bug#49742: 28.0.50; previous-single-property-change sometimes wrong?

[Eli Zaretskii]

> Date: Mon, 26 Jul 2021 08:47:07 -0700
> From:  "T.V Raman" via "Bug reports for GNU Emacs,
>  the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>
> 
> 
> This appears to a corner case where previous-single-property-change
> breaks in a surprizing way.
> 
> 1. ;;; Evaluate the below to get a string with properties.
> (setq s
> 
> #("-L Trace line evolution (-L)
> 
> History simplification" 0 2 (face transient-blue button (t) category 
> transient-button-button command magit-log:-L) 2 25 (button (t) category 
> transient-button-button command magit-log:-L) 25 27 (face 
> transient-inactive-value button (t) category transient-button-button command 
> magit-log:-L) 27 28 (button (t) category transient-button-button command 
> magit-log:-L) 30 52 (face transient-heading)))
> 
> 
> 2.;;; Insert into a new buffer
> (switch-to-buffer "foo")
> (insert s)
> 
> 4. ;;; Place point on the 'H' of "history"
> 
>    5. ;;; eval
>    (previous-single-property-change (point 'face)
> 
>    Instead of returning value of point before the 'H', this returns a
>    value that is unexpected, it returns point  past the ')'

What does "M-x describe-text-properties RET" say at the position
returned by the above previous-single-property-change call?  Does that
value explain the result?  If not, why not?