Fixes #5017

[ptosco]

Reference Issue

This PR fixes #5017

What does this implement/fix? Explain your changes.

patchPandasrepr() patches pd.io.formats.html.HTMLFormatter._write_cell(), then it restores the original function before returning. However, while doing so, it also sets an _rdkitpatched attribute on pandas.io.formats.html.HTMLFormatter.
The next time patchPandasrepr() is called, the patch is not applied anymore due to the present of the _rdkitpatched attribute, even though the patch has actually been reverted.
This explains why the DataFrame displays molecules as PNG images only the first time.
The fix is removing the usage of _rdkitpatched attribute entirely as it is not necessary, since the patch is applied and then reverted. Instead, a try..finally block is used to make sure that the patch is reverted before returning even if an exception is raised.

After this PR, running the test described in #5017 yields the desired behavior:

from rdkit import Chem
from IPython.display import HTML
from rdkit.Chem import PandasTools
import pandas as pd

import rdkit
rdkit.__version__
'2022.03.1pre'

df = pd.DataFrame(dict(a=[1], s=['CCN']))
PandasTools.AddMoleculeColumnToFrame(df, smilesCol='s', molCol='m')
df

df

PandasTools.RenderImagesInAllDataFrames(False)
df

df

PandasTools.RenderImagesInAllDataFrames(True)
df

df

[ptosco]

The macOS_x64_java failure was caused by Avalon failing to download from SourceForge, nothing to do with this PR.

[greglandrum]

LGTM

[greglandrum]

@ptosco : I merged this without actually trying it, and it doesn't seem to actually work in jupyter:

I've tried it in both jupyter notebook and jupyterlab (v3.4.4) and get the same behavior in both of them

Which version/OS did you test this on?

[greglandrum]

@ptosco : I merged this without actually trying it, and it doesn't seem to actually work in jupyter:

The same problem occurs with pandas 1.4.2

[ptosco]

@greglandrum The fix works, the problem is that I inadvertently reverted it in PR #5132:
https://github.com/rdkit/rdkit/pull/5132/files#diff-4d3e0e17d686a9de85fb42b78a37f8fe69e95258f4e6be45899873fa292a2090L200
because I did not merge #5103 into my rdkit-structure-renderer branch before submitting #5132.
Sorry for the inconvenience, I have just submitted #5381 to restore the #5103 fix.

[greglandrum]

@ptosco : sorry, I should have caught that in the review... the PandasTools code is a real brain bender though.

It would be super shiny if we could figure out some way to actually automatically test this stuff, but I don't see how that would work.