pyranges: Range join (also overlap, intersect) fails until I recreate the PyRanges from its .df

I encountered this today, and I’m completely baffled by it.

I have an object test_gr that looks like this:

+--------------+-----------+-----------+----------------------+-------+
|   Chromosome |     Start |       End | Value                | +4    |
|   (category) |   (int32) |   (int32) | (object)             | ...   |
|--------------+-----------+-----------+----------------------+-------|
|            1 | 161332203 | 161332344 | POOR_MAPPING_QUALITY | ...   |
+--------------+-----------+-----------+----------------------+-------+
Unstranded PyRanges object has 1 rows and 8 columns from 1 chromosomes.
For printing, the PyRanges was sorted on Chromosome.
4 hidden columns: sample_name, vendor, experiment, promised_exon

And I have an object test_exons that I want to join, as below

+--------------+-----------+-----------+
| Chromosome   | Start     | End       |
| (category)   | (int32)   | (int32)   |
|--------------+-----------+-----------|
| 1            | 15764898  | 15764905  |
| 1            | 15764946  | 15765013  |
| 1            | 15766782  | 15766901  |
| 1            | 15766966  | 15766973  |
| ...          | ...       | ...       |
| X            | 154002863 | 154002997 |
| X            | 154003456 | 154003561 |
| X            | 154004448 | 154004612 |
| X            | 154005060 | 154005156 |
+--------------+-----------+-----------+
Unstranded PyRanges object has 2,177 rows and 3 columns from 23 chromosomes.
For printing, the PyRanges was sorted on Chromosome.

If I run test_gr.join(test_exons), I get Empty PyRanges, which is not correct because there is an overlap in there.

If I do this pr.PyRanges(test_gr.df).join(test_exons), using presumably the same data, then the overlap works.

+--------------+-----------+-----------+----------------------+-------+
|   Chromosome |     Start |       End | Value                | +6    |
|   (category) |   (int32) |   (int32) | (object)             | ...   |
|--------------+-----------+-----------+----------------------+-------|
|            1 | 161332203 | 161332344 | POOR_MAPPING_QUALITY | ...   |
+--------------+-----------+-----------+----------------------+-------+
Unstranded PyRanges object has 1 rows and 10 columns from 1 chromosomes.
For printing, the PyRanges was sorted on Chromosome.
6 hidden columns: sample_name, vendor, experiment, promised_exon, Start_b, ... (+ 1 more.)

How can this be? I’ve tried versions all the way back to 0.0.79, and the problem remains.

test_gr was created by (filters removed):

test_df=failed_loci_per_assay_df[
    (failed_loci_per_assay_df.vendor == "XXX")
    & (failed_loci_per_assay_df.experiment == "YYY")
    & (failed_loci_per_assay_df.promised_exon=="ZZZ")
].copy()
test_gr = pr.PyRanges(test_df.head(1))

Thanks very much!