next topic »

[Reply #15 on:]

You are correct that break is still a valid statement, and that this would be an ever greater problem:

Yep, that's a real problem, and potentially an easy trap for the unwary to fall into. Initially, I thought it was an unlikely bug to cause real problems, but now I think this is a shortcoming with potential to cause some real grief. I don't think it unreasonable or unlikely that someone would want to protect only a section of code within a "for" or "while" loop. If that section contains a break or continue, things certainly won't work as expected anymore if using ATOMIC_BLOCK macros.

In which case, I think Pyro's alternative is looking better all the time... I'll have to get around to having a play.

[Reply #16 on:]

The only caveat (as far as I can tell) is placement
...
The Libc macro does not have the same potential problem.

This is a problem indeed, but I believe the macros do suffer too as the atomic operation doesn't take effect until the macro is encountered. I think this is entirely up to the programmer to order their code carefuly, but worth a mention in the documnetation for sure.

@WizenedEE, that is a nice example of a 'break' failure.

Its good to see a few things being discussed weren't even something I had considered.

Now that I can consider the library mildly stable after doing some tests, I can start experimenting with some new ideas.
I have implemented some nice extensions coming soon, with entirely new ways of doing things for more flexibility.

[Reply #17 on:]

G'day all,

I have finished a major update to the library, it is more of an extension as the original functionality hasn't changed.
The library now supports in-line atomic operations on any kind of element* giving you greater control of how long interrupts are held. Some situations may warrant the atomic/critical section to be as short as possible, accessing 'safe' data may be better with short multiple accesses, compared to one long blocking operation. This creates a very easy interface to do so.

 For example ( explanations in updated reply #1 ):

• Atomic reads.
• Atomic writes.
• Atomic function calls.
• Atomic pointer manipulation.

You can download the new code from the first post [here]. The documentation has been updated there as well as a full explanation inside the AtomicBlock.h file. There is more documentation to come, I just wanted to get a quick run down of the new features.

This uses some high level C++ techniques to accomplish safe and efficient operation. If you would like explanations of any parts of the code feel free to ask.


*Object member functions are not currently supported, next update.

[Reply #18 on:]

As the new Due has come out, I have recently updated my AtomicBlock system to now support a few more platforms.

AtomicBlock V1.2 (beta)

Now supports:

AVR 8-bit processors

• Arduino 8-bit family ( Compiled / Tested )
• Teensy 2.0 ( Not Compiled / Not Tested )
• Teensy++ 2.0 ( Not Compiled / Not Tested )

ARM Cortex-M3

• Arduino DUE ( Compiled / Not Tested )
• Olimex STM32 / Leaflabs Maple. ( Compiled / Not Tested )

ARM Cortex-M4

• Teensy 3.0 ( Compiled / Not Tested )

Experimental:

AVR32

• Initial implementation. ( Not Compiled / Not Tested )
     ASF ( AVR Software Framework )
     Reference: http://asf.atmel.com/docs/2.11.1/index.html
  GNU Toolchain support will be available soon.

PIC32

• C++ compiler required.

The new code is posted on google code here: http://code.google.com/p/arduino-extensions/downloads/list
I have also uploaded the file to the original post [here].
   

[Reply #19 on:]

Standard atomic block also has some problems with different C standards as well, I actually made a new macro that works as a one-shot to avoid the issue instead. My application for it is to wrap malloc/free/realloc so that it is IRQ safe.

You can look at what I did here: https://github.com/xxxajk/xmem2/blob/master/malloc.c