Re: CSV parsing and other issues (Re: LC_NUMERIC)

[Maxim Nikulin]

Eli, Boruch, you are overreacting (both).

On 11/06/2021 13:19, Eli Zaretskii wrote:
> There's no need to
> introduce into Emacs features that are useful for a few people.

I think that expectation of users and developers in respect to support 
of locales evolves in time. Proper formatting of numbers is useful more 
widely then for a few people.

Boruch, till your last messages, I believed that you were convinced that 
adding support of "'" and "I" is not so easy.

Support of locale-dependent format specifiers through printf looks 
attractive but it can not be directly used by `format' or other elisp 
functions in a safe way.

Some code calling `format' implicitly expects that it generates 
locale-independent numbers, so changing its behavior is not backward 
compatible.

libc can only work with single global locale at any moment. I expect 
that attempt to "temporary" call setlocale(LC_NUMERIC, "") will be 
permanent source of bugs: forgotten reverting call, call of a function 
that needs universal format in locale-specific context, threads started 
at inappropriate moment, etc.

Another implementation of locale functions is necessary with ability to 
perform parsing and formatting without touching of global variables.

Personally I expect basic level functions with explicit locale context 
(random names):

    (locale-format-number-with-ctx
     (locale-get-current-context :group-separator 'suppress)
     1234567890)

or with explicit locale instead of `locale-get-current-context'. It is 
better to add some convenience helpers that inspect text properties, 
buffer-local and global settings to determine context:

    (locale-format-number 1234567890)

and maybe even `locale-format[-with-ctx]' that accepts printf-like 
format string.

On 11/06/2021 03:20, Boruch Baum wrote:
> Then don't make them locale specific. Implement the
> single-quote specifier the same way you currently handle the
> floating-point specifier '%f', a locale-specific format that
> has existed in emacs without complaint since ...

You are confusing something. "%f" is not locale-specific inside Emacs,
it uses "universal" format with dot "." as decimal separator even in
locales with "," in this role. At the same time "'" is highly
locale-dependent in libc. Group sizes and group separator widely
vary. I posted this example earlier:

LC_NUMERIC=C.UTF-8 /usr/bin/printf "%'d\n" 1234567890
1234567890
LC_NUMERIC=en_US.UTF-8 /usr/bin/printf "%'d\n" 1234567890
1,234,567,890
LC_NUMERIC=es_ES.UTF-8 /usr/bin/printf "%'d\n" 1234567890
1.234.567.890
LC_NUMERIC=ru_RU.UTF-8 /usr/bin/printf "%'d\n" 1234567890
1 234 567 890
LC_NUMERIC=en_IN.UTF-8 /usr/bin/printf "%'d\n" 1234567890
1,23,45,67,890

> It's not your responsibilty.
>
> I can say that in the use-case that prompted my request, I'm
> confident it will *never* be an issue. I ask format to give
> me a string and I display it. End of story. Whether just 99%
> or 99.99%, the overwhelming majority of cases will be the
> same. Your concerns are total non-issues.

I would prefer to avoid idiosyncrasy when "%'d" is locale-dependent but 
"%f" is not.

P.S.

With some limitation (printf binary is available and you do not need to 
work with floating point numbers), you can leverage libc formatting 
facilities with the following crutch:

(shell-command-to-string (format "/usr/bin/printf \"%%'d\" %d"
1234567890))