528206 - Warn on unnecessary float->double conversion

Status: RESOLVED WONTFIX

(Developer Infrastructure :: Source Code Analysis, enhancement, P5)

Description

[Jeff Muizelaar [:jrmuizel]]

qcms has a bunch of code like:

float p;
...
p *= 65535.;

Since the type of 65535. is a double this code will promote p to a double to do the multiply. This causes the, likely unintended, conversion to double and conversion back to float.

It would be nice if we could warn about situations like this.

Comment 1

[Prarthana]

I am a student from SJCE, Mysore India and would like to take up this bug.

Comment 2

[Nathan Froyd [:froydnj]]

FWIW, GCC >= 4.6 already has an option to do this (-Wdouble-promotion).  The implementation was straightforward; I don't think it would be too hard to port it to clang, if it's not already there.

Comment 3

[(dormant account)]

Sounds like a good minor reason to bump gcc...if anyone feels bored :)

Comment 4

[Chris Peterson [:cpeterson]]

clang also supports the -Wdouble-promotion warning:

> warning: implicit conversion increases floating-point precision: A to B

https://clang.llvm.org/docs/DiagnosticsReference.html#wdouble-promotion

Comment 5

[Chris Peterson [:cpeterson]]

P5 because this warning is very noisy and, AFAIU, is a performance problem, not a correctness bug.

Comment 6

[Sylvestre Ledru [:Sylvestre]]

I don't think we will ever fix that, closing.