Including both linux/time.h and linux/input.h causes failure

Including both linux/time.h and linux/input.h causes failure

[Elichai Turkel]

Hi,
The following header: <wrapper.h>:
```
#include <linux/input.h>
#include <linux/time.h>
```
Will cause the compiler to fail because of redefinition of a lot of
time related structs, that are declared once in `linux/time.h` and
then again via `bits/types/struct_timeval.h` which is included through
`linux/input.h > sys/time.h -> bits/types/struct_timeval.h`



Command: `gcc -I./headers_install/x86/include wrapper.h`
output:
```
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:16:8: error: redefinition
of ‘struct timeval’
   16 | struct timeval {
      |        ^~~~~~~
In file included from /usr/include/sys/time.h:25,
                 from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/bits/types/struct_timeval.h:8:8: note: originally defined here
    8 | struct timeval
      |        ^~~~~~~
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:21:8: error: redefinition
of ‘struct timezone’
   21 | struct timezone {
      |        ^~~~~~~~
In file included from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/sys/time.h:52:8: note: originally defined here
   52 | struct timezone
      |        ^~~~~~~~
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:30: warning: "ITIMER_REAL" redefined
   30 | #define ITIMER_REAL  0
      |
In file included from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/sys/time.h:92: note: this is the location of the previous
definition
   92 | #define ITIMER_REAL ITIMER_REAL
      |
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:31: warning:
"ITIMER_VIRTUAL" redefined
   31 | #define ITIMER_VIRTUAL  1
      |
In file included from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/sys/time.h:95: note: this is the location of the previous
definition
   95 | #define ITIMER_VIRTUAL ITIMER_VIRTUAL
      |
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:32: warning: "ITIMER_PROF" redefined
   32 | #define ITIMER_PROF  2
      |
In file included from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/sys/time.h:99: note: this is the location of the previous
definition
   99 | #define ITIMER_PROF ITIMER_PROF
      |
In file included from wrapper.h:13:
./headers_install/x86/include/linux/time.h:39:8: error: redefinition
of ‘struct itimerval’
   39 | struct itimerval {
      |        ^~~~~~~~~
In file included from ./headers_install/x86/include/linux/input.h:13,
                 from wrapper.h:8:
/usr/include/sys/time.h:104:8: note: originally defined here
  104 | struct itimerval
      |        ^~~~~~~~~

```

-- 
PGP: 5607C93B5F86650C

Re: Including both linux/time.h and linux/input.h causes failure

[Carlos O'Donell]

On 1/29/20 10:14 AM, Elichai Turkel wrote:
> Hi,
> The following header: <wrapper.h>:
> ```
> #include <linux/input.h>
> #include <linux/time.h>
> ```
> Will cause the compiler to fail because of redefinition of a lot of
> time related structs, that are declared once in `linux/time.h` and
> then again via `bits/types/struct_timeval.h` which is included through
> `linux/input.h > sys/time.h -> bits/types/struct_timeval.h`
 
This is another header-coordination issue between the kernel and libc.

In this case it's linux/time.h vs. sys/time.h.

"Synchronizing Headers"
https://sourceware.org/glibc/wiki/Synchronizing_Headers
- Just added your case.

It's not immediate clear to me if the UAPI header is clean enough to
use directly in glibc or not, and that's often the simplest way to fix
things.

When you can't just use the kernel header definitions then you have
to work through the conflict and decide if you want to allow both headers
to be potentially included and in which orders to support it.

Why do you need both headers included?

-- 
Cheers,
Carlos.

Re: Including both linux/time.h and linux/input.h causes failure

[Elichai Turkel]

Mostly I need the uid_t type for syscalls(getuid/setuid)
Which is defined in sys/types.h but that header also includes time
related stuff.

For context I'm writing a libc equivilant library (in rust).

Thanks,
Elichai.

On Wed, Jan 29, 2020 at 7:16 PM Carlos O'Donell <codonell-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
>
> On 1/29/20 10:14 AM, Elichai Turkel wrote:
> > Hi,
> > The following header: <wrapper.h>:
> > ```
> > #include <linux/input.h>
> > #include <linux/time.h>
> > ```
> > Will cause the compiler to fail because of redefinition of a lot of
> > time related structs, that are declared once in `linux/time.h` and
> > then again via `bits/types/struct_timeval.h` which is included through
> > `linux/input.h > sys/time.h -> bits/types/struct_timeval.h`
>
> This is another header-coordination issue between the kernel and libc.
>
> In this case it's linux/time.h vs. sys/time.h.
>
> "Synchronizing Headers"
> https://sourceware.org/glibc/wiki/Synchronizing_Headers
> - Just added your case.
>
> It's not immediate clear to me if the UAPI header is clean enough to
> use directly in glibc or not, and that's often the simplest way to fix
> things.
>
> When you can't just use the kernel header definitions then you have
> to work through the conflict and decide if you want to allow both headers
> to be potentially included and in which orders to support it.
>
> Why do you need both headers included?
>
> --
> Cheers,
> Carlos.
>


-- 
PGP: 5607C93B5F86650C