Re: New Linux build breakage: at-spi2-core

[Michael Forney <mforney%mforney.org@localhost>]

Benny Siegert <bsiegert%gmail.com@localhost> wrote:
> Hi!
> 
> I am currently running pkgrrxx on my Linux box, and I get this build 
> failure for devel/at-spi2-core:
> 
> [111/287] Compiling C object 
> tests/at-spi2-atk/dummyatk/libdummyatk.a.p/my-atk-action.c.o
> FAILED: [code=1] 
> tests/at-spi2-atk/dummyatk/libdummyatk.a.p/my-atk-action.c.o
> cc -Itests/at-spi2-atk/dummyatk/libdummyatk.a.p 
> -Itests/at-spi2-atk/dummyatk -I../tests/at-spi2-atk/dummyatk -I. -I.. 
> -Iatk -I../atk -I/usr/pkg/include/glib-2.0 
> -I/usr/pkg/lib/glib-2.0/include -I/usr/pkg/include 
> -I/usr/pkg/include/gio-unix-2.0 -I/usr/include 
> -I/usr/pkg/include/libxml2 -fdiagnostics-color=always 
> -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -std=c99 
> -DG_DISABLE_SINGLE_INCLUDES -DATK_DISABLE_SINGLE_INCLUDES -O2 -fPIC -MD 
> -MQ tests/at-spi2-atk/dummyatk/libdummyatk.a.p/my-atk-action.c.o -MF 
> tests/at-spi2-atk/dummyatk/libdummyatk.a.p/my-atk-action.c.o.d -o 
> tests/at-spi2-atk/dummyatk/libdummyatk.a.p/my-atk-action.c.o -c 
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c: In function 
> 'my_atk_action_description_get':
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c:124:10: error: implicit 
> declaration of function 'strdup'; did you mean 'g_strdup'? 
> [-Wimplicit-function-declaration]
>    124 |   return strdup (info->description);
>        |          ^~~~~~
>        |          g_strdup
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c:124:10: error: returning 
> 'int' from a function with return type 'const gchar *' {aka 'const char 
> *'} makes pointer from integer without a cast [-Wint-conversion]
>    124 |   return strdup (info->description);
>        |          ^~~~~~~~~~~~~~~~~~~~~~~~~~
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c: In function 
> 'my_atk_action_name_get':
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c:161:10: error: returning 
> 'int' from a function with return type 'const gchar *' {aka 'const char 
> *'} makes pointer from integer without a cast [-Wint-conversion]
>    161 |   return strdup (info->name);
>        |          ^~~~~~~~~~~~~~~~~~~
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c: In function 
> 'my_atk_action_localized_name_get':
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c:178:10: error: returning 
> 'int' from a function with return type 'const gchar *' {aka 'const char 
> *'} makes pointer from integer without a cast [-Wint-conversion]
>    178 |   return strdup (info->name);
>        |          ^~~~~~~~~~~~~~~~~~~
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c: In function 
> 'my_atk_action_get_keybinding':
> ../tests/at-spi2-atk/dummyatk/my-atk-action.c:207:10: error: returning 
> 'int' from a function with return type 'const gchar *' {aka 'const char 
> *'} makes pointer from integer without a cast [-Wint-conversion]
>    207 |   return strdup (info->keybinding);
>        |          ^~~~~~~~~~~~~~~~~~~~~~~~~
> 
> The file in question does include string.h. If I fix this file (by 
> replacing strdup with g_strdup), there are more errors about undefined 
> functions in other files.
> 
> Has anyone seen this? I never had build failures in this package before.

strdup wasn't part of ISO C until C23, but the compile flags include
-std=c99, which hides the strdup declaration by default.

The error is perhaps due to the recently added patch-meson.build
which removed -D_POSIX_C_SOURCE=200809L and -D_DEFAULT_SOURCE from
the compile flags.

See https://mail-index.netbsd.org/pkgsrc-users/2025/09/24/msg042013.html
for more discussion.

I believe c_std=c99 is coming from the top of meson.build
(`default_options: ['c_std=c99']`).

With -std=c99, adding -D_GNU_SOURCE would likely work on Linux, but
according to that mail, it doesn't cause timersub to be exposed on
NetBSD.