The problem is that the original image of the text is "in front" of the actual digital text of the PDF.
Figure 1: How layers are stacked in a PDF
PDF is composed of overlapping visible and invisible layers. OCR attempts to use algorithms to identify text placement and size, then place text objects under the original images of the pages.
See these ...
PDFs are not designed to be converted from. They have no structure that allows Calibre to determine for sure how many columns of text it has or what's a footnote. The Calibre manual describes the topic in more depth.
I found that copying the text to a text editor, cleaning it up and then pasting it into the Calibre e-book editor gives the best results but it ...
If you are starting with an .epub file, there is an editor program in the Calibre suite which can change anything in the .epub, including the fonts:
But if you convert the file from .epub to .mobi, with any program including Calibre, it's then almost impossible to edit the resulting .mobi file. So you get stuck with ...
I found the answer: the original .html file had table inside with fixed width, and that's why the text was running out of the screen. The solution: just use the option --linearize-tables !
ebook-convert 08-microprocesseur-son-environnement.html 08-microprocesseur-son-environnement.epub --linearize-tables