svn commit: r303146 - head/usr.bin/sed

[Pedro Giffuni]

On 07/22/16 02:13, Bruce Evans wrote:
> On Fri, 22 Jul 2016, Alexey Dokuchaev wrote:
>
>>> Log:
>>>   sed(1): Appease older GCC.
>
> "Appease" actually seems to be the correct wording here since gcc's
> detection
> of a variable that might be used unitialized seems to report one that is
> not used uninitialized.
>
>> Isn't it also being dictated by style(9) and common sense? :)
>
> You missed that this combines a style fix in previous gcc appeasement
> (or just excessive paranoia) in one variable with appeasement for
> another variable, since copying the previous appeasement would copy
> its style bug.  The 2 variables are used in exactly the same limited
> way.
>

Yes, the first one is just a style fix while I was there.

The oldpsanl bogusness was breaking the build with gcc42. gcc48+, clang
and coverity all agree it was a false positive.
It was likely a side effect of raising the WARNS level to 5.

>>> Modified:
>>>   head/usr.bin/sed/process.c
>>>
>>> @@ -97,11 +97,12 @@ process(void)
>>>  {
>>>      struct s_command *cp;
>>>      SPACE tspace;
>>> -    size_t oldpsl = 0;
>>> +    size_t oldpsl;
>>>      char *p;
>>>      int oldpsanl;
>>>
>>>      p = NULL;
>>> +    oldpsanl = oldpsl = 0;
>
> Multiple assignments on a single line is not very good style and is
> probably
> not KNF.  Here it is further from being good style since the variables
> have different types.  Since both types are integral and the value is 0
> the implicit type conversions don't change the value.  However, compilers
> should warn about down-converting a size_t to an int unless they do the
> analysis that this is safe because the value in the size_t is known to
> fit in the int.
>

Doing the multiple assignment seemed natural and readable as both are
(perhaps equally bogus) initializations.

I did notice the different types after committing. Perhaps swapping the
assignment would have been preferable? I suspect the compiler manages to 
optimize out the casting.

Pedro.