Jeff King writes: > There's some small value in leaving > %X alone if we do not understand "X" (not to mention the backwards > %compatibility you mentioned), but I think %() is a pretty > deliberate indication that a placeholder was meant there. Good point. > We already do this for ref-filter expansions: > > $ git for-each-ref --format='%(foo)' > fatal: unknown field name: foo > > We don't for "--pretty" formats, but I do wonder if anybody would be > really mad (after all, we have declared ourselves free to add new > placeholders, so such formats are not future-proof). Oh my. I wasn't aware that there was a totally separate string interpolation implementation used for ref filters. That one has separated parsing, making it more amenable to good error handling. I wonder if that could be generalized and reused for pretty formats. However I doubt I will have time to dig deeper into that in near time.
