Compare commits

..
Author SHA1 Message Date
Niels Lohmann 0395ad0992 Pass the complete text of a UBJSON/BJData high-precision number to SAX
For a high-precision number that is a floating-point number, the
UBJSON/BJData reader passed lexer::get_token_string() as the source
text to SAX number_float(). That helper is meant for error messages and
shows a token of more than 8192 characters only by its first and last
4096 characters around "...", so a SAX handler received a shortened
string for a long number.

Pass the number text the lexer collected while scanning the number
(lexer::get_string()) instead. It is complete and, as before, holds
only the number, without any whitespace or byte order mark around it
in the payload. Error messages still use get_token_string().

Signed-off-by: Niels Lohmann <mail@nlohmann.me>
2026-11-16 08:27:07 +01:00
10 changed files with 56 additions and 189 deletions

No files matched your search

@@ -100,8 +100,7 @@ points apply to every edit:
[`root()`](root.md). New values (and the element sequences of an edited array/object) go to storage owned by the
document, allocated on demand.
- **A view keeps referring to the same value.** After [`set`](set.md) replaces the value a view refers to, that view
sees the new value; a view of a value that a later edit replaces or drops keeps showing what it last held. Such a
view can still be edited: the edit changes only that detached value, never anything reachable from `root()`. An edit
sees the new value; a view of a value that a later edit replaces or drops keeps showing what it last held. An edit
of an array or object, however, **invalidates the iterators taken over it** (its members may now live in a
different sequence), and a string obtained with [`get_string()`](../basic_json_view/get_string.md) stays valid even
as further edits happen (earlier buffers of edited text are kept alive, not overwritten).
@@ -164,9 +164,7 @@ general.
The same switch happens, for the same reason, when overload 1. replaces a multi-node array/object value with a
scalar: the *parent's* element sequence is what has to switch to links, not `target` itself, because the parent
originally stepped over `target`'s whole subtree by its node count, which no longer applies once `target` is a
one-node scalar. This also holds when `target` is no longer part of the document (an edit erased or replaced it, or
a value that contains it): if the container that `target` was in still uses its parsed layout, it switches to links
(otherwise nothing steps over `target` any more), and only the dropped values change.
one-node scalar.
## Examples
@@ -4096,6 +4096,10 @@ class binary_reader
auto number_lexer = detail::lexer<BasicJsonType, ia_type>(detail::input_adapter(number_vector), false);
const auto result_number = number_lexer.scan();
const auto number_string = number_lexer.get_token_string();
// the number as the lexer read it, for SAX number_float(): unlike
// get_token_string(), which is meant for error messages and shortens
// long tokens, it is complete; the next scan() replaces it
const string_t number_text = number_lexer.get_string();
const auto result_remainder = number_lexer.scan();
using token_type = typename detail::lexer_base<BasicJsonType>::token_type;
@@ -4122,10 +4126,7 @@ class binary_reader
number_string,
out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", number_string, '\''), nullptr));
}
// number_string is a std::string, while the SAX interface takes a
// string_t; convert explicitly, as the two are only implicitly
// convertible for some string types
return sax->number_float(parsed_float, string_t(number_string.data(), number_string.size()));
return sax->number_float(parsed_float, number_text);
}
case token_type::uninitialized:
case token_type::literal_true:
+2 -15
View File
@@ -384,22 +384,9 @@ class editor
// for the container is linear in the size of the document, so
// links (which are marked in the slot) avoid it.
node* const p = parent_known ? parent : find_parent(m_doc, slot);
if (p != nullptr)
if (p != nullptr && ((p->flags & node_flags::moved) == 0 || moved_capacity(m_doc, p) == 0))
{
if ((p->flags & node_flags::moved) == 0 || moved_capacity(m_doc, p) == 0)
{
block_of(m_doc, p, 0);
}
}
else
{
// the slot is no longer part of the document, but the
// (dropped) container it is in may still step over it
node* const q = find_layout_parent(m_doc, slot);
if (q != nullptr)
{
block_of(m_doc, q, 0);
}
block_of(m_doc, p, 0);
}
}
*slot = e.scalar;
+1 -38
View File
@@ -194,9 +194,7 @@ inline node* block_of(document_data& d, node* n, std::size_t extra)
h->len = n->len;
h->next = static_cast<std::uint32_t>(used);
node* o = h + 1;
// (the bound on o keeps the copy inside the block even if the elements
// disagreed with the element count)
for (const node* c = d.first_child_edited(n), *e = d.child_end_edited(n), *const full = h + used; c != e && o != full;)
for (const node* c = d.first_child_edited(n), *e = d.child_end_edited(n); c != e;)
{
if (object)
{
@@ -278,41 +276,6 @@ inline node* find_parent(const document_data& d, const node* target)
}
}
/// The container whose contiguous layout (in the parsed index, or in a new
/// value) steps over `target`, also when an edit dropped the container or a
/// value that encloses it; nullptr if there is none, or if the container no
/// longer walks that layout (its elements moved). An edit never gives a node a
/// larger extent than it was parsed or encoded with (a scalar is one node, an
/// assigned array/object keeps the extent of the slot), so the containers whose
/// extent covers `target` are its original ancestors: the nearest one before it
/// is its original parent. The root of a new value that is the element sequence
/// of another container is walked by that container (which find_parent() finds
/// while it still uses the sequence), not by itself.
inline node* find_layout_parent(const document_data& d, const node* target)
{
const std::less<const node*> lt{};
const node* lo = d.nodes;
if (lt(target, lo) || !lt(target, d.nodes + d.node_count))
{
auto it = d.edits->regions.upper_bound(target);
if (it == d.edits->regions.begin())
{
return nullptr; // LCOV_EXCL_LINE (an array/object with elements is in the index or a new value)
}
--it;
lo = it->second != nullptr ? it->first + 1 : it->first;
}
for (const node* c = target; c != lo;)
{
--c;
if (is_container(*c) && lt(target, c + c->next))
{
return (c->flags & node_flags::moved) == 0 ? const_cast<node*>(c) : nullptr; // NOLINT(cppcoreguidelines-pro-type-const-cast): the nodes belong to the document
}
}
return nullptr;
}
} // namespace view
} // namespace detail
NLOHMANN_JSON_NAMESPACE_END
+5 -4
View File
@@ -22774,6 +22774,10 @@ class binary_reader
auto number_lexer = detail::lexer<BasicJsonType, ia_type>(detail::input_adapter(number_vector), false);
const auto result_number = number_lexer.scan();
const auto number_string = number_lexer.get_token_string();
// the number as the lexer read it, for SAX number_float(): unlike
// get_token_string(), which is meant for error messages and shortens
// long tokens, it is complete; the next scan() replaces it
const string_t number_text = number_lexer.get_string();
const auto result_remainder = number_lexer.scan();
using token_type = typename detail::lexer_base<BasicJsonType>::token_type;
@@ -22800,10 +22804,7 @@ class binary_reader
number_string,
out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", number_string, '\''), nullptr));
}
// number_string is a std::string, while the SAX interface takes a
// string_t; convert explicitly, as the two are only implicitly
// convertible for some string types
return sax->number_float(parsed_float, string_t(number_string.data(), number_string.size()));
return sax->number_float(parsed_float, number_text);
}
case token_type::uninitialized:
case token_type::literal_true:
+3 -53
View File
@@ -5564,9 +5564,7 @@ inline node* block_of(document_data& d, node* n, std::size_t extra)
h->len = n->len;
h->next = static_cast<std::uint32_t>(used);
node* o = h + 1;
// (the bound on o keeps the copy inside the block even if the elements
// disagreed with the element count)
for (const node* c = d.first_child_edited(n), *e = d.child_end_edited(n), *const full = h + used; c != e && o != full;)
for (const node* c = d.first_child_edited(n), *e = d.child_end_edited(n); c != e;)
{
if (object)
{
@@ -5648,41 +5646,6 @@ inline node* find_parent(const document_data& d, const node* target)
}
}
/// The container whose contiguous layout (in the parsed index, or in a new
/// value) steps over `target`, also when an edit dropped the container or a
/// value that encloses it; nullptr if there is none, or if the container no
/// longer walks that layout (its elements moved). An edit never gives a node a
/// larger extent than it was parsed or encoded with (a scalar is one node, an
/// assigned array/object keeps the extent of the slot), so the containers whose
/// extent covers `target` are its original ancestors: the nearest one before it
/// is its original parent. The root of a new value that is the element sequence
/// of another container is walked by that container (which find_parent() finds
/// while it still uses the sequence), not by itself.
inline node* find_layout_parent(const document_data& d, const node* target)
{
const std::less<const node*> lt{};
const node* lo = d.nodes;
if (lt(target, lo) || !lt(target, d.nodes + d.node_count))
{
auto it = d.edits->regions.upper_bound(target);
if (it == d.edits->regions.begin())
{
return nullptr; // LCOV_EXCL_LINE (an array/object with elements is in the index or a new value)
}
--it;
lo = it->second != nullptr ? it->first + 1 : it->first;
}
for (const node* c = target; c != lo;)
{
--c;
if (is_container(*c) && lt(target, c + c->next))
{
return (c->flags & node_flags::moved) == 0 ? const_cast<node*>(c) : nullptr; // NOLINT(cppcoreguidelines-pro-type-const-cast): the nodes belong to the document
}
}
return nullptr;
}
} // namespace view
} // namespace detail
NLOHMANN_JSON_NAMESPACE_END
@@ -6559,22 +6522,9 @@ class editor
// for the container is linear in the size of the document, so
// links (which are marked in the slot) avoid it.
node* const p = parent_known ? parent : find_parent(m_doc, slot);
if (p != nullptr)
if (p != nullptr && ((p->flags & node_flags::moved) == 0 || moved_capacity(m_doc, p) == 0))
{
if ((p->flags & node_flags::moved) == 0 || moved_capacity(m_doc, p) == 0)
{
block_of(m_doc, p, 0);
}
}
else
{
// the slot is no longer part of the document, but the
// (dropped) container it is in may still step over it
node* const q = find_layout_parent(m_doc, slot);
if (q != nullptr)
{
block_of(m_doc, q, 0);
}
block_of(m_doc, p, 0);
}
}
*slot = e.scalar;
+19
View File
@@ -1278,6 +1278,25 @@ TEST_CASE("BJData")
CHECK(j.dump() == "3.141592653589793");
}
SECTION("long floating-point number")
{
// the SAX parser receives the number text as it is in the
// input, also when it is longer than the part of a token
// that error messages show
const std::vector<std::size_t> lengths = {8192, 8193, 10000};
for (const auto length : lengths)
{
CAPTURE(length)
const std::string number = "1." + std::string(length - 2, '3');
std::vector<std::uint8_t> vec = {'H', 'I', static_cast<std::uint8_t>(length & 0xFF), static_cast<std::uint8_t>(length >> 8)};
vec.insert(vec.end(), number.begin(), number.end());
SaxEventLogger logger;
CHECK(json::sax_parse(vec, &logger, json::input_format_t::bjdata));
CHECK(logger.events == std::vector<std::string>({"number_float(" + number + ")"}));
CHECK(json::from_bjdata(vec).is_number_float());
}
}
SECTION("errors")
{
// error while parsing length
-70
View File
@@ -589,76 +589,6 @@ TEST_CASE("json_view edits: a container that was added is replaced by a scalar")
CHECK(d.root().materialize() == json::parse(R"({"a":[1,6,{"k":"s"}]})"));
}
TEST_CASE("json_view edits: a scalar replaces an array or object that is no longer part of the document")
{
// the dropped container that the array/object is in still steps over its
// parsed extent: that container switches to links first, as a live one does
SECTION("members of an erased object")
{
json_editable_document d = json_editable_document::parse(R"({"p": {"a": [1, [2]], "b": 3, "e": []}, "q": 0})");
const json_editable_view p = d.root()["p"];
const json_editable_view a = p["a"];
d.erase(d.root(), "p");
d.set(a[1], 5);
CHECK(p.dump() == R"({"a":[1,5],"b":3,"e":[]})");
d.set(a, 5);
CHECK(p.size() == 3);
CHECK(p.dump() == R"({"a":5,"b":3,"e":[]})");
d.push_back(p["e"], 6);
CHECK(p.dump() == R"({"a":5,"b":3,"e":[6]})");
CHECK(d.root().dump() == R"({"q":0})");
}
SECTION("an array whose elements moved")
{
json_editable_document d = json_editable_document::parse(R"({"p": {"a": [1, 2], "b": 3}, "q": 0})");
const json_editable_view p = d.root()["p"];
const json_editable_view a = p["a"];
d.erase(d.root(), "p");
d.push_back(a, 4); // (the extent of a stays)
d.set(a, 5);
CHECK(p.dump() == R"({"a":5,"b":3})");
CHECK(d.root().dump() == R"({"q":0})");
}
SECTION("an element of an erased array, which grows afterwards")
{
json_editable_document d = json_editable_document::parse(R"({"p": [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], "q": 0})");
const json_editable_view p = d.root()["p"];
d.erase(d.root(), "p");
d.set(p[0], 5);
d.push_back(p, 1);
CHECK(p.dump() == "[5,1]");
CHECK(d.root().dump() == R"({"q":0})");
}
SECTION("values in an erased new value")
{
json_editable_document d = json_editable_document::parse("[0]");
const json_editable_view x = d.push_back(d.root(), json::parse("[[[1, 2]], 3]"));
const json_editable_view y = x[0];
d.erase(d.root(), 1);
d.set(y[0], 5);
CHECK(x.dump() == "[[5],3]");
d.set(y, 5);
CHECK(x.dump() == "[5,3]");
CHECK(d.root().dump() == "[0]");
}
SECTION("values in the elements that an array had before it was replaced")
{
json_editable_document d = json_editable_document::parse("[0]");
d.set(d.root()[0], json::parse("[[[1, 2]]]"));
const json_editable_view y = d.root()[0][0];
d.set(d.root()[0], json::array({3}));
d.set(y[0], 5);
CHECK(y.dump() == "[5]");
d.set(y, 5);
CHECK(y.dump() == "5");
CHECK(d.root().dump() == "[[3]]");
}
}
#if !defined(JSON_NOEXCEPTION)
TEST_CASE("json_view edits: the size of a text arena")
{
+19
View File
@@ -793,6 +793,25 @@ TEST_CASE("UBJSON")
CHECK(j.dump() == "3.141592653589793");
}
SECTION("long floating-point number")
{
// the SAX parser receives the number text as it is in the
// input, also when it is longer than the part of a token
// that error messages show
const std::vector<std::size_t> lengths = {8192, 8193, 10000};
for (const auto length : lengths)
{
CAPTURE(length)
const std::string number = "1." + std::string(length - 2, '3');
std::vector<std::uint8_t> vec = {'H', 'I', static_cast<std::uint8_t>(length >> 8), static_cast<std::uint8_t>(length & 0xFF)};
vec.insert(vec.end(), number.begin(), number.end());
SaxEventLogger logger;
CHECK(json::sax_parse(vec, &logger, json::input_format_t::ubjson));
CHECK(logger.events == std::vector<std::string>({"number_float(" + number + ")"}));
CHECK(json::from_ubjson(vec).is_number_float());
}
}
SECTION("errors")
{
// error while parsing length