fix: expand format placeholder followed by a literal '}'

[upuddu]

A format placeholder immediately followed by a literal } wasn't expanded: fd --format '{}}' emitted a literal {} and dropped the trailing brace instead of producing <path>}. The parser fell through to a literal branch whenever a placeholder was followed by }; removing it lets the } be treated as ordinary literal text like everywhere else. Since --exec shares the parser, fd -x echo '{.}}' was affected too.

[tmccombs]

What you describe is expected behavior. Currently, "}}" is the way to escape a "}" character, so the "}}" is replaced witha literal "}", and the "{" is unmatched, so not replaced.

I haven't looked at this deeply, but I am guessing this is effectively changing the eacape syntax, which would be a breaking change. Such a change should probably have more discussion in an issue before it is merged.

[upuddu]

Thanks for taking a look, and fair point, I don't want to quietly change the escaping rules either.

Digging into the parser, I think the change is narrower than "changing the escape syntax." {{ → { and }} → } are handled exactly as before, and the documented way to emit a literal {} ({{}}) is unchanged. The only strings whose output changes are a placeholder immediately followed by a single }: {}}, {/}}, {.}}, etc.

The reason I filed it as a bug is consistency of the lone }. A single, un-doubled } is already treated as ordinary literal text everywhere else in a format string (e.g. --format 'a}b' emits a}b). The one exception was directly after a placeholder, where the trailing } silently swallowed the placeholder, so {}} produced the literal text {} instead of <path>}. Removing that special case is what makes a lone } behave the same in that position as everywhere else.

That said, you're right that it's technically a behavior change for {}} specifically, and I'm happy to follow your process on it. Want me to open an issue laying this out (bug-vs-breaking-change, the consistency argument, and the fact that the {{}} escape is unaffected) so it can be decided there before merging? Happy to hold the PR until then.

[tavianator]

I think the sane thing to do here is maximum munch tokenization, same as most lexers/parsers do. Although IMO {.}} should be a syntax error, you should be forced to write {.}}}