Compare commits

..
Author SHA1 Message Date
Niels Lohmann d027f06a42 Make serializer's indent_string lazily allocated
The serializer constructor unconditionally allocated a 512-byte
indent_string, even though it is only ever read inside the
pretty_print branches of dump(). This wasted a heap allocation (and
its matching deallocation) on every compact (i.e. default, non-pretty)
dump() call.

indent_string is now default-constructed empty and lazily grown to
512 bytes, filled with indent_char, the first time a pretty-print
branch actually needs it. The existing doubling/growth logic for
larger indents is otherwise untouched, so output remains byte-identical
to before -- including in the pre-existing edge case where growth
beyond the initial buffer fills with ' ' instead of indent_char
(tracked separately by open PR #5186, which is left alone here).

The second, larger optimization mentioned in #5413 (removing the
shared_ptr-based output adapter) is intentionally out of scope, as it
overlaps open PR #5285.

Fixes #5413

Signed-off-by: Niels Lohmann <mail@nlohmann.me>
2026-09-05 20:59:58 +02:00
6 changed files with 163 additions and 358 deletions
+14 -2
View File
@@ -71,7 +71,7 @@ class serializer
, thousands_sep(loc->thousands_sep == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->thousands_sep))) , thousands_sep(loc->thousands_sep == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->thousands_sep)))
, decimal_point(loc->decimal_point == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->decimal_point))) , decimal_point(loc->decimal_point == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->decimal_point)))
, indent_char(ichar) , indent_char(ichar)
, indent_string(512, indent_char) , indent_string()
, error_handler(error_handler_) , error_handler(error_handler_)
{} {}
@@ -126,6 +126,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -199,6 +203,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -260,6 +268,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -1010,7 +1022,7 @@ class serializer
/// the indentation character /// the indentation character
const char indent_char; const char indent_char;
/// the indentation string /// the indentation string (lazily allocated on first use by a pretty-print branch)
string_t indent_string; string_t indent_string;
/// error_handler how to react on decoding errors /// error_handler how to react on decoding errors
+17 -124
View File
@@ -3573,7 +3573,6 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
{ {
using std::swap; using std::swap;
swap(*(m_data.m_value.array), other); swap(*(m_data.m_value.array), other);
set_parents();
} }
else else
{ {
@@ -3590,7 +3589,6 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
{ {
using std::swap; using std::swap;
swap(*(m_data.m_value.object), other); swap(*(m_data.m_value.object), other);
set_parents();
} }
else else
{ {
@@ -5159,139 +5157,34 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
case value_t::object: case value_t::object:
{ {
// first pass: record, for every source key, whether it is // first pass: traverse this object's elements
// common to both objects (in source's iteration order) or
// was deleted (i.e., in source but not in target) -- this is
// a by-product of the target.find() call already needed to
// tell the two cases apart, so it adds no extra lookups. The
// "remove" ops themselves are emitted later, interleaved
// with the recursive per-key diffs in the fast path below,
// to match source's original iteration order (as the
// original, pre-reordering-aware implementation did) instead
// of grouping all removes before all recursive diffs.
std::vector<typename object_t::key_type> common_keys_source_order;
for (auto it = source.cbegin(); it != source.cend(); ++it) for (auto it = source.cbegin(); it != source.cend(); ++it)
{ {
// escape the key name to be used in a JSON patch
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
if (target.find(it.key()) != target.end()) if (target.find(it.key()) != target.end())
{ {
common_keys_source_order.push_back(it.key()); // recursive call to compare object values at key it
auto temp_diff = diff(it.value(), target[it.key()], path_key);
result.insert(result.end(), temp_diff.begin(), temp_diff.end());
}
else
{
// found a key that is not in o -> remove it
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
} }
} }
// second pass: find keys that were added (i.e., in target but // second pass: traverse other object's elements
// not in source), and record the keys common to both, in
// target's iteration order -- again a by-product of the
// source.find() call already needed to detect added keys. At
// the same time, determine whether every added key comes
// after every common key in target's order (a precondition
// for the fast path below, which only ever appends new keys
// at the very end): for an object_t whose iteration order is
// a pure function of the key set (e.g. the default std::map,
// which always iterates in sorted key order), the order
// check further below is always true and this whole
// mechanism is effectively a no-op; it only matters for a
// reorderable object_t such as the one backing `ordered_json`.
// patch ops for keys that were added (i.e., in target but not
// in source); built here so the fast path below can reuse
// them without a second source.find() per target key. Only
// used by the fast path -- the slow (reordering) path
// rebuilds "add" ops for every key itself.
std::vector<typename object_t::key_type> common_keys_target_order;
basic_json added_ops(value_t::array);
bool new_keys_form_suffix = true;
bool seen_new_key = false;
for (auto it = target.cbegin(); it != target.cend(); ++it) for (auto it = target.cbegin(); it != target.cend(); ++it)
{ {
if (source.find(it.key()) == source.end()) if (source.find(it.key()) == source.end())
{ {
seen_new_key = true; // found a key that is not in this -> add it
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
added_ops.push_back(
{
{"op", "add"}, {"path", path_key},
{"value", it.value()}
});
}
else
{
common_keys_target_order.push_back(it.key());
if (seen_new_key)
{
new_keys_form_suffix = false;
}
}
}
if (common_keys_source_order == common_keys_target_order && new_keys_form_suffix)
{
// fast path: order of common keys already matches (or the
// object_t's iteration order does not depend on
// insertion history), so a plain per-key recursive diff
// is correct and minimal, as before. common_keys_source_order
// is, by construction, the subsequence of source's keys
// that are common to both objects, in source's iteration
// order -- so it can be walked in lockstep with `source`
// using a cheap key comparison instead of another lookup.
// Deleted keys (those source keys not in common_keys_source_order)
// are interleaved here too, in source's original order, to
// match the historical (pre-reordering-aware) output order.
auto common_it = common_keys_source_order.cbegin();
for (auto it = source.cbegin(); it != source.cend(); ++it)
{
if (common_it != common_keys_source_order.cend() && it.key() == *common_it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
auto temp_diff = diff(it.value(), target[it.key()], path_key);
result.insert(result.end(), temp_diff.begin(), temp_diff.end());
++common_it;
}
else
{
// found a key that is not in target -> remove it
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
}
}
// append the "add" ops for brand-new keys collected above
// during the pass over target -- no second source.find()
// per target key needed
result.insert(result.end(), added_ops.begin(), added_ops.end());
}
else
{
// slow path: the common keys are in a different relative
// order in source and target (only possible for a
// reorderable object_t like ordered_map). Building a
// minimal reordering patch is a nontrivial (LCS-like)
// problem; instead, remove every source key -- both
// deleted keys (which must be removed regardless) and
// common keys (removed so they can be re-added in
// target's order) -- and re-add every key that should
// remain, with its final target value, in target's
// order. basic_json::patch()'s "add" operation on an
// object uses operator[], which appends at the end for a
// vector-backed insertion-ordered map when the key does
// not already exist -- so removing a key and then adding
// it moves it to the end, fixing its position.
for (auto it = source.cbegin(); it != source.cend(); ++it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
}
// add every key that is either common (just removed
// above) or brand new, in target's iteration order, so
// that the final order after applying the patch matches
// target exactly
for (auto it = target.cbegin(); it != target.cend(); ++it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key())); const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back( result.push_back(
{ {
+31 -126
View File
@@ -20150,7 +20150,7 @@ class serializer
, thousands_sep(loc->thousands_sep == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->thousands_sep))) , thousands_sep(loc->thousands_sep == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->thousands_sep)))
, decimal_point(loc->decimal_point == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->decimal_point))) , decimal_point(loc->decimal_point == nullptr ? '\0' : std::char_traits<char>::to_char_type(* (loc->decimal_point)))
, indent_char(ichar) , indent_char(ichar)
, indent_string(512, indent_char) , indent_string()
, error_handler(error_handler_) , error_handler(error_handler_)
{} {}
@@ -20205,6 +20205,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -20278,6 +20282,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -20339,6 +20347,10 @@ class serializer
// variable to hold indentation for recursive calls // variable to hold indentation for recursive calls
const auto new_indent = current_indent + indent_step; const auto new_indent = current_indent + indent_step;
if (JSON_HEDLEY_UNLIKELY(indent_string.empty()))
{
indent_string.resize(512, indent_char);
}
if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent)) if (JSON_HEDLEY_UNLIKELY(indent_string.size() < new_indent))
{ {
indent_string.resize(indent_string.size() * 2, ' '); indent_string.resize(indent_string.size() * 2, ' ');
@@ -21089,7 +21101,7 @@ class serializer
/// the indentation character /// the indentation character
const char indent_char; const char indent_char;
/// the indentation string /// the indentation string (lazily allocated on first use by a pretty-print branch)
string_t indent_string; string_t indent_string;
/// error_handler how to react on decoding errors /// error_handler how to react on decoding errors
@@ -25001,7 +25013,6 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
{ {
using std::swap; using std::swap;
swap(*(m_data.m_value.array), other); swap(*(m_data.m_value.array), other);
set_parents();
} }
else else
{ {
@@ -25018,7 +25029,6 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
{ {
using std::swap; using std::swap;
swap(*(m_data.m_value.object), other); swap(*(m_data.m_value.object), other);
set_parents();
} }
else else
{ {
@@ -26587,139 +26597,34 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
case value_t::object: case value_t::object:
{ {
// first pass: record, for every source key, whether it is // first pass: traverse this object's elements
// common to both objects (in source's iteration order) or
// was deleted (i.e., in source but not in target) -- this is
// a by-product of the target.find() call already needed to
// tell the two cases apart, so it adds no extra lookups. The
// "remove" ops themselves are emitted later, interleaved
// with the recursive per-key diffs in the fast path below,
// to match source's original iteration order (as the
// original, pre-reordering-aware implementation did) instead
// of grouping all removes before all recursive diffs.
std::vector<typename object_t::key_type> common_keys_source_order;
for (auto it = source.cbegin(); it != source.cend(); ++it) for (auto it = source.cbegin(); it != source.cend(); ++it)
{ {
// escape the key name to be used in a JSON patch
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
if (target.find(it.key()) != target.end()) if (target.find(it.key()) != target.end())
{ {
common_keys_source_order.push_back(it.key()); // recursive call to compare object values at key it
auto temp_diff = diff(it.value(), target[it.key()], path_key);
result.insert(result.end(), temp_diff.begin(), temp_diff.end());
}
else
{
// found a key that is not in o -> remove it
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
} }
} }
// second pass: find keys that were added (i.e., in target but // second pass: traverse other object's elements
// not in source), and record the keys common to both, in
// target's iteration order -- again a by-product of the
// source.find() call already needed to detect added keys. At
// the same time, determine whether every added key comes
// after every common key in target's order (a precondition
// for the fast path below, which only ever appends new keys
// at the very end): for an object_t whose iteration order is
// a pure function of the key set (e.g. the default std::map,
// which always iterates in sorted key order), the order
// check further below is always true and this whole
// mechanism is effectively a no-op; it only matters for a
// reorderable object_t such as the one backing `ordered_json`.
// patch ops for keys that were added (i.e., in target but not
// in source); built here so the fast path below can reuse
// them without a second source.find() per target key. Only
// used by the fast path -- the slow (reordering) path
// rebuilds "add" ops for every key itself.
std::vector<typename object_t::key_type> common_keys_target_order;
basic_json added_ops(value_t::array);
bool new_keys_form_suffix = true;
bool seen_new_key = false;
for (auto it = target.cbegin(); it != target.cend(); ++it) for (auto it = target.cbegin(); it != target.cend(); ++it)
{ {
if (source.find(it.key()) == source.end()) if (source.find(it.key()) == source.end())
{ {
seen_new_key = true; // found a key that is not in this -> add it
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
added_ops.push_back(
{
{"op", "add"}, {"path", path_key},
{"value", it.value()}
});
}
else
{
common_keys_target_order.push_back(it.key());
if (seen_new_key)
{
new_keys_form_suffix = false;
}
}
}
if (common_keys_source_order == common_keys_target_order && new_keys_form_suffix)
{
// fast path: order of common keys already matches (or the
// object_t's iteration order does not depend on
// insertion history), so a plain per-key recursive diff
// is correct and minimal, as before. common_keys_source_order
// is, by construction, the subsequence of source's keys
// that are common to both objects, in source's iteration
// order -- so it can be walked in lockstep with `source`
// using a cheap key comparison instead of another lookup.
// Deleted keys (those source keys not in common_keys_source_order)
// are interleaved here too, in source's original order, to
// match the historical (pre-reordering-aware) output order.
auto common_it = common_keys_source_order.cbegin();
for (auto it = source.cbegin(); it != source.cend(); ++it)
{
if (common_it != common_keys_source_order.cend() && it.key() == *common_it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
auto temp_diff = diff(it.value(), target[it.key()], path_key);
result.insert(result.end(), temp_diff.begin(), temp_diff.end());
++common_it;
}
else
{
// found a key that is not in target -> remove it
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
}
}
// append the "add" ops for brand-new keys collected above
// during the pass over target -- no second source.find()
// per target key needed
result.insert(result.end(), added_ops.begin(), added_ops.end());
}
else
{
// slow path: the common keys are in a different relative
// order in source and target (only possible for a
// reorderable object_t like ordered_map). Building a
// minimal reordering patch is a nontrivial (LCS-like)
// problem; instead, remove every source key -- both
// deleted keys (which must be removed regardless) and
// common keys (removed so they can be re-added in
// target's order) -- and re-add every key that should
// remain, with its final target value, in target's
// order. basic_json::patch()'s "add" operation on an
// object uses operator[], which appends at the end for a
// vector-backed insertion-ordered map when the key does
// not already exist -- so removing a key and then adding
// it moves it to the end, fixing its position.
for (auto it = source.cbegin(); it != source.cend(); ++it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back(object(
{
{"op", "remove"}, {"path", path_key}
}));
}
// add every key that is either common (just removed
// above) or brand new, in target's iteration order, so
// that the final order after applying the patch matches
// target exactly
for (auto it = target.cbegin(); it != target.cend(); ++it)
{
const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key())); const auto path_key = detail::concat<string_t>(path, '/', detail::escape(it.key()));
result.push_back( result.push_back(
{ {
-31
View File
@@ -273,36 +273,5 @@ TEST_CASE("Regression tests for extended diagnostics")
CHECK(j1["numbers"]["two"] == 2); CHECK(j1["numbers"]["two"] == 2);
CHECK(j1["string"] == "t"); CHECK(j1["string"] == "t");
} }
SECTION("Regression test - swap(array_t&)/swap(object_t&) must update JSON_DIAGNOSTICS parent pointers")
{
// swap(array_t&)
{
json j = json::array();
json::array_t arr = {json::array({1})};
j.swap(arr);
// parent pointers of the moved-in elements must point into j, not
// into the now-defunct free-standing array_t
CHECK_THROWS_WITH_AS(j[0][0].get<std::string>(), "[json.exception.type_error.302] (/0/0) type must be string, but is number", json::type_error);
// must not trigger assert_invariant() in a debug/assert-enabled build
json const k = j;
CHECK(k == j);
}
// swap(object_t&)
{
json o = json::object();
json::object_t obj = {{"a", json::array({1})}};
o.swap(obj);
CHECK_THROWS_WITH_AS(o["a"][0].get<std::string>(), "[json.exception.type_error.302] (/a/0) type must be string, but is number", json::type_error);
// must not trigger assert_invariant() in a debug/assert-enabled build
json const p = o;
CHECK(p == o);
}
}
} }
-81
View File
@@ -81,84 +81,3 @@ TEST_CASE("regression test for issue #3732 - iteration_proxy_value<iter_impl<ord
}; };
static_cast<void>(fn); static_cast<void>(fn);
} }
TEST_CASE("regression test - diff() must account for ordered_json member order")
{
SECTION("pure reorder, no value changes")
{
ordered_json a = {{"a", 1}, {"b", 2}};
ordered_json b = {{"b", 2}, {"a", 1}};
CHECK(a != b); // order-sensitive equality
CHECK(a.patch(ordered_json::diff(a, b)) == b);
}
SECTION("new key must land at the front")
{
ordered_json c = {{"b", 2}};
ordered_json e = {{"a", 1}, {"b", 2}};
CHECK(c.patch(ordered_json::diff(c, e)) == e);
}
SECTION("reorder plus a value change on one of the reordered keys")
{
ordered_json a = {{"a", 1}, {"b", 2}};
ordered_json b = {{"b", 20}, {"a", 1}};
CHECK(a != b);
CHECK(a.patch(ordered_json::diff(a, b)) == b);
}
SECTION("reorder plus a deleted key")
{
ordered_json a = {{"a", 1}, {"b", 2}, {"c", 3}};
ordered_json b = {{"b", 2}, {"a", 1}};
CHECK(a != b);
CHECK(a.patch(ordered_json::diff(a, b)) == b);
}
SECTION("reorder plus a nested value that itself needs a recursive diff")
{
ordered_json a = {{"a", {{"x", 1}, {"y", 2}}}, {"b", 2}};
ordered_json b = {{"b", 2}, {"a", {{"x", 1}, {"y", 99}}}};
CHECK(a != b);
CHECK(a.patch(ordered_json::diff(a, b)) == b);
}
SECTION("three or more keys shuffled into a different order")
{
ordered_json a = {{"a", 1}, {"b", 2}, {"c", 3}, {"d", 4}};
ordered_json b = {{"d", 4}, {"b", 2}, {"a", 1}, {"c", 3}};
CHECK(a != b);
CHECK(a.patch(ordered_json::diff(a, b)) == b);
}
SECTION("matching order still produces a minimal patch (fast path unaffected)")
{
ordered_json a = {{"a", 1}, {"b", 2}, {"c", 3}};
ordered_json b = {{"a", 1}, {"b", 20}, {"c", 3}};
auto p = ordered_json::diff(a, b);
// only the changed value should be touched, not a wholesale remove+add
CHECK(p.size() == 1);
CHECK(p[0]["op"] == "replace");
CHECK(p[0]["path"] == "/b");
CHECK(a.patch(p) == b);
}
SECTION("plain json (std::map-backed) is unaffected by same-key-different-insertion-order")
{
json a;
a["b"] = 2;
a["a"] = 1;
json b;
b["a"] = 1;
b["b"] = 2;
// std::map iteration is always sorted by key, so a == b regardless of
// insertion order, and diff() must still produce the same minimal
// (empty) result as before this fix
CHECK(a == b);
auto p = json::diff(a, b);
CHECK(p.empty());
CHECK(a.patch(p) == b);
}
}
+107
View File
@@ -14,6 +14,40 @@ using nlohmann::json;
#include <array> #include <array>
#include <sstream> #include <sstream>
#include <iomanip> #include <iomanip>
#include <cstdlib>
#include <new>
namespace
{
// heap allocation counter used by the regression test for issue #5413
// (https://github.com/nlohmann/json/issues/5413); disabled (and thus a
// no-op besides the counting) unless explicitly toggled on
bool count_heap_allocations = false; // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
std::size_t heap_allocations = 0; // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
} // namespace
void* operator new (std::size_t size) // NOLINT(cppcoreguidelines-owning-memory,misc-new-delete-overloads)
{
if (count_heap_allocations)
{
++heap_allocations;
}
if (void* ptr = std::malloc(size)) // NOLINT(cppcoreguidelines-no-malloc,cppcoreguidelines-owning-memory)
{
return ptr;
}
throw std::bad_alloc(); // NOLINT(hicpp-exception-baseclass)
}
void operator delete (void* ptr) noexcept // NOLINT(cppcoreguidelines-owning-memory,misc-new-delete-overloads)
{
std::free(ptr); // NOLINT(cppcoreguidelines-no-malloc,cppcoreguidelines-owning-memory)
}
void operator delete (void* ptr, std::size_t /*size*/) noexcept // NOLINT(cppcoreguidelines-owning-memory,misc-new-delete-overloads)
{
std::free(ptr); // NOLINT(cppcoreguidelines-no-malloc,cppcoreguidelines-owning-memory)
}
TEST_CASE("serialization") TEST_CASE("serialization")
{ {
@@ -382,3 +416,76 @@ TEST_CASE("dump for basic_json with long double number_float_t")
check_same(100.0L, 100.0); check_same(100.0L, 100.0);
} }
} }
TEST_CASE("regression test for issue #5413 - lazily allocated indent_string")
{
// the serializer used to unconditionally allocate a 512-byte
// indent_string in its constructor, even though it is only ever read
// inside the pretty_print branches of dump(). This wasted a heap
// allocation (and its matching deallocation) on every single compact
// (i.e. non-pretty, the default) dump() call. indent_string is now
// allocated lazily, the first time a pretty-print branch actually
// needs it -- so a compact dump() must perform strictly fewer heap
// allocations than a pretty dump() of the same value.
const json j = {{"level", "info"}, {"msg", "hello world"}, {"id", 12345}};
// warm up anything unrelated to indentation (e.g., one-time locale
// lookups) that might otherwise allocate on first use regardless of
// pretty-printing, so it does not skew the counts measured below
const auto warmup = j.dump();
const auto warmup_pretty = j.dump(4);
CHECK(!warmup.empty());
CHECK(!warmup_pretty.empty());
SECTION("compact dump() has a stable, minimal allocation count")
{
count_heap_allocations = true;
heap_allocations = 0;
const auto compact1 = j.dump();
const auto allocs_compact1 = heap_allocations;
heap_allocations = 0;
const auto compact2 = j.dump(-1);
const auto allocs_compact2 = heap_allocations;
count_heap_allocations = false;
CHECK(compact1 == compact2);
// dump() and dump(-1) both take the compact code path and must
// never touch indent_string, so they allocate identically often
CHECK(allocs_compact1 == allocs_compact2);
}
SECTION("first pretty dump() allocates more than a compact dump()")
{
// use a tiny value whose compact ({"a":1}, 7 bytes) and pretty
// ({"a": 1} with 1-space indent, 11 bytes) serializations both stay
// well inside every common std::string small-string-optimization
// buffer (>= 15 bytes on libstdc++/MSVC STL, >= 22 on libc++), so
// building the result string itself causes no heap allocation
// either way -- isolating indent_string as the only thing that can
// possibly account for a difference in allocation count
const json tiny = {{"a", 1}};
count_heap_allocations = true;
heap_allocations = 0;
const auto compact = tiny.dump();
const auto allocs_compact = heap_allocations;
heap_allocations = 0;
const auto pretty = tiny.dump(1);
const auto allocs_pretty = heap_allocations;
count_heap_allocations = false;
CHECK(compact == "{\"a\":1}");
CHECK(pretty == "{\n \"a\": 1\n}");
// a fresh serializer is created per dump() call; the pretty branch
// lazily allocates indent_string on its first use, so it must
// allocate at least once more than the compact branch, which never
// touches indent_string at all
CHECK(allocs_pretty > allocs_compact);
}
}