Infinite loop when adding multi_cell twice

[skawouter]

When creating a simple pdf page and calling multi_cell twice in a row an infinite loop occurs.

Minimal code

from fpdf import FPDF
pdf = FPDF()
pdf.add_page()
pdf.set_font('Helvetica', '', 10)
pdf.multi_cell(w=0, h=5, txt='test')
pdf.multi_cell(w=0, h=5, txt='test')

I've traced the error to this piece of code in fpdf.py line 2860

        if w == 0:
            w = self.w - self.r_margin - self.x
        if h is None:
            h = self.font_size
        maximum_allowed_emwidth = (w - 2 * self.c_margin) * 1000 / self.font_size

where the resulting maximum_allowed_emwidth becomes negative. This is the result of self.x being almost equal to self.w

Environment

• Linux Mint 20
• Python 3.8.10
• fpdf2 2.5.2 -> bug also present on master

[Lucas-C]

Wow, this is a very annoying bug 😢
Thank you for reporting it @skawouter
I see, the infinite loop occurs in the while (text_line) is not None: loop just after the line you mentioned.
I'm going to try to fix this during the evening.
Ping @gmischler for information, as you worked on that part of the code too.

[Lucas-C]

I opened #390 to fix this

[Lucas-C]

Now your example would raise the following error:

$ ./issue_389.py
Traceback (most recent call last):
  File "./issue_389.py", line 13, in <module>
    pdf.multi_cell(w=0, h=5, txt='test')
  File "/opt/fpdf2/fpdf/fpdf.py", line 265, in wrapper
    return fn(self, *args, **kwargs)
  File "/opt/fpdf2/fpdf/fpdf.py", line 2864, in multi_cell
    "Not enough horizontal space to render cell. Consider introducing a line break."
fpdf.errors.FPDFException: Not enough horizontal space to render cell. Consider introducing a line break.

Note that you probably want to use new_x=XPos.LEFT:

pdf.multi_cell(w=0, h=5, txt='test', new_x=XPos.LEFT)
pdf.multi_cell(w=0, h=5, txt='test', new_x=XPos.LEFT)

[gmischler]

Makes you wonder about the wisdom of defaulting new_y to RIGHT...

@Lucas-C, Have you checked what happens if w ends up larger than c_margin but still smaller than the width of the next character?

[Lucas-C]

Good catch, I'm going to test that

[Lucas-C]

I have found another case where an infinite loop can still occur:

def test_multi_cell_with_limited_horizontal_space(tmp_path):  # issue #389
    pdf = FPDF()
    pdf.add_page()
    pdf.set_font("Helvetica", "", 10)
    pdf.multi_cell(w=pdf.epw - 2 * pdf.c_margin - 1, h=5, txt="test")
    assert pdf.x == pdf.l_margin + pdf.epw - 2 * pdf.c_margin - 1
    pdf.multi_cell(w=0, h=5, txt="test")
    assert_pdf_equal(pdf, HERE / "multi_cell_with_limited_horizontal_space.pdf", tmp_path, generate=True)

I think we should raise an error in MultiLineBreak.get_line_of_given_width() when this method return twice a manual_break() without consuming a character:
https://github.com/PyFPDF/fpdf2/blob/2.5.2/fpdf/line_break.py#L262

My instinct would be to simply add a dummy boolean state attribute to this class...
As you wrote it @gmischler, what do you think?

[gmischler]

As you wrote it @gmischler, what do you think?

@Lucas-C, I only modified that code a bit, it was really written by @oleksii-shyman.

The approach of checking for repeated lines with no content sounds about right.
It might be helpful if we can make sure that the new_x|y settings still get handled, even when an exception is raised.

Unfortunately, it gets a little more convoluted now that we need to handle the problem deeper down in the call chain...
The later phases of #339 might result in further tweeks (eg. changing the boolean to a counter), but let's cross that bridge when we get to it. 😉

[Lucas-C]

Thank you for your answer @gmischler

I submitted #392 to fix this