The commit https://git.savannah.gnu.org/gitweb/?p=gettext.git;a=commitdiff;h=e4a1c66c1ddb05fa67f00678bda9ec24af5e4de6 provides a fix, in the following sense:

• In C++ mode, where possible, overriding the *printf functions is now done through inline functions instead of through C macros.

• The "where possible" restriction applies to mingw and MSVC, which define some *printf functions as inline functions in a header file. Such inline functions can ONLY be overridden through C macros. In particular, this affects 'asprintf' on mingw.

• In order to nevertheless avoid a conflict with Qt, you can define a macro _INTL_NO_DEFINE_MACRO_ASPRINTF, that will inhibit that macro definition.

Oh, I completely misunderstood then. I thought it was both or neither. Thank you for that clarification!

Oh, I just realised that once we fix up this support, it would be very helpful for our programmers to make sure that they actually use the position markers where things need to be translated.

Are there any tools that can assist with enforcing that?

(sorry for the comment spam)

> if neither the programmer nor the translator use the position markers, will gettext implicitly add them in any scenario?

No, gettext will not add %1$ or such. It returns what the translator has provided in the PO file, assuming it has been checked through "msgfmt -c".

There is no tool that tests whether a translator has used this syntax, and frankly we don't want to have such a tool. It would be bad to tell the translators "you cannot reorder arguments, as required for your language, because one of the target platforms of the application is Windows and the programmer has not been able rectify this platform's printf deficiencies."

> I'm wondering why there *printf() replacements are unconditionally included, with no way to turn them off?

They are included on platforms that need them, in particular on native Windows.

The reason is that translators need the freedom to reorder arguments within format strings, see https://www.gnu.org/software/gettext/manual/html_node/c_002dformat-Flag.html

If a system's printf implementation does not support this, the output will be non-intelligible to users.

> we have convenience functions that take format strings in many places

You have to look whether the format string interpreter actually supports the %1$, %2$, etc. syntax.

gettext tries to redirect the various *printf() functions to its own implementation if the platform version doesn't comply to the POSIX behaviour.

Unfortunately, it does this using #define:s, which isn't very compatible with C++ and the namespacing there. It is more careful about C++ when handling the various *gettext() functions, but not here.

A specific scenario where this breaks is when using Qt's QString::asprintf():

D:/a/tigervnc/tigervnc/vncviewer/vncconnection.cxx:1727:24: error: 'libintl_asprintf' is not a member of 'QString'
 1727 |   infoText += QString::asprintf(_("Security method: %s\n"), rfb::secTypeName(m_securityType));
      |                        ^~~~~~~~
make[2]: * [vncviewer/CMakeFiles/vncviewer.dir/build.make:255: vncviewer/CMakeFiles/vncviewer.dir/vncconnection.cxx.obj] Error 1

More complete build log here:

https://github.com/CendioOssman/tigervnc/actions/runs/4626832263/jobs/8184028650