Apparently, terms with a common prefix are *not* connected by implicit "OR"

[David Bremner]

Jorge P. de Morais Neto <jorge+list at disroot.org> writes:

> [ I had replied to David Bremner alone so I didn't have to worry about
> private information leakage.  But now I decided to clean the private
> information and reply to the list too. ]
>
> Em 2019-08-11T20:08:58-0300, David Bremner escreveu:
>> Thanks for the report. As a test, can you try with
>>
>>      $ notmuch count '(to:pontodosconcursos.com.br to:"jorge+cp+concurso at disroot.org")'
>>
>> I suspect that will work around the problem, which I believe is related
>> to the way that notmuch uses the xapian parser (in order to provide
>> regexp matching for some prefixes). In particular, if I try that with
>> NOTMUCH_DEBUG_QUERY=yes in the environment I can see the implicit OR.

Thanks for the detailed report. There are (at least) two different
things going on (in addition to the strange expansion that I focussed on
before, but seems not to be the most important issue).

One is that the combining with implicit-OR was only intended to work for
"boolean prefixes" like tag:. So this is a documentation bug.

A second thing is due to some implimentation details in notmuch, from:
was being treated (for purposes of combining) as a filter. I think it's
clear we want from: and to: to behave similarly, so I propose the
following patch

diff --git a/lib/database.cc b/lib/database.cc
index 24b7ec43..4db1b465 100644
--- a/lib/database.cc
+++ b/lib/database.cc
@@ -400,7 +400,7 @@ _setup_query_field (const prefix_t *prefix, notmuch_database_t *notmuch)
        /* we treat all field-processor fields as boolean in order to get the raw input */
        if (prefix->prefix)
            notmuch->query_parser->add_prefix ("", prefix->prefix);
-       notmuch->query_parser->add_boolean_prefix (prefix->name, fp);
+       notmuch->query_parser->add_boolean_prefix (prefix->name, fp, !(prefix->flags & NOTMUCH_FIELD_PROBABILISTIC));
     } else {
        _setup_query_field_default (prefix, notmuch);
     }


This will make

    to:a to:b

and

    from:a from:b

expand as

    to:a AND to:b

and

    from:a AND from:b


I don't think it's possible to have

  to:a to:b

expand to

   to:a OR to:b

without also having

   a b

expand to

   a OR b

which I think most people would find surprising.

At the moment I'm not sure I see the benefit of having tag: combine
with implicit OR (other than being slightly easier to document).