Re: [PATCH] Fix a warning when initializing symbol_flags

[Saagar Jha <saagar@saagarjha.com>]

Ah, so I just realized that this file is actually compiled as C and not a C++ file in .c clothing, so even if this didn’t warn in Clang the patch was still incorrect as {} is an invalid initializer in C and GCC is correct in flagging it (though perhaps not for the right reasons). Newer versions of Clang do correctly identify { 0 } as the correct zero initialization idiom: https://github.com/llvm/llvm-project/commit/817a3bfcdd0bbac8d74fdfdb83a08484d8f63a30, although most distributions seem to carry a version older than this.

Eleven zeroes is, as far as I can tell, the only widely-compatible, guaranteed-to-work way to initialize this without warnings. memset would generally “work” but if we’re being pedantic I believe this has only been guaranteed as of a recent C standard (C11?) So I guess I’ll have to defer to whatever the decision is here on portability versus readability. (Third option: disable the warning around this code, perhaps after checking theg version.)

> On Sep 26, 2020, at 12:21, Hans-Peter Nilsson <hp@bitrange.com> wrote:
> 
> On Sat, 26 Sep 2020, Alan Modra via Binutils wrote:
> 
>> On Fri, Sep 25, 2020 at 05:42:24PM -0700, Saagar Jha wrote:
>>> Clang on certain platforms (such as arm64) will warn if a struct is
>>> partially initialized:
>>> 
>>> symbols.c:199:35: error: missing field 'written' initializer [-Werror,-Wmissing-field-initializers]
>>> 2895  symbol_entry_t needle = { { { 0 }, hash, name, 0, 0, 0 } };
>>> 
>>> If we use empty braces we have the same effect but avoid the warning.
>> 
>> But on gcc-4.9 this results in:
>> 
>> symbols.c: In function ?symbol_entry_find?:
>> symbols.c:199:3: error: missing initializer for field ?local_symbol? of ?struct symbol_flags? [-Werror=missing-field-initializers]
>>   symbol_entry_t needle = { { { }, hash, name, 0, 0, 0 } };
>>   ^
>> symbols.c:39:16: note: ?local_symbol? declared here
>>   unsigned int local_symbol : 1;
>>                ^
>> 
>> So I don't think we should apply your patch.
> 
> How about a memset 0 and setting .hash and .name?
> 
> I had to do that locally for an older gcc, that had a similar
> warning.  I prefered that over 11 (IIRC) "0," and having to
> keep them straight. :)
> 
> brgds, H-P