Skip to content

Binary reader for BJData creates incorrect SAX events #3503

Closed

Description

@nlohmann

The BJData reader seem to open an object with fixed size of 3 which is never closed, yielding a problem in the SAX parser. I fuzzed this branch (together with the assertions from #3498) and got the following failing input: crash2.bjdata.zip

This generates the following SAX events:

<array>
    <array>
        <object>
            <key key="��" />
            <array>
                <number_integer val="27739" />
                <boolean val="true" />
                <object size="3">                   <!-- open an object with size 3 -->
                    <key key="_ArraySize_" />
                    <array size="2">
                        <number_integer val="6689" />
                        <number_integer val="0" />
                    </array>                        <!-- the first object entry -->
                    <array size="0">                <!-- ERROR: a value without key -->
                    </array>
                    <boolean val="true" />          <!-- ERROR: a value without key -->
                    <boolean val="true" />          <!-- ERROR: a value without key -->
                    <boolean val="true" />          <!-- ERROR: a value without key -->
                    <boolean val="true" />          <!-- ERROR: a value without key -->
                </array>                            <!-- ERROR: closing array (not object) -->
                <key key="" />
                <number_unsigned val="30069" />
                <parse_error id="71" token="<end of file>" />

Again, for line

if (JSON_HEDLEY_UNLIKELY(!sax->start_object(3) || !sax->key(key) || !sax->start_array(dim.size())))

we need to check

  • Why is the size 3 hard-coded?
  • Where to add the required sax->end_object() call?

Originally posted by @nlohmann in #3502 (comment)

Activity

  1. nlohmann commented on May 20, 2022

    @nlohmann
    OwnerAuthor

    I just ran the fuzzer again and wanted to add another (shorter) example: minimized-from-48f6d80fc879d78470bda5e99e4a1213d4d963a1.zip

    <object>
        <key key="N" />
        <array>
            <object size="3">
                <key key="_ArraySize_" />
                <array size="2">
                    <number_integer val="9216" />
                    <number_integer val="792633534417207296" />
                </array>
                <array size="0">. <!-- expected a key here -->
                </array>
            </array>              <!-- expected a end_object() here -->
  2. fangq commented on May 20, 2022

    @fangq
    Contributor

    hi @nlohmann, sorry for the late reply

    Why is the size 3 hard-coded?

    it is sort of a "specification" I want to ensure a bi-directional mapping between a bjdata ND array to a JSON object, as described here.

    Where to add the required sax->end_object() call?

    it is added at the end of this block. This if-branch uses the bit-8 of size_and_type.second (type marker) to tell if an object-based ND array has been partially written, and continue finishing it in the above block. However, looks like this is not robust, and does not always enter this branch even an object has been partially written.

    I will work on a more robust way to implement this.

  3. added a commit that references this issue on May 22, 2022
    7269b88
  4. added this to the Release 3.11.0 milestone on May 22, 2022
  5. self-assigned this
    on May 22, 2022
  6. added a commit that references this issue on May 23, 2022
    ede6667
  7. added a commit that references this issue on Sep 18, 2024
    fcfc848
  8. added a commit that references this issue on Jun 26, 2025
    bbd1e2d
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Projects

No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

    ⌂ 换网址