Handling no RTOS is done wrong

[embeddedteam103]

Description

This is a very annoying little bug:
In certain in mbed there is conditioning on whether rtos is present or not (or any other module). The problem is that in many place the conditional to check for it is wrong (which causes a compile time error).
The check should be

#if MBED_CONF_RTOS_PRESENT
....
#endif

But in many places it is defined as #ifdef or #if defined() which give the wrong result.

The way to remove rtos or anything else optional in mbed (the way I understand it, maybe I'm wrong) is to have an mbed_app.json file and have there module-name : {present: "0"} that overrides the defaulted "1" value.
When we do this - the macro MBED_CONF_RTOS_PRESENT is defined to be zero and therefore the tests for it should be the value and not if it is defined.
I see https://github.com/ARMmbed/mbed-os/pull/8957/files which encounters a similar problem.

Some of the files affected (we used a number of regex searches in mbed 5.8.3 with some modules missing to produce the list, so it might not be complete):

• events\mbed_shared_queues.h
• events\mbed_shared_queues.cpp
• events\equeue\equeue_platform.h
• events\equeue\equeue_mbed.cpp
• platform\mbed_retarget.cpp
• platform\mbed_sdk_boot.cpp
• platform\Platformmutex.h
• platform\SingletonPtr.h
• platform\mbed_poll.cpp
• platform\mbed_wait_api_rtos.cpp

Issue request type

[ ] Question
[ ] Enhancement
[x] Bug

[kjbracey]

I think you can override with present: null - I think a null will override a non-null fine, and JSON null leads to the value not being defined.

But in general you're right, tests on MBED_CONF should normally be #if on the assumption that they will be either true or false . (null would lead to not defined which #if treats as 0)

(JSON true and false are also now permitted rather than 0 or 1 - this wasn't the case in early versions of the config tools).

[0xc0170]

But in general you're right, tests on MBED_CONF should normally be #if on the assumption that they will be either true or false . (null would lead to not defined which #if treats as 0)

👍

@ARMmbed/mbed-os-tools Please review . As mentioned, this touches as well 8957. This ifdef vs if confuses people. I can't spot this documented in https://os.mbed.com/docs/latest/reference/configuration.html, or is it documented elsewhere?

[kjbracey]

The way to remove rtos or anything else optional in mbed (the way I understand it, maybe I'm wrong) is to have an mbed_app.json file and have there module-name : {present: "0"} that overrides the defaulted "1" value.

That isn't actually the direction it was intended to work, I believe - it was really a detection mechanism for the case where someone had removed the directory. It was for code that needs to work on both Mbed OS 2 and Mbed OS 5, rather than an Mbed OS 5 config mechanism.

But I guess it can work in the way you describe - if you undefine the value, no-one will call it, so the linker will end up excluding it.

Code which is intended to be excluded at compile time is flagged via feature or component or target switch gating. The RTOS is not currently officially excludable for Mbed OS 5, but a lot of people do it unofficially by having an .mbedignore file that skips the directory.

(Build config wasn't the intent of mbedignore - it was to ignore files totally irrelevant to mbed OS like other OS port directories, so people didn't have to restructure their upstream repos if going in as a subtree).

[bridadan]

@kjbracey-arm is correct, setting the MBED_CONF_RTOS_PRESENT macro (or the equivalent configuration option) is not the intended way of removing the RTOS. It is possible to build Mbed OS without the RTOS by adding the rtos directory to your .mbedignore file, as was previously stated.

@0xc0170 I'll review on #8957.

[embeddedteam103]

Just a comment: I think the reason this was left undetected/unchanged was the fact that adding rtos to .mbedignore is the semi-official way to disable RTOS (which avoids the problem).

[0xc0170]

@embeddedteam103 Is there anything left here for building without RTOS? We will look at how to improve it and make it official way.

[deepikabhavnani]

Just a comment: I think the reason this was left undetected/unchanged was the fact that adding rtos to .mbedignore is the semi-official way to disable RTOS (which avoids the problem).

Partially because of .mbedignore and partially because few components and features do not have RTOS dependency handled fully.

[embeddedteam103]

@embeddedteam103 Is there anything left here for building without RTOS? We will look at how to improve it and make it official way.

Actually we want much more :)
I want to be able to replace (or remove) the cmsis implementation layer using the mbed-cli or some similar tool, just like we can replace targets using the custom_targets.json mechanism.
See #9139 for the details.
We are willing to help here.

[bmcdonnell-ionx]

Maybe relevant:
https://github.com/janjongboom/mbed-os-example-blinky-no-rtos

[0xc0170]

@embeddedteam103 How this changes with baremetal profile introduction : https://os.mbed.com/docs/mbed-os/v5.13/reference/mbed-os-bare-metal.html ?