Hi Antonin

On 30/10/2025 09:22, Antonin Delpeuch wrote:

On 29/10/2025 11:16, Phillip Wood wrote:

quoted

Unfortunately XDF_DIFF_ALGORITHM_MASK does not include 
XDF_NEED_MINIMAL so if the user has a config file that looks like

    [diff]
        algorithm = minimal
        algorithm = myers

We'll parse it as "minimal" rather than "myers"

As we need to reset the diff algorithm in a number of places I think 
it would be best to define a macro

        #define CLEAR_DIFF_ALGORITHM(flags) \
        flags &= ~(XDF_DIFF_ALGORITHM_MASK | XDF_NEED_MINIMAL)

Ouch, good catch! This problem is affecting other places as well.

I'm wondering if we couldn't even add XDF_NEED_MINIMAL to 
XDF_DIFF_ALGORITHM_MASK. I've reviewed all the places where 
XDF_DIFF_ALGORITHM_MASK is used, and it seems that in all cases it would 
either preserve the existing behaviour (potentially allowing us to 
remove an accompanying "DIFF_XDL_CLR(opts, NEED_MINIMAL);" macro which 
becomes redundant), or in some other cases it would fix a similar issue 
(for instance, in merge-file.c).

Thanks for taking the time to look at the other uses of 
XDF_DIFF_ALGORITHM_MASK. I think adding XDF_NEED_MINIMAL to 
XDF_DIFF_ALGORITHM_MASK is a good idea as I think we always treat 
"minimal" as another diff algorithm, rather than a variant of "myers" 
which is how it is implemented in xdiff.

Is your suggestion to introduce a new macro motivated by stability 
concerns? I'm aware that xdiff is used in other code bases as a library, 
so I guess changing XDF_DIFF_ALGORITHM_MASK can indeed be seen as a 
breaking change.

While it is a breaking change, tweaking XDF_DIFF_ALGORITHM_MASK seems 
like a better idea to me. I wondered about it yesterday but worried it 
would end up being too much work to review all the existing uses.

Thanks

Phillip

Antonin