possible bug with symlink following

[BurntSushi]

To reproduce:

$ git clone --depth 1 git:https://github.com/BurntSushi/linux linux-burntsushi
Cloning into 'linux-burntsushi'...
remote: Counting objects: 58881, done.
remote: Compressing objects: 100% (55208/55208), done.
remote: Total 58881 (delta 4182), reused 21851 (delta 2810), pack-reused 0
Receiving objects: 100% (58881/58881), 155.87 MiB | 8.87 MiB/s, done.
Resolving deltas: 100% (4182/4182), done.
Checking out files: 100% (55499/55499), done.
$ cd linux-burntsushi/
$ rg -n -u -tc -w -L '[A-Z]+_SUSPEND' | wc -l
403
$ LC_ALL=en_US.UTF-8 egrep -R -n --include='*.c' --include='*.h' -w '[A-Z]+_SUSPEND' | wc -l
403
$ LC_ALL=C egrep -R -n --include='*.c' --include='*.h' -w '[A-Z]+_SUSPEND' | wc -l
403
$ ucg --type=cc -w '[A-Z]+_SUSPEND' | wc -l
ucg: warning: './arch/powerpc/boot/dts/include/dt-bindings': recursive directory loop
ucg: warning: './arch/metag/boot/dts/include/dt-bindings': recursive directory loop
ucg: warning: './arch/arm64/boot/dts/include/dt-bindings': recursive directory loop
ucg: warning: './arch/arm/boot/dts/include/dt-bindings': recursive directory loop
ucg: warning: './arch/cris/boot/dts/include/dt-bindings': recursive directory loop
ucg: warning: './arch/mips/boot/dts/include/dt-bindings': recursive directory loop
391

In particular, I don't believe there are any recursive directory loops.

Are you perhaps not descending into the symlinked directory because it links to a directory that has already been searched? (Either way, it's at least not a loop.)

[BurntSushi]

Version info:

ucg --version
UniversalCodeGrep 0.3.0
Copyright (C) 2015-2016 Gary R. Van Sickle.

This program is free software; you can redistribute it and/or modify
it under the terms of version 3 of the GNU General Public License as
published by the Free Software Foundation.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program. If not, see http:https://www.gnu.org/licenses/.

Build info

Repo version: 0.3.0-4-g6867e8e

Compiler info:
 Name ($(CXX)): g++
 Version string: "g++ (GCC) 6.2.1 20160830"

ISA extensions in use:
 sse4.2: yes
 popcnt: yes

libpcre info:
 Version: 8.39 2016-06-14
 JIT support built in?: yes
 JIT target architecture: x86 64bit (little endian + unaligned)
 Newline style: LF

libpcre2-8 info:
 Version: 10.22 2016-07-29
 JIT support built in?: yes
 JIT target architecture: x86 64bit (little endian + unaligned)
 Newline style: LF

[gvansickle]

(note to self, that's current master).

Hey @BurntSushi ,

Coincidentally, I'm right in the middle of reworking the directory tree traversal, so this may get changed in the process. It's not really a bug though, but I suppose it depends on how you look at it. E.g. try this:

$ ucg --type=cc -w '[A-Z]+_SUSPEND' ~/src/TestCorpus/linux --dirjobs=1 | wc -l
404
$ ucg --type=cc -w '[A-Z]+_SUSPEND' ~/src/TestCorpus/linux --dirjobs=2 | wc -l
392

Now do something similar with grep (note the -R (follow symlinks) vs. -r (don't follow symlinks)):

$ egrep -R -n --include='*.c' --include='*.h' -w '[A-Z]+_SUSPEND' ~/src/TestCorpus/linux | wc -l
404
$ egrep -r -n --include='*.c' --include='*.h' -w '[A-Z]+_SUSPEND' ~/src/TestCorpus/linux | wc -l
390

Are you perhaps not descending into the symlinked directory because it links to a directory that has already been searched?

That's exactly what's going on, but as always, it's a bit more complicated than that. When running with more than 1 dir tree thread, I have to handle symlink loop detection myself. I do that by not re-visiting directories I've already visited. But with only one thread, the fts library I use is able to handle this for me; however, it appears to be detecting only real cycles, and not just previously-visited dirs. So in that case, any "forward-crosslinks" (gotta be an actual term for that) get visited more than once.

grep uses the same fts library, which similarly explains its behavior. If you do a physical traversal, you hit each file exactly once, and you get the lower number. If you do a logical traversal, you're safe from cycles, but any files on the other side of any forward-crosslinks will get visited multiple times.

I noticed this when I first went to multithreaded directory traversal, and made the decision that:

1. For the use-case at hand, it seems like not re-searching files that have already been searched is the expected and preferred behavior.
2. Since this would be the behavior with the default 2 threads, plus my (maybe naive) expectation that symlinks would be rare in the trees being searched, it wasn't worth worrying about the duplicate hits in the 1 thread case.

So it's not a bug of ommission.

(Either way, it's at least not a loop.)

Indeed. It probably would make sense to at least reword the warning in this case.

Thanks for the report Andrew. Looks like you've been pretty busy!

GRVS

[BurntSushi]

I admit, it's a little strange for the results returned to vary based on the number of threads you're using, but I can see why skipping previously visited directories might be helpful (if a bit surprising, based on the behavior of others tools).

That's exactly what's going on, but as always, it's a bit more complicated than that. When running with more than 1 dir tree thread, I have to handle symlink loop detection myself. I do that by not re-visiting directories I've already visited. But with only one thread, the fts library I use is able to handle this for me; however, it appears to be detecting only real cycles, and not just previously-visited dirs. So in that case, any "forward-crosslinks" (gotta be an actual term for that) get visited more than once.

I just recently added a lock-free parallel recursive directory iterator myself, which also of course has to handle symlinks manually, but it only detects true loops (which matches single threaded behavior). Rust being fairly new, I also had to write the single threaded iterator. :P They both check for symlink loops in exactly the same way.

[gvansickle]

This has been resolved since release 0.3.2 in a few ways:

• The message has been changed to "warning: '': already visited this directory, possible recursive directory loop?"
• Replacement of fts lib with hand-rolled DirTree, which behaves the same wrt this issue regardless of the number of threads.

Closing as resolved.