Document the internal _ and N_ macros
Commit Message
-----
Tested by building glibc on x86_64-linux-gnu.
include/libintl.h | 33 +++++++++++++++++++++++++--------
1 file changed, 25 insertions(+), 8 deletions(-)
Comments
On 5/6/20 5:23 AM, Florian Weimer wrote:
> -----
> Tested by building glibc on x86_64-linux-gnu.
>
Your comment indicates that N_ is deprecated, but doesn't list what we
have switched to. This doesn't follow your recommendation to Andreas,
which is to use some new macro?
I think this patch should include such a new macro, and switch all of
the N_ entries which we won't translate to that new macro, and mention
the use of the new macro in the comment below (leaving N_ for the
actual intended use).
> include/libintl.h | 33 +++++++++++++++++++++++++--------
> 1 file changed, 25 insertions(+), 8 deletions(-)
>
> diff --git a/include/libintl.h b/include/libintl.h
> index 9a11367ec6..4b0a0e8674 100644
> --- a/include/libintl.h
> +++ b/include/libintl.h
> @@ -37,17 +37,34 @@ extern char *__bind_textdomain_codeset (const char *__domainname,
> extern const char _libc_intl_domainname[];
> libc_hidden_proto (_libc_intl_domainname)
>
> -/* Define the macros `_' and `N_' for conveniently marking translatable
> - strings in the libc source code. We have to make sure we get the
> - correct definitions so we undefine the macros first. */
> +/* _ marks its argument, a string literal, for translation, and
> + performs translation at run time if the LC_MESSAGES locale category
> + has been set. The MSGID argument is extracted, added to the
> + translation database, and eventually submitted to the translation
> + team for processing. New translations are periodically
> + incorporated into the glibc source tree as part of translation
> + updates. */
> +# undef _
> +# define _(msgid) __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
>
> +/* N_ marks its argument, a string literal, for translation, so that
> + it is extracted and added to the translation database (similar to
> + the _ macro above). It does not translate the string at run time.
> + The first, primary use case for N_ is a context in which a string
> + literal is required, such as an initializer. Translation will
> + happen later, for example using the __gettext function.
> +
> + The second, historic, use case involves strings which may be
> + translated in a future version of the library, but cannot be
> + translated in current releases due to some technical limitation
> + (e.g., gettext not being available in the dynamic loader). No
> + translation at run time happens in such cases. This usage is
> + deprecated because using N_ for strings which are not translated
> + after all creates unnecessary work for the translation team: glibc
> + does not currently use those translations, and that may never
> + change. */
> # undef N_
> # define N_(msgid) msgid
>
> -# undef _
> -/* This is defined as an optimizing macro, so use it. */
> -# define _(msgid) \
> - __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
> -
> # endif /* !_ISOMAC */
> #endif
>
* Carlos O'Donell:
> On 5/6/20 5:23 AM, Florian Weimer wrote:
>> -----
>> Tested by building glibc on x86_64-linux-gnu.
>>
>
> Your comment indicates that N_ is deprecated, but doesn't list what we
> have switched to. This doesn't follow your recommendation to Andreas,
> which is to use some new macro?
We don't have consensus on what to do instead. I think there is
consensus that it's deprecated, though.
> I think this patch should include such a new macro, and switch all of
> the N_ entries which we won't translate to that new macro, and mention
> the use of the new macro in the comment below (leaving N_ for the
> actual intended use).
I think we should document what we have today, so that someone who
encounters these macros is less confused by them.
On 5/6/20 5:23 AM, Florian Weimer wrote:
> -----
> Tested by building glibc on x86_64-linux-gnu.
>
Please post v2 for review.
> include/libintl.h | 33 +++++++++++++++++++++++++--------
> 1 file changed, 25 insertions(+), 8 deletions(-)
>
> diff --git a/include/libintl.h b/include/libintl.h
> index 9a11367ec6..4b0a0e8674 100644
> --- a/include/libintl.h
> +++ b/include/libintl.h
> @@ -37,17 +37,34 @@ extern char *__bind_textdomain_codeset (const char *__domainname,
> extern const char _libc_intl_domainname[];
> libc_hidden_proto (_libc_intl_domainname)
>
> -/* Define the macros `_' and `N_' for conveniently marking translatable
> - strings in the libc source code. We have to make sure we get the
> - correct definitions so we undefine the macros first. */
> +/* _ marks its argument, a string literal, for translation, and
> + performs translation at run time if the LC_MESSAGES locale category
> + has been set. The MSGID argument is extracted, added to the
> + translation database, and eventually submitted to the translation
> + team for processing. New translations are periodically
> + incorporated into the glibc source tree as part of translation
> + updates. */
OK.
> +# undef _
> +# define _(msgid) __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
>
> +/* N_ marks its argument, a string literal, for translation, so that
> + it is extracted and added to the translation database (similar to
> + the _ macro above). It does not translate the string at run time.
> + The first, primary use case for N_ is a context in which a string
> + literal is required, such as an initializer. Translation will
> + happen later, for example using the __gettext function.
> +
> + The second, historic, use case involves strings which may be
> + translated in a future version of the library, but cannot be
> + translated in current releases due to some technical limitation
> + (e.g., gettext not being available in the dynamic loader). No
> + translation at run time happens in such cases.
OK up to here.
This usage is
> + deprecated because using N_ for strings which are not translated
> + after all creates unnecessary work for the translation team: glibc
> + does not currently use those translations, and that may never
> + change. */
Suggest:
In the future this historic usage of N_ may become deprecated. Strings
which are not translated create unnecessary work for the translation
team. We continue to use N_ because it helps mark translatable strings.
> # undef N_
> # define N_(msgid) msgid
>
> -# undef _
> -/* This is defined as an optimizing macro, so use it. */
> -# define _(msgid) \
> - __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
> -
> # endif /* !_ISOMAC */
> #endif
>
* Carlos O'Donell:
> On 5/6/20 5:23 AM, Florian Weimer wrote:
>> -----
>> Tested by building glibc on x86_64-linux-gnu.
>>
>
> Please post v2 for review.
>> + deprecated because using N_ for strings which are not translated
>> + after all creates unnecessary work for the translation team: glibc
>> + does not currently use those translations, and that may never
>> + change. */
>
> Suggest:
>
> In the future this historic usage of N_ may become deprecated. Strings
> which are not translated create unnecessary work for the translation
> team. We continue to use N_ because it helps mark translatable strings.
Thanks, v2 posted:
<https://sourceware.org/pipermail/libc-alpha/2020-May/113667.html>
@@ -37,17 +37,34 @@ extern char *__bind_textdomain_codeset (const char *__domainname,
extern const char _libc_intl_domainname[];
libc_hidden_proto (_libc_intl_domainname)
-/* Define the macros `_' and `N_' for conveniently marking translatable
- strings in the libc source code. We have to make sure we get the
- correct definitions so we undefine the macros first. */
+/* _ marks its argument, a string literal, for translation, and
+ performs translation at run time if the LC_MESSAGES locale category
+ has been set. The MSGID argument is extracted, added to the
+ translation database, and eventually submitted to the translation
+ team for processing. New translations are periodically
+ incorporated into the glibc source tree as part of translation
+ updates. */
+# undef _
+# define _(msgid) __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
+/* N_ marks its argument, a string literal, for translation, so that
+ it is extracted and added to the translation database (similar to
+ the _ macro above). It does not translate the string at run time.
+ The first, primary use case for N_ is a context in which a string
+ literal is required, such as an initializer. Translation will
+ happen later, for example using the __gettext function.
+
+ The second, historic, use case involves strings which may be
+ translated in a future version of the library, but cannot be
+ translated in current releases due to some technical limitation
+ (e.g., gettext not being available in the dynamic loader). No
+ translation at run time happens in such cases. This usage is
+ deprecated because using N_ for strings which are not translated
+ after all creates unnecessary work for the translation team: glibc
+ does not currently use those translations, and that may never
+ change. */
# undef N_
# define N_(msgid) msgid
-# undef _
-/* This is defined as an optimizing macro, so use it. */
-# define _(msgid) \
- __dcgettext (_libc_intl_domainname, msgid, LC_MESSAGES)
-
# endif /* !_ISOMAC */
#endif