[BusyBox] The mount rewrite is in.

[Rob Landley]

On Wednesday 17 August 2005 03:15, Thomas Fröhlich wrote:

> > P.S.  We haven't used CVS for over a year.
>
> Oh, I meant subversion, don't be so closely ;) I think, you knew what I
> meant.
> I'm at revision 11182

Just making sure. :)

> I have in .config:
> CONFIG_FEATURE_BUFFERS_USE_MALLOC=y
> ...
> CONFIG_FEATURE_SH_IS_ASH=y
> # CONFIG_FEATURE_SH_IS_HUSH is not set
> # CONFIG_FEATURE_SH_IS_LASH is not set
> # CONFIG_FEATURE_SH_IS_MSH is not set
> # CONFIG_FEATURE_SH_IS_NONE is not set
> ...
> CONFIG_ASH=y
> CONFIG_MOUNT=y
> CONFIG_UMOUNT=y
>  # CONFIG_FEATURE_MTAB_SUPPORT is not set

Yeah, just built with that, and it worked.  What compiler version are you 
using?  (I'm using gcc 3.3.5.)

> compilation failed with
> busybox/util-linux/util-linux.a(umount.o)(.text+0x2b8): In function
>
> `umount_main':
> : undefined reference to `erase_mtab'

And for me, it didn't.

Notice that this is a link-time error.  Although this function has a 
prototype, all the calls to it are optimized out by 
ENABLE_FEATURE_MTAB_SUPPORT.  (In the case of the umount call, the local 
variable is initialized to ENABLE_FEATURE_MTAB_SUPPORT and constant 
propogation zaps it.)  There is no place where that variable is ever set to a 
nonzero value when ENABLE_FEATURE_MTAB_SUPPORT is 0, and the compiler I'm 
using is smart enough to pick up on that.

Apparently yours isn't.  Now I can make a stub for it compilers, but that's 
another thing that would go in a quarrantine.h file to support C libraries 
and old compilers...

> if I clean the
>  #ifconfig CONFIG_FEATURE_MTAB_SUPPORT
>  #endif
> in mtab.c it compiles...

You need it even though both of the call sites are designed to let it be 
optimized out using compiler optimization techniques that were well-known 
more than 10 years ago...

> I didn't think there is a problem on the shell...
> It's great for you, if it works fine, but I had the same problem on a new
> checkout...I give you my .config, perhaps a uClibc problem?!?

It's not a library problem, it's a compiler problem.  When we get to link 
time, there are still calls to a function that can never be called.

I can add a broken_compilers.h file that does a version test and adds an empty 
version of this function under the #ifdef...

> regards
> Thomas

Rob