Current upstream is 3.0.4, which I have packaged in Salsa.

(This is mostly a placeholder to get a bug number - more to follow when
I get test results)

Reverse depends/recommends:  amd-debug-tools apache-arrow asammdf
asf-search astropy augur azure-kusto-python beangrow bmtk boinor bqplot
busco camelot-py cfgrib changeo cnvkit dask dask-image dask.distributed
datalab debian-astro debian-pan debian-science dials dioptas dipy
domdf-python-tools drms dxchange dyda edlio emperor esda
exchange-calendars extra-data flox folium glueviz guidata harlequin
harmonypy hdmf hickle imbalanced-learn influxdb-python ipython
jitterdebugger joblib joypy json-tricks jsonpickle jupyter-cache
libpysal lmfit-py loki-ecmwf macsyfinder macsylib mapclassify mcaller
mdtraj metaphlan metpy metview-python mirtop mkdocs-test mlpack monty
myst-nb nanofilt napari nbsphinx networkx nipype openpyxl
orange-spectroscopy orange3 osmnx pairtools pandas parmed partd patsy
pdb2pqr pg-statviz pint-xarray plasmidid pointpats poretools presto
psychopy pybel pychopper pycoqc pydevd pyensembl pyinstaller pymatgen
pymatgen-core pymecavideo pynwb pyodc pyranges pyreadstat pyrle
pytest-regressions python-airr python-altair python-anndata
python-apptools python-azure python-bioframe python-biom-format
python-bioregistry python-calendra python-cgelib python-clevercsv
python-cobra python-cogent python-cooler python-curies python-datacache
python-django-pint python-dynasor python-fastparquet python-fluids
python-geopandas python-geotiepoints python-gsd python-gtfparse
python-hypothesis python-igraph python-influxdb-client python-iow
python-loompy python-lsp-server python-maggma python-mne python-momepy
python-nanoget python-nanomath python-narwhals python-ncls
python-numpy-groupies python-opencascade-pywrap python-orjson
python-pandas-flavor python-parsl python-pauvre python-peakutils
python-petl python-polsarpro python-pyani python-pybedtools
python-pyfakefs python-pymeasure python-pyomop python-pyproj
python-pyvista python-rdata python-salib python-sercol python-sigima
python-skbio python-streamz python-tablib python-tabulate
python-testbook python-treetime python-ulmo python-upsetplot
python-vega-datasets python-xarray pytorch-geometric pytroll-schedule
pyxnat q2-cutadapt q2-demux q2-diversity-lib q2-emperor q2-metadata
q2-quality-control q2-quality-filter q2-taxa q2-types q2templates
qcumber qiime r-bioc-mofa rapidfuzz rdkit recan refnx rickslab-gpu-utils
rpy2 sarsen scikit-learn scikit-optimize scikit-rf seaborn sentineldl
sherlock skimage sklearn-pandas skorch skyfield slm snakemake
sorted-nearest spaghetti sphinx-book-theme spopt spyder spyder-kernels
statsmodels stimfit sunpy textual-fastdatatable topplot tqdm traittypes
umap-learn umis xarray-safe-rcm xarray-safe-s1 xarray-sentinel xgboost
xpore xraylarch yanagiba

pandas 3.0 is now available in experimental.  As that version number
suggests, it can be a breaking change.

These packages appear to be broken *by* pandas 3.0: apache-arrow
azure-kusto-python beangrow bmtk boinor bqplot cfgrib cnvkit datalab
glueviz influxdb-python joypy json-tricks jsonpickle loki-ecmwf mdtraj
partd pybel pychopper pycoqc pymatgen-core pynwb pyranges pyrle
pytest-regressions python-altair python-anndata python-bioframe
python-bioregistry python-cgelib python-cobra python-dynasor
python-fluids python-pyani python-pyvista python-sigima python-upsetplot
python-vega-datasets q2-emperor q2-metadata q2-taxa r-bioc-mofa rpy2 slm

These packages have pre-existing failures (which are probably RC bugs,
but ones unrelated to this transition): asammdf dials emperor
ipython(sometimes) q2-diversity-lib q2-quality-filter traittypes umap-learn

Autopkgtest results:
https://qa.debian.org/excuses.php?experimental=1&package=pandas
Build test results (packages without (non-trivial) autopkgtests):
https://launchpad.net/~rebecca-palmer/+archive/ubuntu/pandas3p0/+builds?build_text=&build_state=failed
(And no, I don't know why many of these don't actually link to a log, or
whether they will start doing so if we wait longer.)

I suggest checking whether your upstream have already fixed this.

If that doesn't help, some common issues:
- The string dtype (separate from the object dtype) is now used by
default, and is not a numpy dtype
- pandas no longer depends on pytz - if you need it, you need to depend
on it explicitly

For more information, see the upstream release notes:
https://pandas.pydata.org/docs/whatsnew/v3.0.0.html

The "failed with no log" state seems to have been an infrastructure
issue not a real failure: beangrow bqplot loki-ecmwf pychopper pynwb
python-cgelib succeeded when retried.  (cfgrib did fail again, but at
least we can now see how.)

Also, pymatgen-core and rpy2 have now been fixed.

Popcon suggests apache-arrow influxdb-python partd jsonpickle glueviz
(in that order) are the most important of the remaining ones, but is
obviously a very imperfect tool for assessing that.