does not compile

[mikrocoder]

Hi,

your lib looks very promising. Unfortunately there are big problems with it. If I use it with Arduino then the controller always resets.

If I use it in Atmel Studio 7 it always compiles with error messages. Your vector example does not compile either. It doesn't matter if I use an avr-gcc toolchain with gcc 9.3.0 or 11.2.0. Also in the Arduino IDE.

I can include vector without errors, but when I use it, it doesn't compile anymore.

Now I wanted to ask in which IDE do you program? In which environment did you test all this? According to your description it should work with avr-gcc.

2 Examples with Atmel Studio 7.

#include <vector>
#include <cstdio>
#include <algorithm>

int main()
{
    std::vector<uint8_t> test{10, 1, 2, 42, 3};
    test.push_back(4);
    test.erase(test.begin());

    if(auto it = std::find(test.begin(), test.end(), 42); it != test.end()) {
        test.erase(it);
    }
}

Output

Error undefined reference to `operator new(unsigned int)'   avr-libstdcpp-Test_001  <artificial>    1
Error <artificial>:(.text.startup+0x3e): undefined reference to `operator new(unsigned int)'    avr-libstdcpp-Test_001      1
Error <artificial>:(.text.startup+0x6c): undefined reference to `operator delete(void*, unsigned int)'  avr-libstdcpp-Test_001      1
Error <artificial>:(.text.startup+0xd2): undefined reference to `operator delete(void*, unsigned int)'  avr-libstdcpp-Test_001      1
Error C:\Users\Worker\AppData\Local\Temp\cc6G2oHd.ltrans0.ltrans.o: in function `main': avr-libstdcpp-Test_001      

---

#include <avr/io.h>
#include <vector>
#include <algorithm>

using namespace std;

vector<int16_t> v1 {1, 2, 3, 4};
vector<int16_t> v3 (19);

int main()
{  }

Output

Error undefined reference to `operator new(unsigned int)'   avr-libstdcpp-Test_001  <artificial>    1
Error <artificial>:(.text.startup+0x8a): undefined reference to `operator new(unsigned int)'    avr-libstdcpp-Test_001      1
Error C:\Users\Worker\AppData\Local\Temp\ccYoUbRD.ltrans0.ltrans.o: in function `std::_Vector_base<int, std::allocator<int> >::~_Vector_base()':    avr-libstdcpp-Test_001      1
Error C:\Users\Worker\AppData\Local\Temp\ccYoUbRD.ltrans0.ltrans.o: in function `_GLOBAL__sub_I_v1':    avr-libstdcpp-Test_001      1
Error ld returned 1 exit status avr-libstdcpp-Test_001  collect2.exe    0
Error undefined reference to `operator delete(void*, unsigned int)' avr-libstdcpp-Test_001  <artificial>    1

Translated with www.DeepL.com/Translator (free version)

[rleh]

Have taken a look at the vector example in the examples/ directory? You can't just copy the vector header file to you project, include it and expect it to work. There is a Makefile that specifies all compiler flags, include directories and source directories necessary for the avr-libstdc++ to work.

All examples get tested by our CI (Github Action) on every commit automatically and work fine: https://github.com/modm-io/avr-libstdcpp/runs/5250992849?check_suite_focus=true

[mikrocoder]

Hi,

I have now specified all the flags according to the make file in Atmel Studio. It still does not compile with the same errors.

About your objection. How else are you supposed to do it? That is the usual way. Make lib folder known, then include and apply.

You tested this only with the g++? But you write it would also work with the avr-gcc. Which it doesn't. Did you also test it with avr-gcc?

[rleh]

I have now specified all the flags according to the make file in Atmel Studio. It still does not compile with the same errors.

Then your build probably still misses the source files in $(LIB_DIR)/src/*.cc, see line 18 of the Makefile.

You tested this only with the g++? But you write it would also work with the avr-gcc. Which it doesn't. Did you also test it with avr-gcc?

[avr-]gcc/[avr-]g++ are the exact same compiler, except that a call to g++/avr-g++ enables C++ language support by default. You might have to pass -x c++ to avr-gcc command to switch on C++ language support.

Which exact compiler version are you using? Which environment (operation system, ...) are you using? Where did you get your avr-gcc compiler? Have you tried our precompiled avr-gcc v11.2.0? (Only for Linux, looks like you are using Windows.)

What additional compiler flags does you IDE(s) pass to the avr-gcc compiler? Have you tried compiling the unmodified vector example from this Repo using the Makefile (by just calling make in the example directory from shell)?

[rleh]

Looking at the error messages you posted in you first comment the problem seems easy: You are missing the operator new() and operator delete() implementations. In the vector example of this repository they come from examples/common/new.cpp. This file is added to the sources in line 18 of the Makefile ($(COMMON_DIR)/*.cpp), so you probably only have to copy that file (and maybe also cxxabi.cpp and uart.cpp) to you project, ensure the compiler compiles and links them for your error to disappear.

[mikrocoder]

Hi,

can we speak in german?

[CombiesGit]

I would like to thank you very much for the question.

And thanks for the replies.

So I managed to put this std++ implementation into operation in Arduino.

A few tests still need to be done. Then I would be happy to provide a short guide

[ckormanyos]

Looking at the error messages you posted in you first comment the problem seems easy: You are missing the operator new() and operator delete() implementations.

I just looked at this post and it's going in the right direction. Raphael (@rleh) is correct and on the right track for this issue.

Indeed, one of the tricky things about std::vector is that is uses dynamic menory (such as an allocator that reduces to using operators new/delete under the hood). So

If you look at the full template signature of std::vector, you will see that the second parameter is the allocator used for memory allocation.

Things you can try...

• Just to get an example running, you might try the advice from Raphael (@rleh) above.
• Sometimes it can be helpful to make a similar example with std::array.
• Write/use a custom allocator (this is a slightly advanced topic). This can be a great way to go on the metal.

I have examples for most of this stuff if you'd like to dive deeper into any of these ideas...

[mikrocoder]

Hello,

it was about how to get the avr-libstdcpp lib running in Atmel Studio on Windows. I was able to solve this with the help of another forum. If you are interested in this ...

You have to specify the path to the include directory in the project properties under directories.

Under Miscellaneous the usual: -std=c++20 -flto -Wno-volatile Under Symbol I enter F_CPU.

And in Solution Explorer you have to add the .cc and .cpp files.

After that you can start with the avr-libstdcpp lib.

If you want to show examples, feel free to do so. :-)

[ckormanyos]

it was about how to get the avr-libstdcpp lib running in Atmel Studio on Windows

Understood. Sorry, I had not understood that point. Yes. Simply add the include path on the compiler command line with the GUI (which boils down to the GCC command line -I include path syntax).

Not every compiler allows the specification of a path for its STL. GCC, however, odes a fine job on this for AVR and also for other toolchain targets such as ARM. So if you ever get a cool embedable STL, you can use it not only with AVR in Microchip's ATMEL Studio, but also with other targets on other environments.

I do have a related question: I notice you are mentioning the flag -std=c++20. Have you upgraded the GCC tollchain in your ATMEL Studio? Or does a more modern version or ATMEL Studio install a toolchain beyond GCC 5.4 AVR?

[ckormanyos]

If you want to show examples, feel free to do so

Hi @mikrocoder in another repository, I have some detailed benchmarks running on AVR. Some of them make rather heavy use of C++/STL.

For each benchmark, portd.3 measures the runtime on the scope.

You might enjoy the depth and scope of some of those benchmarks and their enclosing repo. In the root of theproject, there is a workspace for ATMEL Studio.

[mikrocoder]

Hi,

I'll have a look at your examples. Let's see if I understand them. Anyway, thanks for that.

AS always uses its outdated avr-gcc 5.4.0. There is nothing newer. I use my own toolchain with avr-gcc 11.2.0 with binutils 2.38. From the http://packs.download.atmel.com/ are added, see below ... the standard controllers are included anyway.

In Atmel Studio you still have to add the packages in Device Manager. If you want can upload my toolchain to dropbox. Can be included in Atmel Studio and also in Arduino IDE.

Zak Kemble https://blog.zakkemble.net/avr-gcc-builds/ also offers updated toolchains in larger intervals. Only his description for Arduino does not work.

atmega328pb

atmega808   
atmega809   
atmega1608      
atmega1609
atmega3208      
atmega3209  
atmega4808  
atmega4809  

attiny202   
attiny204   
attiny212   
attiny214   
attiny402   
attiny404   
attiny406   
attiny412   
attiny414   
attiny416   
attiny417   
attiny804   
attiny806   
attiny807   
attiny814   
attiny816   
attiny817   
attiny1604          
attiny1606          
attiny1607          
attiny1614
attiny1616
attiny1617
attiny3216
attiny3217
attiny1624
attiny1626
attiny1627
attiny3224
attiny3226
attiny3227

avr32da28
avr32da32
avr32da48
avr64da28
avr64da32
avr64da48
avr64da64
avr128da28
avr128da32
avr128da48
avr128da64

avr32db28
avr32db32
avr32db48
avr64db28
avr64db32
avr64db48
avr64db64
avr128db28
avr128db32
avr128db48
avr128db64

avr16dd14
avr16dd20
avr16dd28
avr16dd32
avr32dd14
avr32dd20
avr32dd28
avr32dd32
avr64dd14
avr64dd20
avr64dd28
avr64dd32

Translated with www.DeepL.com/Translator (free version)

[ckormanyos]

I use my own toolchain with avr-gcc 11.2.0 with binutils 2.38

That is a really good idea. I am glad to hear oyu did that. I'll get back to you if needed.

We are straying just a bit off topic, but for more completeness, one more link...

You might like my build of avr-gcc found here as a multi-file ZIP archive for windows (but you'll need a copy of libwinpthread-1.dll).

I solved some of the missing device file problems (but not all of them). I made a bunch of notes on building here.

There are some other target GCC builds there that might interest you as well.

[mikrocoder]

Hello,

there is a new problem. gcc 11.3.0 is out. In it there were changes in the math lib. Now sadly with avr-libstdcpp there is the problem of conflictings declarations and previous declarations. Could this be fixed at some point? My test environment is Windows 10 64bit, Microchip Studio 7 with my toolchains. With avr-gcc 11.2.0 everything was still ok.

Severity    Code    Description Project File    Line
Message     'float copysignf(float, float)' previously defined here avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  389
Message     'float fabsf(float)' previously defined here    avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  163
Error       conflicting declaration of C function 'bool isfinitef(float)'   avrLibStdCpp    C:\avrToolchain\avrLibStdCpp\include\cmath  144
Error       conflicting declaration of C function 'bool isfinitef(float)'   avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\math.cc   142
Error       conflicting declaration of C function 'bool isinff(float)'  avrLibStdCpp    C:\avrToolchain\avrLibStdCpp\include\cmath  143
Error       conflicting declaration of C function 'bool isinff(float)'  avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\math.cc   137
Error       conflicting declaration of C function 'bool isnanf(float)'  avrLibStdCpp    C:\avrToolchain\avrLibStdCpp\include\cmath  142
Error       conflicting declaration of C function 'bool isnanf(float)'  avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\math.cc   132
Message     previous declaration 'int isfinitef(float)' avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  365
Message     previous declaration 'int isfinitef(float)' avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  365
Message     previous declaration 'int isinff(float)'    avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  358
Message     previous declaration 'int isinff(float)'    avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  358
Message     previous declaration 'int isnanf(float)'    avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  348
Message     previous declaration 'int isnanf(float)'    avrLibStdCpp    c:\avrtoolchain\avr-gcc-11.3.0_mingw32_binutils2.38\avr\include\math.h  348
Error       recipe for target 'math.o' failed   avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\Release\Makefile  141
Error       redefinition of 'float copysignf(float, float)' avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\math.cc   147
Error       redefinition of 'float fabsf(float)'    avrLibStdCpp    C:\Users\Worker\Documents\Atmel Studio\7.0\WorkSpace_AVR128DB48\avrLibStdCpp\avrLibStdCpp\math.cc   32

[ckormanyos]

gcc 11.3.0 is out

Do you have a build of avr-gcc 11.3 that you're using?

Now sadly with avr-libstdcpp there is the problem of conflictings declarations and previous declarations. Could this be fixed at some point?

This topic is something i can look into. But it might take a bit of time to straighten out the float, double and long double C/C++ functions for both 32 and 64 bit. It is something I've looked into before, but not fully completed yet.

[mikrocoder]

Hi,

gcc 11.3.0 is out

Do you have a build of avr-gcc 11.3 that you're using?

Yes.

Now sadly with avr-libstdcpp there is the problem of conflictings declarations and previous declarations. Could this be fixed at some point?

This topic is something i can look into. But it might take a bit of time to straighten out the float, double and long double C/C++ functions for both 32 and 64 bit. It is something I've looked into before, but not fully completed yet.

Okay that would be very nice. It does not push. I have plenty of time.

[ckormanyos]

Do you have a build of avr-gcc 11.3 that you're using?

Yes.

Cool. I actually mean, do you have one available that we could use for testing? I can spin one up in a few hours, but if you've got one, that's easier...

[mikrocoder]

Hi,

I can provide. For Windows in .rar and .zip, each as he likes and for Linux. Whereby the Linux version is not tested. I use only the Windows version. I delete again in a few days. Link

[ckormanyos]

I can provide. For Windows in .rar and .zip, each as he likes and for Linux.

Awesome builds @mikrocoder. Many thanks.

OK, so here is the deal... Lines like this are actually incorrect. This is because the proper C-language signature of such functions returns int instead of bool.

I have not figured out all the inconsistencies yet. And I'm not sure if compiler option or complete repair is best (as these signatures originate in <math.h>.

So I have opened a new issue #18 to handle this. I'll try a few things there and report back when done.

[ckormanyos]

Can we now close this issue @mikrocoder (assuming we follow the latest point in #18) ?

Cc: @salkinium and @rleh and @chris-durand

[salkinium]

I've given you triage rights @ckormanyos, so that you can manage issues directly.

[ckormanyos]

The first point is finished:

When using this STL library with ATMEL Studio, two thngs are needed:

• Use ATMEL Studio with a compiler version higher than the installed version (such as get above GCC 8).
• Add (via -I compiler command line option) the include path of avr-libstdcpp to the command line.

The second point (<cmath> in cimbination with a new GCC 11.3) is being handled in #18.

[ckormanyos]

given you triage rights

Thank you Niklas (@salkinium).

[mikrocoder]

I think I forgot to thank you for your work. Better late than never. :-)