py3: Fix table display in Jupyter notebook with Sage kernel

[slelievre]

As reported in

• ​Ask Sage question 46005: Issue printing tables in Jupyter Notebook

the following fails to display the expected table properly in the Jupyter Notebook with the Python3-based SageMath kernel:

%display latex
rows = [('A', 'B', 'C'), (0, 1, 2), (3, 4, 5)]
table(rows)

It works fine with the Python2-based SageMath kernel.

Observed with SageMath 8.8.beta1, on MacOS and on CoCalc.

The table displays fine in text mode in the Sage REPL:

sage: rows = [('A', 'B', 'C'), (0, 1, 2), (3, 4, 5)]
sage: table(rows)
  A   B   C
  0   1   2
  3   4   5

[jhpalmieri]

Please provide OS information and the Sage version. The command works just fine for me in the Jupyter notebook with Sage 8.8.beta1 on OS X, with both Safari and Firefox.

[jhpalmieri]

Oh, is it a Python 3 problem? I see this issue with Python 3 but not Python 2.

[slelievre]

Right, forgot to mention this is with Python 3.

[jhpalmieri]

This reminds me of #24784. I would guess that the problem is in the _html_ method for tables, possibly in the use of StringIO.

[jhpalmieri]

Maybe the problem is really in HtmlFragment. Note that this code produces similar gibberish with Python 3 but not Python 2:

A = ([1, 1], [3, 1], [-1, -1])
b = (1000, 1500, -400)
c = (10, 5)
P = InteractiveLPProblemStandardForm(A, b, c)
P.run_simplex_method()

The run_simplex_method also calls HtmlFragment.

Or try this:

a = 123
html.eval('<sage>a</sage>')

[slelievre]

Hopefully a Python3 expert can help with this.

[embray]

I didn't realize this was only on Python 3. Mostly likely the problem is something returning bytes when it should return str.

[jhpalmieri]

I don't know enough about how the Jupyter notebook displays things. The failure of

sage.misc.html.HtmlFragment('hello')

says that the problem is in how the notebook displays instances of HtmlFragment. This is Python 3 only, so Erik is likely right that something is producing bytes instead of str. Can someone who understands the Jupyter notebook shed any light on this?

[embray]

I haven't tested it yet, but this is likely the fix we want:

src/sage/repl/rich_output/backend_ipython.py

@@ -539 +539 @@
                      u'text/plain': plain_text.text.get_unicode(),
             }, {})
         elif isinstance(rich_output, OutputHtml):
-            return ({u'text/html':  rich_output.html.get(),
-                     u'text/plain': plain_text.text.get(),
+            return ({u'text/html':  rich_output.html.get_unicode(),
+                     u'text/plain': plain_text.text.get_unicode(),
             }, {})
         elif isinstance(rich_output, OutputImagePng):
             return ({u'image/png':  rich_output.png.get(),
@@ -548 +548 @@
             }, {})
         elif isinstance(rich_output, OutputImageGif):
             return ({u'text/html':  rich_output.html_fragment(),
-                     u'text/plain': plain_text.text.get(),
+                     u'text/plain': plain_text.text.get_unicode(),
             }, {})
         elif isinstance(rich_output, OutputImageJpg):
             return ({u'image/jpeg':  rich_output.jpg.get(),

[jhpalmieri]

That works for me.

[jhpalmieri]

New commits:

​9214d1c

trac 27656: fix unicode issue with HtmlFragment

[embray]

Thanks for making the branch and reviewing it.

[chapoton]

This was also proposed long ago in #19186.

[slelievre]

Sorry! I looked for an existing ticket before opening this one but missed #19186.