Reason for -fno-builtin introduced in SDK 7 ?

[marto]

Hello,

I was surprised that gcc did not optimized my strlen("FOO") construct (worse, in a code review, I argued that the compiler would do the optimization…)

It seems that -fno-builtin (in example Makefiles) prevent it to use __builtin_strlen() and compute the string size at compile-time.

This setting has been added between SDK6.1.0 and SDK7.0.0 but without mention in the release note. So my question is:

- why this has been added (what problem was "solved" by adding this flag)

- could I remove it to it to enjoy the gcc optimizations

- (i.e. what's your advise regarding this feature; should we activate it only for a few builtin…)

Thanks,

Marc.

[Kenneth]

Hi,

I can pass this along to the SDK team. But can you refer to a specific example in for instance nRF5 SDK v15.3 example that will compile differently depending on whether using S132 SD v6.1.1 vs S132 SD v7.0.1? Reason for asking is that there is no such release as SDK6.1.0 or SDK7.0.0.

Kenneth

[marto]

I'm referring to SDK found here: https://developer.nordicsemi.com/nRF5_SDK/ not to SoftDevices version.

I did not check the result w/o the `-fno-builtin`

[Kenneth]

I don't think anyone remember back 4 years why it was used, but there is no problem to remove it. I can find this information from Keil, maybe it was related to SDK team wanting to have control if printf could be redirected to uart, rtt, usb etc.:
http://www.keil.com/support/man/docs/armclang_ref/armclang_ref_edd1505401721650.htm

[marto]

Kenneth said:

I don't think anyone remember back 4 years why it was used,

No one want to/can remember anything for that long, that's why we use commit messages to put information out of brains :D. However we do not have access to your git (or whatever version control system you use).

I just hoped that the commit message would give a hint on such change.