Thanks for your useful explanations.
This example shows that, sometimes, the
gumbo parser is more royalist than the King (Epubcheck). The main point is that it's right and sound. Maybe it deserves better than the stern warning Sigil issues before using it if you select the option to mend on opening: it says roughtly "Sigil can correct automatically your html but you can lose data in the process", which always put me off up to now.
I did a second try from the same odt file. I asked ODTImport to produce an
EPUB3 without asking Sigil to mend anything. This time, nearly everything was correct for the table, including
colgroup and
tbody tags. Mystery...
Two small things about this EPUB3:
1- One that can be easily corrected using Sigil is changing the extension name of all the files (from html(5) to xhtml) to please Epubcheck. (the gumbo parser is undisturbed about this trifle).
2- Epubcheck signalls also an error about a deprecated (for EPUB3 only) "cellspacing" attribute on this table, which does not disturb the gumbo parser. It's also easy to take out this attribute though maybe not for a beginner. (I use a saved regex to weed it out).