drivers: sensor: lsm6dl fixing -Wdouble-promotion warning

[FRASTM]

Fixing this "revealed" bug which got introduced when a PR added the -Wdouble-promotion flag to GCC builds
Based on the recent PR #68237

Fixes #68284

[yperess]

Is this really what we want? Shouldn't the fix be to change SENSI_GRAIN_XL to a float instead of a double by adding f to the constant:

#define SENSI_GRAIN_XL (61 / 1000.0f)

I think it's better than to force using doubles here.

[avisconti]

Is this really what we want? Shouldn't the fix be to change SENSI_GRAIN_XL to a float instead of a double by adding f to the constant:

#define SENSI_GRAIN_XL (61 / 1000.0f)

I think it's better than to force using doubles here.

I tend to agree.

[FRASTM]

Is this really what we want? Shouldn't the fix be to change SENSI_GRAIN_XL to a float instead of a double by adding f to the constant:

#define SENSI_GRAIN_XL (61 / 1000.0f)

I think it's better than to force using doubles here.

I tend to agree.

Sure but I do not see this is enough. Warning messages remain

./samples/sensor/lsm6dsl/src/main.c:150:59: warning: implicit conversion from 'float' to 'double' when passing argument to function [-Wdouble-promotion]
  150 |                                                           out_ev(&accel_x_out),

[avisconti]

I approved here, as it is the simplest solution. Nevertheless I guess that all that sensitivity values there could be expressed as ug/LSB using just a uint32_t. Obviously all the "GRAINS" definition should be expressed in ug/LSB as well (e.g. #define SENSI_GRAIN_XL (61LL)) and the conversion to mg (or mdps and so on) should be done in the correspondent get_channel() routine. What do you think?

[erwango]

@yperess The problem is that this warning shows up un CI in other PRs and hence is blocking them. I suggest approving this one and raising an issue for improvement (or a bug as you wish).
This can be fixed later on. Issue was there initially anyway..

I know that this is exactly why warning are there, but point is now detected. It doesn't have to be blocking.

[yperess]

Is this really what we want? Shouldn't the fix be to change SENSI_GRAIN_XL to a float instead of a double by adding f to the constant:

#define SENSI_GRAIN_XL (61 / 1000.0f)

I think it's better than to force using doubles here.

I tend to agree.

Sure but I do not see this is enough. Warning messages remain

./samples/sensor/lsm6dsl/src/main.c:150:59: warning: implicit conversion from 'float' to 'double' when passing argument to function [-Wdouble-promotion]
  150 |                                                           out_ev(&accel_x_out),

I'm sorry, I need a little more detail, out_ev is defined:

static inline float out_ev(struct sensor_value *val)
{
	return (val->val1 + (float)val->val2 / 1000000);
}

In this case val1 and val2 are int32_t types. I don't see how this PR solves this issue in samples/sensor/lsm6dsl/src/main.c unless it's trying to convert 1000000 to a double in which case you can just add .0f.

In my experience when we cover up these warnings, nobody ever goes back and fixes them. I would really prefer to just see a proper solution instead of moving to doubles which are much more costly.

[avisconti]

@erwango @FRASTM @yperess
I created the PR #68549 which does not touch the driver and should remove the warning. Would you please take a look at it? It may replace this PR.

[henrikbrixandersen]

Can this be closed now that #68549 was merged?

[avisconti]

Can this be closed now that #68549 was merged?

I guess so. It seems to me that the blocked PR have eventually been merged

[erwango]

Closing