Bug/regression in regex-replace (was fix for bug#67124)

Bug/regression in regex-replace (was fix for bug#67124)

[Campbell Barton]

Hi, the commit 47b497b4dac91e5ea56102018223bdeb5e21a93b by Stefan 
Monnier (an alternate fix for bug#67124) has broken the evil-numbers 
package [0], causing it's tests to fail.

 From a user perspective:
`evil-numbers/dec-at-pt` on `0` to result in `1` instead of `-1`.

The change seems to be caused by (set-match-data ...)
using passing in a pair of identical points (a match with zero length), 
then a call to regex-replace to insert text into the empty match (the 
numbers "sign" in this case).

I'm not sure if this would be considered a bug, or if creating an empty 
group is considers incorrect use of the API, in that case - invalid 
input might raise an error instead of failing silently.

Whatever the case this used to work - and the commit doesn't indicate 
the change was intentional.

[0]: https://github.com/juliapath/evil-numbers

Re: Bug/regression in regex-replace (was fix for bug#67124)

[ideasman42]

On 11/30/24 12:24 PM, Campbell Barton <ideasman42@gmail.com> wrote:
> Hi, the commit 47b497b4dac91e5ea56102018223bdeb5e21a93b by Stefan 
> Monnier (an alternate fix for bug#67124) has broken the evil-numbers 
> package [0], causing it's tests to fail.
> 
>  From a user perspective:
> `evil-numbers/dec-at-pt` on `0` to result in `1` instead of `-1`.
> 
> The change seems to be caused by (set-match-data ...)
> using passing in a pair of identical points (a match with zero length), 
> then a call to regex-replace to insert text into the empty match (the 
> numbers "sign" in this case).
> 
> I'm not sure if this would be considered a bug, or if creating an empty 
> group is considers incorrect use of the API, in that case - invalid 
> input might raise an error instead of failing silently.
> 
> Whatever the case this used to work - and the commit doesn't indicate 
> the change was intentional.
> 
> [0]: https://github.com/juliapath/evil-numbers

To follow up on this problem, this looks like a more typical bug since swapping the order,
so the empty match is replaced after the later, non-empty match still works,
so I assume it's preferred to handle this via the bug-tracker instead of the mailing list.