accept() checks strings without copying them

json::accept() parses through json_sax_acceptor, which ignores the
values of strings and object keys, but the lexer still copied every
string token into its token buffer. That copy was 18 % of the
instructions of json::accept() on twitter.json (x86, GCC 16).

The lexer's flag for accept(), renamed from discard_number_values to
discard_values, now also selects a second instantiation of
scan_string() that performs the same checks (escapes, surrogates,
UTF-8, control characters, with the same error messages) without
writing to the token buffer. scan() chooses the instantiation once per
string: a run-time check inside the string loop made number-heavy
files up to 3 % slower on Apple M1.

json::accept() over 19 files, time after/before (geomean): 0.876 on x86
with GCC 16.2, 0.884 with Clang 23, 0.932 on Apple M1; up to 0.74 on
single files; files with few strings (canada, mesh, numbers) unchanged.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>
This commit is contained in:
Niels Lohmann committed 2026-10-31 00:07:57 +01:00
1 parent 755bb16d7a
commit 74f5ecbd35
5 files changed
+218 -90

No files matched your search

+2 -2
View File
@@ -184,11 +184,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
const bool allow_exceptions = true,
const bool ignore_comments = false,
const bool ignore_trailing_commas = false,
const bool discard_number_values = false
const bool discard_values = false
)
{
return ::nlohmann::detail::parser<basic_json, InputAdapterType>(std::move(adapter),
std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_number_values);
std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_values);
}
private: