Linus Björnstam writes:

> On Tue, 3 Aug 2021, at 00:50, Vivien Kraus via General Guile related discussions wrote:
>> (syntax-case
>>     (call-with-input-string "(a\r b)" read-syntax) ()
>>   ((a b)
>>    (values (syntax-source #'a) (syntax-source #'b))))
>> 
>> =>
>> 
>> $1 = ((line . 0) (column . 1))
>> $2 = ((line . 0) (column . 1))
>> 
>> This is obviously because of #\return.
>
> Well, if anything this has confused the guile reader. It is probably a
> bug: I suspect it resets the column because it anticipates a
> windows-style newline.

Maybe the correct behavior should be to entirely ignore #\return when
computing the line/column?

I see in libguile/ports.c (update_port_position) that #\return and
#\backspace may break the monotony of the position by strictly reducing
it.

I read here:
https://www.gnu.org/prep/standards/html_node/Errors.html#Errors

that the recommended solution is to use wcwidth, to give every non-ASCII
character at least a null width. Is there a reason not to use wcwidth?
If not, what do you think of the attached patch?

>  But are you often dealing with lonely "\r" in sources? Using either
>  \n (unix-style) or \r\n (window-style) will give you the correct
>  results.

Agreed. I don’t control the input, but I can silently drop all weird
characters (#\return, #\backspace and #\alarm) and almost noone [1] will
notice.

Vivien

[1] If it’s in a string literal, we can replace it with \r, \b or \a
anyway.