Re: gcc -Wconversion

[Richard Sandiford <richard.sandiford@arm.com>]

David Aldrich <david.aldrich.ntml@gmail.com> writes:
> Hi
>
> For our large 64-bit C++ project we use gcc compile options: -Wall -pedantic
>
> We also build the project with Visual C++ and get lots of instances of
> warning C4267, for example:
>
> 'initializing': conversion from 'size_t' to 'unsigned int', possible loss
> of data
>
> To get a similar warning with gcc it seems that I need -Wconversion.
> However, that warning option seems very strict. For example with this code:
>
> typedef struct
> {
>     uint8 x : 4,
>              y  : 4;
> } myHdr;
>
> unsigned X=1;
> myHdr hdr;
>
> hdr.x = static_cast<uint8_t>(X);
>
> I get warning:
>
> warning: conversion to ‘unsigned char:4’ from ‘uint8_t {aka unsigned char}’
> may alter its value [-Wconversion]

Yeah, it's a long-standing wart that the warning can't be disabled
for bitfields even when, like here, there's an explicit cast to the
underlying type (which is as close as an explicit cast can be).  See:

   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=39170

(warning: some of the initial responses were somewhat tetchy)

clang's -Wconversion doesn't warn for this case either, so it looks
like GCC is being stricter than both clang and Visual C++.

> I have two questions:
>
> 1) What static_cast would I use to fix the above example warning?

I don't think there is one, but you can use:

   hdr.x = X & 0xf;

to make the truncation explicit.  Neither GCC nor clang warn then.

I realise that might not be particularly desirable though.

> 2) Is -Wconversion recommended or is it too fussy in practice? Is there a
> better option?

-Wconversion is the right option to use.  Unfortunately there's
not yet any way of disabling or relaxing the warning for bitfields,
but the PR above is tracking that as a future enhancement.

Thanks,
Richard