(In reply to Florencia Di Ciocco, NI to vbalducci from comment #1)
Hi, I was able to spot the day this behavior was introduced, it was the 18th of December 2019.
I get the same regression range.
I couldn't finish the bisect because my mozregression throws an error (both in windows and Ubuntu).
I don't think it really throws an error - it just tries to bisect further within a single day and fails to do so because we don't have any finer-grained builds cached that far back. It's quite noisy about this which is unfortunate, but if you scroll up you can still get the last pushlog which it provided for the last bisect operation that it was able to do, which is:
The relevant bug in there would be bug 1602430.
Nonetheless, I want to point out that before this bug, the printing was worse, I attach how it used to print, it's missing the middle section.
Indeed, this wasn't really "good" before the regression. We just failed to clamp the height of the scrollable area at all; whereas after, we do clamp the height but we also let it create additional pages for some reason.
(In this case, the
visibility:hidden on the content meant that it mostly didn't paint, which is why we get a mostly-blank page in good.pdf. The behavior is clearer with my attached testcase, which prints with both bordered boxes being visible on all of the pages, in builds before 2019-12-18.)
Given that, I'm removing the regression keywords, since our behavior now is essentially better than before, even if it's slightly more confusing in some cases. Bottom-line, this is a "wart" that was left behind when bug 1602430 was fixed.