Re: column numbers and ranges in error messages

[Per Bothner]

On 9/30/18 4:06 PM, Per Bothner wrote:
> Two related issues about section "4.4 Formatting Error Messages"
> (https://www.gnu.org/prep/standards/html_node/Errors.html):
>
> (1) "The error message can also give both the starting and ending positions of the 
> erroneous text."
>
> This is ambiguous wrt respect to whether end positions are exclusive or 
> inclusive.
> In most APIs I'm familiar that talk about ranges (such as a substring function), an 
> "end position"
> is exclusive (but is zero-based).  However, it seems clear to me that we want 
> *inclusive* ranges;
> one reason is that otherwise you would never have a duplicate line numbers.

I think I'm wrong here: Exclusive end positions are ok.  (I was thinking in 
terms
of a line range being delimited by an exclusive end line number, but that is 
not relevant.)

I haven't found real examples of code that emits

   sourcefile:line1.column1-line2.column2: message
or
   sourcefile:line1:column1-line2:column2: message

However, gcc's and clang's -fdiagnostics-parseable-fixits (and clang's 
-fdiagnostics-print-source-range-info)
options emit sources ranges with exclusive end-points.  However, the ranges are 
meant to be
machine-parsable, are inside braces, and count bytes rather than characters, so 
may not be
directly relevant.
--
        --Per Bothner
address@hidden   http://per.bothner.com/