Hi Tobias!

On 2020-10-30T12:16:05+0100, Tobias Burnus <tobias@codesourcery.com> wrote:
> On 30.10.20 11:47, Thomas Schwinge wrote:
>>>> Fixed by introducing a new function; now one only needs to make sure
>>>> that no new code will re-introduce "lb->location":-)
>> ... another*existing instance*  of this problem.
> ...
>>   gfc_set_backend_locus (locus * loc)
>>   {
>>     gfc_current_backend_file = loc->lb->file;
>> -  input_location = loc->lb->location;
>> +  input_location = gfc_get_location (loc);
>>   }
>
> In bare usage, it seems to be fine – which are 23 callers.
>
> However, there is additionally:
>
> gfc_save_backend_locus (locus * loc)
> {
>    loc->lb = XCNEW (gfc_linebuf);
>    loc->lb->location = input_location;
>    loc->lb->file = gfc_current_backend_file;
> }
>
> which is used together with:
>
> gfc_restore_backend_locus (locus * loc)
> {
>    gfc_set_backend_locus (loc);
>    free (loc->lb);
> }
>
> I think the latter needs to be replaced by the previous
> version of "gfc_save_backend_locus" for two related reasons:
>
> * gfc_save_backend_locus operates with incomplete data,
>    i.e. loc->nextc (used by gfc_get_location) might not
>    be set.
> * input_location might/should already contain the column
>    offset – and you do not want to add some random offset
>    to it.
>
> Hence: LGTM – if you update 'gfc_restore_backend_locus'
> by inlining the previous version of 'gfc_set_backend_locus'.

Thanks for the review; absolutely right, sorry for not realizing that on
my own.

Thusly changed, see attached, pushed "Further improve Fortran column
location information [PR92793]" to master branch in commit
5677444f7e7ca15557030902c3d09dab4852fa90, and backported to
releases/gcc-10 branch in commit
a5c5f9e181c1f7548930f1cab91002b9d460cc92.


Grüße
 Thomas


-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Alexander Walter