speed: 950 kHz
target halted due to debug-request, current mode: Thread
xPSR: 0xc1000000 pc: 0x08002338 msp: 0x20000a20
adapter speed: 4000 kHz
** Programming Started **
auto erase enabled
target halted due to breakpoint, current mode: Thread
xPSR: 0x61000000 pc: 0x2000003a msp: 0x20000a20
wrote 16384 bytes from file cortexm0.elf in 0.700133s (22.853 KiB/s)
** Programming Finished **
semihosting is enabled
adapter speed: 950 kHz
kind(p) = ptr
kind(*p) = struct
type(*p) =
*p = {
X. : -123
X. : true
}
If you’ve actually run this code, you noticed that semihosting is slow, especially if you write a byte after byte (buffering helps).
As you can see, there is no type name for *p and all struct fields have the same X. name. Let’s compile this program again, this time without -nt -nf options:
$ egc
$ arm-none-eabi-size cortexm0.elf
text data bss dec hex filename
16052 216 312 16580 40c4 cortexm0.elf
Now the type and field names have been included but only these defined in main.go file main package. The output of our program looks as follows:
kind(p) = ptr
kind(*p) = struct
type(*p) = S
*p = {
A : -123
B : true
}
Reflection is a crucial part of any easy to use serialization library and serialization algorithms like JSONgain in importance in the IOT era.
This is where I finish the second part of this article. I think there is a chance for the third part, more entertaining, where we connect to this board various interesting devices. If this board won’t carry them, we replace it with something a little bigger.
via: https://ziutek.github.io/2018/04/14/go_on_very_small_hardware2.html
作者：Michał Derkacz 译者：译者ID 校对：校对者ID