Stefan Monnier wrote:

> Isn't there a way other than modifying the file to tell
>  Texinfo that the encoding is utf-8?

Not that I know of, no.  Sorry.

> your test will end up using ASCII in non-English locales

No, because the test doesn't look at the current locale.
It looks at what locales are available.
That being said, the test could be improved to not care about
English; that's simpler anyway.  I'll attach a revised patch,
which also fixes some other problems I noticed (e.g., some
'sed' implementations mishandle NUL bytes).

Defaulting to ASCII in all platforms would leave the original
bug unfixed by default on all installations, causing the
regression from `foo' to 'foo' that Drew Adams complained of.
I.e., on all platforms the Emacs documentation would default
to lower quality than it's had for decades.  It's better to
default to ASCII only on platforms that don't support UTF-8
well.

> the config option might like to use a more
> generic name like "--with-liberal-use-of-unicode".

But it's not liberal: it's conservative!  :-)
That is, it conserves the info file's contents.
Plus, it's UTF-8 not Unicode.

I tried names like '--with-some-utf-8-or-another' but
the 'utf-8-' makes it hard to parse visually, so I inverted the
flag to '--with-ascii-info-marks', which just barely fits in
'configure --help' output.  This is included in the revised
patch.