bug#39291: M-: history doesn't store erroneous input

[Eli Zaretskii <eliz@gnu.org>]

> From: Federico Tedin <federicotedin@gmail.com>
> Cc: Juri Linkov <juri@linkov.net>,  39291@debbugs.gnu.org,  pogonyshev@gmail.com
> Date: Wed, 29 Jan 2020 22:32:16 +0100
> 
> I found that the minibuffer input string in `read_minibuf' was being
> added to the history list after we tried to parse an object from it (if
> `expflag' was true). So if the parsing failed, then the value wasn't
> added to the history. I'm attaching a patch with my changes.

Thanks.

> My only doubt is that now, the call to `string_to_object' (which calls
> `read-from-string', which calls `read1', and then `read0') is located
> outside the context (not sure what the right term is) set up with
> `specbind'. Could this be a problem? Do any of these functions depend on
> the context set up in `read_minibuf'?  (They don't appear to, just want
> to be sure).

Why is it a problem to move the call to unbind_to to after the call to
add-to-history?  AFAIU, the original code, before you changed it,
actually did that.  It just didn't call string_to_object, but I'm not
sure why would that change anything.  Does anyone see a problem here?
Stefan, any comments?