Re: Format specifier (re)ordering in Python (was: String change for orca)

[Willie Walker <William Walker Sun COM>]

Hi Wouter:

From my experiences with the Q_ vs. C_ change, for example, I saw 
translators being overly aggressive in their translations, translating 
stuff the comments explicitly told them not to translate.  Hence, I 
definitely supported the change to C_ because it helped prevent this.

At the risk of translators being more likely to get "%s level %d" right 
and "%(role)s level %(level)d" wrong, however, I think I'd prefer to 
reserve the use of format specifiers where disambiguation is necessary.

But, if the gnome-i18n team makes a decree that thou shalt always use 
format specifier reordering, perhaps a rule of thumb might be to use 
cryptic parameter names, like:

_("%(p1)s blah %(p2)d") % {'p1' : 'foo', 'p2' : 3 }

Will

Wouter Bolsterlee wrote:
> Hi all,
>
> The "format specifiers should be in a different order" problem is not new,
> and not specific to Python. Let's look into this specific case once more:
>
> 2009-02-11 klockan 20:58 skrev Willie Walker:
> > At the request of our Hungarian translator, we've changed a string in  
> > Orca.  The old string was this:
> >
> > #. Translators: this is in reference to a heading level
> > #. in HTML (e.g., For <h3>, the level is 3).
> > #.
> > #: ../src/orca/scripts/toolkits/Gecko/speech_generator.py:90
> > #, python-format
> > msgid "level %d"
> >
> > The new string is this:
> >
> > #. Translators: the %d is in reference to a heading
> > #. level in HTML (e.g., For <h3>, the level is 3)
> > #. and the %s is in reference to a previously
> > #. translated rolename for the heading.  If you
> > #. change the order of the %s and %d in the string
> > #. (as needed for Hungarian, for example), Orca will
> > #. detect it and do the right thing.
> > #.
> > #: ../src/orca/scripts/toolkits/Gecko/speech_generator.py:93
> > #, python-format
> > msgid "%s level %d"
>
> While this may work in this specific case, this is by no means a generally
> acceptable solution to this kind of problem. What happens if both arguments
> are strings? Or both are numbers? Or if you had 4 format specifiers? Well,
> it would break horribly. So we need something more reliable :)
>
> In C, sprintf() supports reordering format specifiers using "%n$f", where n
> is a number. Python does not support the "%1$d" syntax to its built-in
> sprintf-like string formatting operator, %, but it does support named
> parameter substitution when a mapping (e.g. a dictionary instance) is passed
> to the % operator.
>
> This may sound cryptic, so let's illustrate this with a few examples:
>
>   >>> 'the role is %(role)s and the level is %(level)d' % {'role': 'test', 'level': 3}
>   'the role is test and the level is 3'
>
> If translators decide the order has to be changed, this will result in
> something like this:
>
>   >>> 'the level is %(level)d and the role is %(role)s' % {'role': 'test', 'level': 3}
>   'the level is 3 and the role is test'
>
> This approach works in all cases, even in the case that the two format
> specifiers are the same (e.g. both are numbers).
>
> There is one downside though: translators MUST NOT translate the keywords
> ("level" and "role" in my case), since those are the keys used to lookup the
> value in the mapping passed to the % operator. But then, translators can
> just as well mess up format specifiers that are used in the traditional,
> positional way, e.g. by translating '%.3f' to %.0f', which will result in
> the fraction being removed upon display. So... a short translator notice
> should be fine to avoid this problem.
>
> Hope this helps others as well.
>
>     — Wouter