Should GCC warn when __LINE__ is treated as %u %d %i?

Martin Sebor msebor@gmail.com
Tue Feb 19 15:21:00 GMT 2019


On 2/19/19 4:27 AM, Jonny Grant wrote:
> 
> 
> On 18/02/2019 22:03, Martin Sebor wrote:
>> On 2/17/19 7:12 AM, Jonny Grant wrote:
>>> Hello
>>>
>>> **Please keep my email address in any replies.
>>>
>>>
>>> 1) should this not give type conversion warnings?
>>>
>>> printf("lineoutput %d:%u:%i\n", __LINE__, __LINE__, __LINE__);
>>>
>>> 2) Should __LINE__ expand as an 'int', or 'unsigned int'?
>>>
>>> Had expected line 9 to expand as '9U', but -save-temps shows it ends 
>>> up as:
>>>
>>> printf("lineoutput %d:%u:%i\n", 9, 9, 9);
>>
>> The bigger problem with __LINE__ is that it need not expand to int
>> at all, such as in the following:
>>
>> $ cat u.c && gcc -O2 -S -Wall -Wextra u.c
>> #line 2147483647
>>
>> void f (void)
>> {
>>    __builtin_printf ("%i", __LINE__);
>> }
>> u.c: In function ‘f’:
>> u.c:-2147483646:21: warning: format ‘%i’ expects argument of type 
>> ‘int’, but argument 2 has type ‘long int’ [-Wformat=]
>>
>> Notice how the line number in the message doesn't correspond to
>> the sum of 2147483647 and 3 (the offset from the #line directive).
>>
>> Martin
> 
> Good point. Looks like a compiler limitation. Although I wonder how many 
> files really have that many lines, but anyway...
> 
> GCC could store the line number as a 'long int', then can display line 
> numbers up to 2^63
> 
> u.c: In function ‘f’:
> u.c:2147483650:21: warning: format ‘%i’ expects argument of type ‘int’, 
> but argument 2 has type ‘long int’ [-Wformat=]
> 
> I can file a on gcc.gnu.org bugzilla, or would you prefer to file yourself?

You're right that having that many physical lines is very unlikely.
The only way for this to be noticed that I can think of is in
conjunction with the #line directive explicitly setting very large
numbers (perhaps in generated code).  I'm not sure it's worth growing
the GCC internal tree node by 4 bytes but detecting the overflow and
issuing a diagnostic, maybe even an error, rather than allowing line
numbers to start to decrease, would be cheap and an improvement.
Either way, if you agree this is worth at least making a record of
in Bugzilla, please go ahead.

Martin



More information about the Gcc-help mailing list