OpenSystemsDevelopment / qpdf

During xref recovery reject /Page objects with multiple errors

When validating the pages tree after xref recovery do not tree dictionaries
as page objects if more than 2 of the following are true:

- the dictionary is direct
- the /Type entry is missing or is not /Page
- the dictionary does not contain a valid /Parent entry
- the dictionary does not contain or inherit a valid /MediaBox
- the dictionary does not contain or inherit a /Resources dictionary

Such dictionaries are very unlikely to be page objects and trying to
process them may cause excessive run time and memory usage.

Refine xref table reconstruction

Bump version to 12.2.1

Fix QPDFFormFieldObjectHelper::getChoices (fixes #1433)

During xref table reconstruction ignore uncompressed object entries found
in xref streams. The xref table gets populated with entries for the
objects actually found in the file. The entries for uncompressed object in
xref streams are redundant and potentially incorrect.

Add additional sanity check during xref recovery

Enhance Windows build for local dev

* Remove dependency on `perl`, now only required for maintenance
  activities and running the test suite.
* Exercise building qpdf and running executables from JetBrains CLion
  without any special additional tooling beyond pre-built external
  libraries.

This replaces the `copy_dlls` with a powershell script, written mostly
by ChatGPT, starting from the bash script below. The copy_dlls script
had a lot of logic that was no longer needed.

```bash
#!/bin/bash
set -eo pipefail
exe="$1"
dest="$2"
mingw_bin_dir="$3"
if [[ $mingw_bin_dir == "" ]]; then
    echo >&2 "Usage: $(basename $0) exe dest mingw-bin-dir"
    exit 2
fi

get_dlls() {
    objdump -p "$1" | grep 'DLL Name:' | awk '{print $NF}'
}
declare -a dlls
dlls=($(get_dlls "$exe"))
declare -A seen
while [[ ${#dlls[@]} -gt 0 ]]; do
    i="${dlls[0]}"
    dlls=("${dlls[@]:1}")
    if [[ ${seen[$i]} == 1 ]]; then
        continue
    fi
    seen[$i]=1
    full="$mingw_bin_dir/$i"
    if [[ -f "$full" ]]; then
        cp "$full" $dest/
        dlls+=($(get_dlls "$full"))
    fi
done
```

Return the display value if the choices entry is an array of strings rather
than a single string.

Test file is need-appearances.pdf modified to contain one array entry.

Do not copy duplicate pages during xref recovery.

During xref reconstruction reject unreasonably large objects

Reject objects containing arrays or dictionaries with more than 5000
elements. We are by definition dealing with damaged files, and such
objects are extremely likely to be invalid or malicious.

Don't set tree root /Limits in NNTreeIterator::resetLimits (fixes #1435)

Bump version to 12.1.1

Fix android build issues (fixes #950)

Implement QPDFFormFieldObjectHelper::isChecked

Detect shared /Kids arrays in pages tree

API was defined, but .cc had no implementation. PDF 2.0 manual is surprisingly unclear as to on/off values, giving /Yes in an example and /Off in descriptive text as "not on".

In QPDF::getAllPagesInternal include the /Kids array in the visited set
for loop detection.

Fixes oss-fuzz issue 411312393

Refactor QPDFFormFieldObjectHelper ValueSetter::writeAppearance

Adjusting for under/overflow by repeated incrementing/decrementing can
cause excessive runtime with invalid BBox coordinates.

Fixes oss-fuzz issue 409905355.

Detection of fseeko on Android ABI level < 24 gets a false positive if
_FILE_OFFSET_BITS is not set first.

Thanks to github user @cdosborn for the basic enhancement.

Fix logic around cleartext metadata (fixes #1368)

Only top-level XMP metadata is supposed to be left unencrypted. All
other metadata is not treated specially.

Allow rotate as array in job JSON (fixes #1401)

Replace JetBrains-specific cmake profiles with CMakePresets.json

Deprecate QPDF::optimize