Re: High CPU use on BTRFS and some other details in the stable release package.

[Bernhard Voelker]

On 10/03/2016 04:30 PM, Steve wrote:
> I'm not sure what was at fault here but I had excessive CPU use with 
> findutils after switching from EXT4 to BTRFS in Void Linux.
> https://github.com/voidlinux/void-packages/issues/4829
> Is this because find isn't optimised for BTRFS, there's a problem with the 
> BTRFS kernel driver, or something else? I built the latest version of 
> findutils from the stable source package and had the same result. 

I don't see how/why find(1) should do something special for BTRFS;
it's just using standard kernel functions to traverse the file system.

> I resolved my personal requirements by switching to using mlocate and ag 
> instead, but also saw some other small errors
in the source package. Perhaps these other observations might be helpful for 
you?
> 
>  File: README-alpha : Says it's an alpha release when it's the latest stable 
> one.

Fixed with:

  [PATCH 1/3] maint: merge README-alpha into README

>  The entire ChangeLog-2013 content is duplicated in ChangeLog

Fixed with

  [PATCH 2/3] maint: remove ChangeLog-2013 from distribution tarball

>  Typo:  README-hacking: "Dejagnu is in fact optional, but it's strongly 
> recommened,"

Already fixed in:

  http://git.sv.gnu.org/cgit/findutils.git/commit/?id=9333e22e91b6

> Compiles in a standard build, but won't build debug version:
> findutils-4.6.0$ ./configure --enable-debug
> $ make
> 
> ftsfind.c: In function 'main':
> ftsfind.c:705:49: warning: passing argument 1 of 'ctime' from incompatible 
> pointer type
>    fprintf (stderr, "cur_day_start = %s", ctime (&options.cur_day_start));
>                                                  ^
> In file included from ../gl/lib/time.h:41:0,
>                  from ../gl/lib/sys/stat.h:44,
>                  from ftsfind.c:39:
> /usr/include/time.h:264:14: note: expected 'const time_t *' but argument is 
> of type 'struct timespec *'
>  extern char *ctime (const time_t *__timer) __THROW;
>               ^
> 
> Fixed that error:
> 
>  find/ftsfind.c
> 
> #ifdef DEBUG
>   fprintf (stderr, "cur_day_start = %s", ctime (&options.cur_day_start));
> #endif /* DEBUG */
> 
> Changed to:
>   fprintf (stderr, "cur_day_start = %s", ctime ((time_t 
> *)&options.cur_day_start.tv_sec));

Already fixed in

  http://git.sv.gnu.org/cgit/findutils.git/commit/?id=bcbfcb86be6b

... but there was a similar one in oldfind.c - fixed with:

  [PATCH 3/3] oldfind: fix compiler warnings

> But it seems to have an outdated binary which is only called in a debug build:
> mv -f .deps/ftsfind.Tpo .deps/ftsfind.Po
> gcc -std=gnu99  -g -O2   -o find ftsfind.o ./libfindtools.a ../lib/libfind.a 
> ../gl/lib/libgnulib.a       -lm -lm
> ../gl/lib/libgnulib.a(parse-datetime.o): In function `parse_datetime':
> /home/james/source/GNU/findutils/git/gnu/findutils/gl/lib/parse-datetime.y:1452:
>  undefined reference to `rpl_mktime'
> /home/james/source/GNU/findutils/git/gnu/findutils/gl/lib/parse-datetime.y:1484:
>  undefined reference to `rpl_mktime'
> /home/james/source/GNU/findutils/git/gnu/findutils/gl/lib/parse-datetime.y:1497:
>  undefined reference to `rpl_mktime'
> /home/james/source/GNU/findutils/git/gnu/findutils/gl/lib/parse-datetime.y:1519:
>  undefined reference to `rpl_mktime'
> collect2: error: ld returned 1 exit status
> make[3]: *** [Makefile:1810: find] Error 1
> make[3]: Leaving directory 
> '/home/steve/projects/gnu-findutils/2dbg/findutils-4.6.0/find'
> make[2]: *** [Makefile:1898: all-recursive] Error 1
> make[2]: Leaving directory 
> '/home/steve/projects/gnu-findutils/2dbg/findutils-4.6.0/find'
> make[1]: *** [Makefile:1706: all-recursive] Error 1
> make[1]: Leaving directory 
> '/home/steve/projects/gnu-findutils/2dbg/findutils-4.6.0'
> make: *** [Makefile:1646: all] Error 2
> 
> The source is different:
> gl/lib/parse-datetime.y
> 1452      Start = mktime (&tm);
> 1484              Start = mktime (&tm);
> 1497          Start = mktime (&tm);
> 1519          Start = mktime (&tm);
> 
> make only updates what it needs to, probably the stable package didn't get 
> built from a clean source. (Sometimes make clean misses stuff too.)
> Building from git repo would probably fix this, I didn't try that.

oops, I see this too.
However, the --enable-debug option has meanwhile been retired:

  http://git.sv.gnu.org/cgit/findutils.git/commit/?id=995151807658

Thanks & have a nice day,
Berny

0001-maint-merge-README-alpha-into-README.patch
Description: Text Data

0002-maint-remove-ChangeLog-2013-from-distribution-tarbal.patch
Description: Text Data

0003-oldfind-fix-compiler-warnings.patch
Description: Text Data