Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> In an nxml-mode buffer, put the following text:
>>
>> <x a="foo" b='bar'/>
>>
>> Only "foo" is fontified in as a string, while bar is not.
>
> IIRC this is technically correct because IIRC in XML (contrary to SGML)
> only "..."  is allowed and not '...'.  Don't take my word for it, tho,
> it's just a vague recollection.

Well, I was pretty sure that XML allows both quotes, but just to make
things definitive, https://www.w3.org/TR/xml/#NT-AttValue:

    [10]    AttValue       ::= '"' ([^<&"] | Reference)* '"'
                            |  "'" ([^<&'] | Reference)* "'"

I found this for SGML http://xml.coverpages.org/sgmlsyn/sgmlsyn.htm#P34:

    [34] attribute value literal =

        ( lit , "
        replaceable character data [46] *,
        lit ) | "
        ( lita , '
        replaceable character data [46] *,
        lita ) '

> This said, it's probably a good idea to understand '...' in nxml-mode,
> since there are various circumstances where you may want to use
> nxml-mode to edit files in a format that's more permissive than strict XML.
>
>> This is a regression since Emacs 25.3.
>
> Even more reason to allow '...'.
> Have you investigated the source of the regression?

I didn't actually checkout and compile before+after, but I'm pretty sure
it's [56e1097584], same as Bug#32003.

[56e1097584]: 2016-01-16 15:03:42 -0500
  lisp/nxml: Use syntax-tables for comments
  https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=56e1097584c13f2b6db85592769db1c6c36e9419

>> The patch below seems solves problem, though I'm not entirely sure
>> about it: there are some confusing comments in sgml-mode.el about
>> drawbacks and tradeoffs of recognizing single quotes that I'm not
>> really following (are they applicable to SGML only, and not XML?).
>
> IIRC the comments aren't related to the issue of accepting '...' itself,
> but to some of the side-effects of doing it naively in cases such as
>
>     <em>That's right!</em>
>
> but we're no so naive any more, so I believe that we can support this
> without the downsides.

>> +  (setq-local syntax-ppss-table nxml-tag-syntax-table)
>> +  (setq-local syntax-propertize-function #'nxml-syntax-propertize)
>
> Hmm... I think it would be better to change `sgml-syntax-propertize` so
> it does what we need.  After all, it's more important to support '...'
> for SGML  than for XML.

s/more/equally/, but otherwise yes.  Patching sgml-mode is shorter and
even fixes Bug#8203 as well.  Still good for emacs-26 I hope, since this
is for an (nxml-mode) regression in 26.1.