Re: Handling qt-format in gettext tools (was: Updating qt-format handling?)

[Bruno Haible]

Chusslove Illich wrote on 2007-04-19 and 2007-07-11
<http://lists.gnu.org/archive/html/bug-gnu-utils/2007-07/msg00030.html>:
> The messages with placeholders (format strings) in KDE4 are going to be 
> proper qt-format (to the point of its own well-definedness), so we'd like 
> to have the POs pass msgfmt -c in the future.
> 
> However, qt-format specs in Gettext are a bit outdated, probably due to its 
> low usage (KDE didn't use it before, and there is also the native 
> QtLinguist format for pure Qt apps).
> 
> The current specifications for placeholder replacement in QString are given 
> here:
> 
> http://doc.trolltech.com/4.3/qstring.html#arg
> http://doc.trolltech.com/4.3/qstring.html#arg-2
> 
> As I read (and tested), it is that:
> 
> 1. The limit on placeholder numbers are no longer 1-9, but 1-99.
> 
> 2. There is no special requirement on all placeholders being unique. 
> Equal-numbered placeholders are allowed, and will be substituted with the 
> same argument.
> 
> Points 1 and 2 are there to amend the following assertion by the Gettext 
> manual:
> 
> "...a directive consists of a `%' followed by a digit. The same directive 
> cannot occur more than once in a format string."

The written specification is not enough. You also need to look into the
source code. As far as I can see, the initial comment in format-qt.c needs
to be changed as follows:

***************
*** 31,44 ****
  #define _(str) gettext (str)
  
  /* Qt format strings are processed by QString::arg and are documented in
!    qt-3.0.5/doc/html/qstring.html.
!    A directive starts with '%' and is followed by a digit ('0' to '9').
!    Each %n must occur only once in the given string.
     The first .arg() invocation replaces the %n with the lowest numbered n,
     the next .arg() invocation then replaces the %n with the second-lowest
     numbered n, and so on.
!    (This is inherently buggy because a '%' in the first replacement confuses
!    the second .arg() invocation.)
     Although %0 is supported, usually %1 denotes the first argument, %2 the
     second argument etc.  */
  
--- 31,51 ----
  #define _(str) gettext (str)
  
  /* Qt format strings are processed by QString::arg and are documented in
!    qt-4.3.0/doc/html/qstring.html.
!    A directive
!      - starts with '%',
!      - is optionally followed by 'L' (indicates locale-dependent processing),
!      - is followed by one or two digits ('0' to '9'). %0n is equivalent to %n.
!    An unterminated directive ('%' or '%L' at the end) is not an error.
     The first .arg() invocation replaces the %n with the lowest numbered n,
     the next .arg() invocation then replaces the %n with the second-lowest
     numbered n, and so on.
!    This is inherently buggy because a '%' in the first replacement confuses
!    the second .arg() invocation.
!    To reduce this problem and introduce another one, there are also .arg()
!    methods that take up to 9 strings and perform the replacements in one 
swoop.
!    But this method works only on strings that contain no 'L' flags and only
!    single-digit argument designators.
     Although %0 is supported, usually %1 denotes the first argument, %2 the
     second argument etc.  */
  

So, effectively we have two kinds of Qt format strings: those that contain
no 'L' flags and only single-digit argument designators, and those that have
'L' flags or two-digit argument designators. The translator must not give a
translation of the second kind for an msgid of the first kind.
Maybe it can still be handled through a single 'format-qt' attribute,
through the format_check function?

> And now suggestions for some heuristics:
> 
> 3. Since same-numbered placeholders represent same piece of data, I would 
> not insist on strict congruence between the number of same-numbered 
> placeholders in msgid and msgstr. It may be that the msgid uses %1 twice 
> due to sentence structure, whereas the translation can get away with a 
> single %1 and a pronoun -- or vice versa. It is enough if each placeholder 
> number from msgid exists in msgstr at least once.

Yes, I agree. Qt3 had a limitation "Each %n must occur only once in the
given string." Qt4 does not have this limitation any more.

> 4. How about letting qt-format use the same heuristic for plural messages 
> as the c-format? Eg. not having to mention the placeholder in msgstr[x], 
> if [x] is used only for the case strictly n==1 (I'm not sure what the full 
> heuristic here is, but since noone complains for the c-format itself...)

Yes, this sounds good. The 'equality' parameter to format_check contains the
information whether to do strict checking, or loose checking for the case
you mention.

> If not too time-intensive, could the maintainer himself make these changes 
> for the next release? *looks at Bruno with puppy eyes*

As you can see, I haven't found the time to do these changes in 4 months.
I would appreciate a patch that takes into account the finding for 1. + 2.
mentioned above.

Bruno