[Jbansal2]

Fixes #2155.

fd currently rejects any search path (positional argument or
--search-path) that is not a directory, with
Search path "..." is not a directory., even though the underlying
walker (ignore::WalkBuilder) already happily accepts a file as a
root and simply yields it as a single entry — fd's own CLI layer is
the only thing standing in the way.

This adds an opt-in --allow-search-files flag that relaxes the
check in Opts::search_paths() to also accept existing files (not
just directories), so a search path that is a file gets matched
against the pattern like any other search root:

$ fd '' a f.txt --allow-search-files
f.txt
a/c.txt

This mirrors the find behavior described in the issue. The flag is
opt-in (default behavior — rejecting non-directory search paths with
an error — is unchanged) to avoid silently changing existing scripts
that rely on the current error for typo detection.

Added test_allow_search_files to tests/tests.rs covering both the
existing error without the flag and the new behavior with it, plus a
CHANGELOG entry.

Note on verification

In my sandbox, cargo check/cargo build for this crate fails to
link several dependencies' build scripts (libc, crossbeam-utils,
proc-macro2, nix, ...) due to an unrelated, pre-existing MSVC
linker invocation problem in the environment, so I could not run
cargo test to completion locally. The change itself is small and
confined to Opts::search_paths()'s existing filter predicate (one
added || condition) plus the new flag definition, both following
the same shape as the neighboring one_file_system flag, so I'm
confident in it, but flagging this limitation for the reviewer.

[tmccombs]

Lot's of failing tests. I'm also not sure this is something we want.

[Jbansal2]

Thanks for taking a look, and sorry for the broken CI.

Found and fixed the actual bug: walk.rs unconditionally skips every depth-0 entry on the assumption it's always the root directory being traversed. With --allow-search-files, a root can now be a file, which has no children, so depth 0 is its only entry — the old code silently dropped it from the results even though it matched. Fixed by only skipping depth-0 when the entry is actually a directory, which leaves existing behavior unchanged.

My sandbox's MSVC linker can't link this crate's dependencies at all (unrelated environment issue, not this crate), so this time I built and ran the full suite under WSL instead of relying on an unverified diff:

cargo test:                                  123 passed; 0 failed
cargo clippy --all-targets -- -D warnings:   clean
cargo fmt --check:                           clean

On whether this is wanted at all: totally understand if the design itself isn't something you want to take on — happy to close this if so. If it's more about validating the design first, I'm glad to adjust (e.g. different flag name/semantics, or folding the file-root case into existing behavior without a flag) based on whatever direction you'd prefer.