bug#65319: compilation error on Android (Termux)

bug#65319: compilation error on Android (Termux)

[Bruno Haible]

Hi,

Compiling a current Emacs checkout from today on Android, inside the Termux
app, produces a compilation error:

../../src/dired.c:1140:16: warning: call to undeclared function 'getpwent'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
  while ((pw = getpwent ()))
               ^
../../src/dired.c:1140:14: error: incompatible integer to pointer conversion assigning to 'struct passwd *' from 'int' [-Wint-conversion]
  while ((pw = getpwent ()))
             ^ ~~~~~~~~~~~

The reason is that the default Android API level in Termux is 24,
the Android <pwd.h> declares getpwent() only starting with API level 26:

  #if __ANDROID_API__ >= 26
  struct passwd* getpwent(void) __INTRODUCED_IN(26);
  ...

yet HAVE_GETPWENT and HAVE_ENDPWENT come out as 1.

I can see two possible fixes:
  a) use
       gl_CHECK_FUNCS_ANDROID([getpwent], [[#include <pwd.h>]])
     instead of
       AC_CHECK_FUNCS(... getpwent ...)
     in configure.ac,
  b) add
       AC_CHECK_DECLS([getpwent], [], [], [[
         #include <sys/types.h>
         #include <pwd.h>
         ]])
     to configure.ac, and a '&& HAVE_DECL_GETPWENT' in
     src/dired.c line 1137.

Bruno

bug#65319: compilation error on Android (Termux)

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Bruno Haible <bruno@clisp.org> writes:

> Hi,
>
> Compiling a current Emacs checkout from today on Android, inside the Termux
> app, produces a compilation error:
>
> ../../src/dired.c:1140:16: warning: call to undeclared function 'getpwent'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>   while ((pw = getpwent ()))
>                ^
> ../../src/dired.c:1140:14: error: incompatible integer to pointer conversion assigning to 'struct passwd *' from 'int' [-Wint-conversion]
>   while ((pw = getpwent ()))
>              ^ ~~~~~~~~~~~
>
> The reason is that the default Android API level in Termux is 24,
> the Android <pwd.h> declares getpwent() only starting with API level 26:
>
>   #if __ANDROID_API__ >= 26
>   struct passwd* getpwent(void) __INTRODUCED_IN(26);
>   ...
>
> yet HAVE_GETPWENT and HAVE_ENDPWENT come out as 1.
>
> I can see two possible fixes:
>   a) use
>        gl_CHECK_FUNCS_ANDROID([getpwent], [[#include <pwd.h>]])
>      instead of
>        AC_CHECK_FUNCS(... getpwent ...)
>      in configure.ac,
>   b) add
>        AC_CHECK_DECLS([getpwent], [], [], [[
>          #include <sys/types.h>
>          #include <pwd.h>
>          ]])
>      to configure.ac, and a '&& HAVE_DECL_GETPWENT' in
>      src/dired.c line 1137.
>
> Bruno

Thanks.  What does config.guess say in Termux?  And are you trying to
build the Android port on Android, or just Emacs itself?

P.S: our bug tracker intercepts all mail and resends individual messages
itself, so you must use the X-Debbugs-Cc header instead of the carbon
copy list when initially opening a bug.  If you don't, I get the message
you send to the bug tracker, instead of the message it resends.

Thanks.

bug#65319: compilation error on Android (Termux)

[Bruno Haible]

Po Lu wrote:
> What does config.guess say in Termux?

Both build-aux/config.guess and exec/config.guess return
  armv7l-unknown-linux-gnueabi
but I configure all packages with
  --host=armv7l-linux-androideabi
so that all tests of "$host_os" see that it's Android and not an
arm system with glibc.

> And are you trying to
> build the Android port on Android, or just Emacs itself?

I took the Emacs sources from git, generated the 'configure' file, and
am compiling it like I compile on other POSIX-like systems. The Emacs
specific configure options that I use are --without-all --without-x.

Bruno

bug#65319: compilation error on Android (Termux)

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Bruno Haible <bruno@clisp.org> writes:

> I took the Emacs sources from git, generated the 'configure' file, and
> am compiling it like I compile on other POSIX-like systems. The Emacs
> specific configure options that I use are --without-all --without-x.

OK, thanks; I'm surprised this configuration functions at all,
considering that it has never been tested.

Does Gnulib guarantee that the gl_CHECK_FUNCS_ANDROID stuff will
continue to exist in the future?  If so, I'm inclined towards using
that.

bug#65319: compilation error on Android (Termux)

[Bruno Haible]

Po Lu wrote:
> Does Gnulib guarantee that the gl_CHECK_FUNCS_ANDROID stuff will
> continue to exist in the future?  If so, I'm inclined towards using
> that.

Yes. This macro is used over 100 times in Gnulib, has not seen any
problem reports in the last 6 months, and is prefixed with 'gl_'
not '_gl_'. Therefore it's very unlikely to go away.

Note this macro's documentation:

...
dnl Thus, the question "does the OS have the function func" has three possible
dnl answers:
dnl   - yes, in all versions starting from the given API level,
dnl   - no, in no version,
dnl   - not in the given API level, but in a later version of Android.
...
dnl This macro sets two variables:
dnl   - gl_cv_onwards_func_<func>   to yes / no / "future OS version"
dnl   - ac_cv_func_<func>           to yes / no / no
dnl The first variable allows to distinguish all three cases.
dnl The second variable is set, so that an invocation
dnl   gl_CHECK_FUNCS_ANDROID([func], [[#include <foo.h>]])
dnl can be used as a drop-in replacement for
dnl   AC_CHECK_FUNCS([func]).

Bruno

bug#65319: compilation error on Android (Termux)

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Bruno Haible <bruno@clisp.org> writes:

> Po Lu wrote:
>> Does Gnulib guarantee that the gl_CHECK_FUNCS_ANDROID stuff will
>> continue to exist in the future?  If so, I'm inclined towards using
>> that.
>
> Yes. This macro is used over 100 times in Gnulib, has not seen any
> problem reports in the last 6 months, and is prefixed with 'gl_'
> not '_gl_'. Therefore it's very unlikely to go away.
>
> Note this macro's documentation:
>
> ...
> dnl Thus, the question "does the OS have the function func" has three possible
> dnl answers:
> dnl   - yes, in all versions starting from the given API level,
> dnl   - no, in no version,
> dnl   - not in the given API level, but in a later version of Android.
> ...
> dnl This macro sets two variables:
> dnl   - gl_cv_onwards_func_<func>   to yes / no / "future OS version"
> dnl   - ac_cv_func_<func>           to yes / no / no
> dnl The first variable allows to distinguish all three cases.
> dnl The second variable is set, so that an invocation
> dnl   gl_CHECK_FUNCS_ANDROID([func], [[#include <foo.h>]])
> dnl can be used as a drop-in replacement for
> dnl   AC_CHECK_FUNCS([func]).
>
> Bruno

Thanks.  I'll install the change you proposed in short order, and am
closing this bug.