Generate PDB file failed, as parsing json crashed in `fakepdb_pdb` when doing the pdb generation.

[WindBlowAssCold]

Describe the bug

In my case, the PE file being analyzed had a segment call .voltbl, which can't be parsed by ida correctly.

Then it generate .json like this.

While the null value is not supported by nlohmann, which is used in this project to generate PDB file. So the fakepdb_pdb crashed (I debuged this). As mentioned here [nlohmann/json/issues/2356].

After I edit the segment prosperity like this to avoid null value, the PDB generation works fine.

Maybe, do more check in python, to inform user when there is something not parsable in json?

[Abbas-MG]

I also tracked the issue here.

I made a quick (and messy) workaround since the main JSON object is const:

// Create a copy of just the segments array
auto segments_copy = nlohmann_json_j.at("segments");

// Modify the copy
for (auto& seg : segments_copy)
{
    if (seg.contains("type") && seg["type"].is_null())
    {
        seg["type"] = "UNKNOWN";
    }
}

// Convert the modified copy
segments_copy.get_to(nlohmann_json_t.segments);

To reproduce the issue, I did this:

failed as expected.
After recompiling FakePDB with the above change, it no longer crashed and successfully generated the final .pdb file.

As mentioned in the linked issue, this needs to be addressed by the developers themselves. In other words, the null handling should be implemented within FakePDB’s internals; it requires @Mixaill 's attention.

update:
@WindBlowAssCold test the temporary fix and see if it helps; fresh build provided Here.

[WindBlowAssCold]

I also tracked the issue here.

I made a quick (and messy) workaround since the main JSON object is const:

// Create a copy of just the segments array
auto segments_copy = nlohmann_json_j.at("segments");

// Modify the copy
for (auto& seg : segments_copy)
{
    if (seg.contains("type") && seg["type"].is_null())
    {
        seg["type"] = "UNKNOWN";
    }
}

// Convert the modified copy
segments_copy.get_to(nlohmann_json_t.segments);

To reproduce the issue, I did this:

failed as expected. After recompiling FakePDB with the above change, it no longer crashed and successfully generated the final .pdb file.

As mentioned in the linked issue, this needs to be addressed by the developers themselves. In other words, the null handling should be implemented within FakePDB’s internals; it requires @Mixaill 's attention.

update: @WindBlowAssCold test the temporary fix and see if it helps; fresh build provided Here.

It works. Appreciate for your work.

[dezmen3]

Simple fix for dumpinfo.py:

    def __process_segments(self):
        segments = list()

        for n in range(0, ida_segment.get_segm_qty()):
            seg = ida_segment.getnseg(n)
            if seg:
                segm = {
                    'align'     : self.__describe_alignment(seg.align),
                    'bitness'   : self.__describe_bitness(seg.bitness),
                    'name'      : ida_segment.get_segm_name(seg),
                    'rva_start' : seg.start_ea - self._base,
                    'rva_end'   : seg.end_ea - self._base,
                    'permission': self.__describe_permission(seg.perm),
                    'selector'  : seg.sel,
                    'type'      : ida_segment.get_segm_class(seg) or "UNKNOWN",
                }
                
                segments.append(segm)

        return segments

This is also useful if you inject dll into exe that you loading in IDA

[Abbas-MG]

My humble opinion is that handling an anomaly in complete silence only increases the risk of another bug elsewhere. Giving a hint about what went wrong can save users trouble in later phases of reverse engineering automation.
I handled it with a simple print().

[dezmen3]

My humble opinion is that handling an anomaly in complete silence only increases the risk of another bug elsewhere. Giving a hint about what went wrong can save users trouble in later phases of reverse engineering automation. I handled it with a simple print().

Yeah, detailed errors are always nice :) Tho, seems that part should be enough
Been remaking fakepdb to work on 9.0+ versions (using 9.2), cutting off old support for anything below 9.0 :/