`notmuch-escape-boolean-term': Broken for non-ascii characters

`notmuch-escape-boolean-term': Broken for non-ascii characters

[Moritz Ulrich]

[-- Attachment #1: Type: text/plain, Size: 953 bytes --]


Hello,

I recently adopted notmuch as my primary way to read mail, so thank you
for this great tool!

Unfortunately, I ran into a problem of the Emacs side of the project
when used in a non-ascii environment:

Having a tag named 'uni-köln', the tag:-completion doesn't work.

This is caused by `notmuch-escape-boolean-term' errornously escaping the
above string:

(notmuch-escape-boolean-term "uni-köln") => "\"uni-köln\""

This is caused by `string-match' with the following errornously matching
my tag:

(string-match "[^!#-'*-~]" "uni-köln") => 5
(string-match "[^!#-'*-~]" "uni-koln") => nil

I'm not exactly sure how to tackle this - the Regexp was crafted to match
(, ), " if I understand it correct. A simple way would be just adding
more characters as a sort-of whitelist. A nicer solution would be
converting it from [^...] to [...] to explicitly mark letters that needs
to be escaped.

Cheers,
Moritz Ulrich

[-- Attachment #2: Type: application/pgp-signature, Size: 818 bytes --]

Re: `notmuch-escape-boolean-term': Broken for non-ascii characters

[Austin T. Clements]

Quoting Moritz Ulrich <moritz@tarn-vedra.de>:
> Hello,
>
> I recently adopted notmuch as my primary way to read mail, so thank you
> for this great tool!
>
> Unfortunately, I ran into a problem of the Emacs side of the project
> when used in a non-ascii environment:
>
> Having a tag named 'uni-köln', the tag:-completion doesn't work.
>
> This is caused by `notmuch-escape-boolean-term' errornously escaping the
> above string:
>
> (notmuch-escape-boolean-term "uni-köln") => "\"uni-köln\""
>
> This is caused by `string-match' with the following errornously matching
> my tag:
>
> (string-match "[^!#-'*-~]" "uni-köln") => 5
> (string-match "[^!#-'*-~]" "uni-koln") => nil
>
> I'm not exactly sure how to tackle this - the Regexp was crafted to match
> (, ), " if I understand it correct. A simple way would be just adding
> more characters as a sort-of whitelist. A nicer solution would be
> converting it from [^...] to [...] to explicitly mark letters that needs
> to be escaped.

notmuch-escape-boolean-term used to use a blacklist, but we switched
to a whitelist because Xapian's own parser has changed over the years
in its handling of non-ASCII characters and invalidated our blacklist.
Ultimately it seemed much safer to go with a whitelist.  Quoting
"uni-köln" isn't erroneous, it's just conservative.

Could you explain in more detail what's broken?  I tried adding the
tag uni-köln to a message in Emacs, then hitting "s" to start a search
then "tag:<TAB>" and that tag (surrounded by quotes) was one of the
completion options.  Upon completing to that tag, the search worked
fine.

Are you objecting to the unnecessary (but legal) quotes in the
completion?  We might be able to include Unicode word characters in
the quoting whitelist, though that seems like a spot fix (probably a
fairly broad one, so maybe that's fine) and might be tricky because of
Emacs' somewhat weird Unicode regexp support (using [[:alpha:]] might
Just Work, but we'd have to be careful of the active syntax table).
Or tab completion could recognize that, say, tag:uni doesn't require
quoting, but still expand it to tag:"uni-köln".

Re: `notmuch-escape-boolean-term': Broken for non-ascii characters

[Moritz Ulrich]

[-- Attachment #1: Type: text/plain, Size: 2889 bytes --]

"Austin T. Clements" <aclements@csail.mit.edu> writes:

> Quoting Moritz Ulrich <moritz@tarn-vedra.de>:
>> Hello,
>>
>> I recently adopted notmuch as my primary way to read mail, so thank you
>> for this great tool!
>>
>> Unfortunately, I ran into a problem of the Emacs side of the project
>> when used in a non-ascii environment:
>>
>> Having a tag named 'uni-köln', the tag:-completion doesn't work.
>>
>> This is caused by `notmuch-escape-boolean-term' errornously escaping the
>> above string:
>>
>> (notmuch-escape-boolean-term "uni-köln") => "\"uni-köln\""
>>
>> This is caused by `string-match' with the following errornously matching
>> my tag:
>>
>> (string-match "[^!#-'*-~]" "uni-köln") => 5
>> (string-match "[^!#-'*-~]" "uni-koln") => nil
>>
>> I'm not exactly sure how to tackle this - the Regexp was crafted to match
>> (, ), " if I understand it correct. A simple way would be just adding
>> more characters as a sort-of whitelist. A nicer solution would be
>> converting it from [^...] to [...] to explicitly mark letters that needs
>> to be escaped.
>
> notmuch-escape-boolean-term used to use a blacklist, but we switched
> to a whitelist because Xapian's own parser has changed over the years
> in its handling of non-ASCII characters and invalidated our blacklist.
> Ultimately it seemed much safer to go with a whitelist.  Quoting
> "uni-köln" isn't erroneous, it's just conservative.
>
> Could you explain in more detail what's broken?  I tried adding the
> tag uni-köln to a message in Emacs, then hitting "s" to start a search
> then "tag:<TAB>" and that tag (surrounded by quotes) was one of the
> completion options.  Upon completing to that tag, the search worked
> fine.
>
> Are you objecting to the unnecessary (but legal) quotes in the
> completion?  We might be able to include Unicode word characters in
> the quoting whitelist, though that seems like a spot fix (probably a
> fairly broad one, so maybe that's fine) and might be tricky because of
> Emacs' somewhat weird Unicode regexp support (using [[:alpha:]] might
> Just Work, but we'd have to be careful of the active syntax table).
> Or tab completion could recognize that, say, tag:uni doesn't require
> quoting, but still expand it to tag:"uni-köln".

Thanks for explaining the reason for the whitelist-approach. Knowing
this is quite helpful.

I can't really explain why, but I just didn't notice tag:"uni-köln" in
the tag-completion - I think my expectations for finding it as
tag:uni-köln must have blinded me.

While it isn't errornous, it's higly unintuitive to quote tags like
this. I can understand that a much more permissive whitelist could cause
other problems which are harder to track down, so maybe it's possible to
make the behavior configurable (e.g. by using a `defvar' for the regex).

-- 
Moritz Ulrich

[-- Attachment #2: Type: application/pgp-signature, Size: 818 bytes --]