[Toybox] [CLEANUP] stat

Rob Landley rob at landley.net
Mon May 27 22:28:42 PDT 2013

On 05/26/2013 03:38:44 AM, Felix Janda wrote:
> Hello,
> some "cleanup" raising the SLOC...
> The first patch makes stat support multiple files, fixes a format
> string, inlines check_file_type() and adds functions do_stat()
> do_statfs(), which do the system call and some preparations, to be
> called via function pointers from main().
> The second patch mainly inlines print_stat_format() splitting of
> the big switch into print_stat() and print_statfs() to be called via
> function pointers from main(). A nice side effect is that it now won't
> segfault if you do something like "stat -f -c %B ."
> For me this finishes the cleanup of stat.

Yay! Applied both, and let's see...

Ok, first thing: clean up the help text. I realize what's there is  
copied verbatim from the man page, but that man page sucks.  
("modification time" vs "change time"?) Took a bit of finagling to fit  
it in 80x24, but just made it.

GLOBALS() indent was still tab, change to two spaces. And I tend to put  
a blank line between options lib/args.c automatically fills out and  
normal globals.

We never do anything with date_stat_format() but immediately print it,  
might as well make the function do it.

The types[] array in do_stat() is a rough edge. Hmmm... there's no else  
case that sets the type in case it was unknown (such as 0). In theory,  
this never happens. In practice it means I can cheat slightly, given  
this observation:

   $ find linux -name stat.h | xargs grep 'S_IF[A-Z]*[ \t]'
   linux/include/uapi/linux/stat.h:#define S_IFMT  00170000
   linux/include/uapi/linux/stat.h:#define S_IFSOCK 0140000
   linux/include/uapi/linux/stat.h:#define S_IFLNK  0120000
   linux/include/uapi/linux/stat.h:#define S_IFREG  0100000
   linux/include/uapi/linux/stat.h:#define S_IFBLK  0060000
   linux/include/uapi/linux/stat.h:#define S_IFDIR  0040000
   linux/include/uapi/linux/stat.h:#define S_IFCHR  0020000
   linux/include/uapi/linux/stat.h:#define S_IFIFO  0010000

I.E. the only place the I_IFBLAH constants occur a stat.h header in  
current linux code is in the generic stuff, it doesn't vary per target.  
(The access permission bits are actually subtly standardized in posix  
due to the command line arguments to chmod, although I'm sure cygwin  
finds a way to break. But the type fields, not so much. But linux has  
to be binary compatible with itself foreverish, and that's all I really  
care about.)

So, we have ALMOST have this going by twos, except there's no 8 and  
there is a 1. so let's make the 1 the default, feed a blank string into  
the 8... No, duh: octal. So it's actually 2, 4, 6, 8, 10, 12. So make  
the loop look like:

   filetype = statf->st_mode & S_IFMT;
   TT.ftname = types;
   for (i = 1; filetype != (i*8192) && i < 7; i++)
     TT.ftname += strlen(TT.ftname)+1;

Yes that's linux-specific, and I think I'm ok with that.

Printing all zeroes and pretending that's nanosecond resolution...  
either support it or don't. Let's see, supporting it is  
stat->st_atim.tv_nsec and similar... no mention of nanoseconds in  
strftime() (et tu, posix2008?) so pass it as a second argument and  
append it by hand... (Need to test that against musl...)

When we hit an unknown type in print_it() we print the literal  
character, which is right for %% but what about an unknown option?

   $ stat -c %q /

Eh, I guess that's a "don't care". It didn't die with an error, that's  
the important thing.

I have a horrible idea for compressing the switch/case blocks, but  
should probably check this in and get some sleep for right now...


More information about the Toybox mailing list