How portable are the fts_*() functions? It will probably require some modifications to configure.ac and meson.build, along with some #ifdefs.

Would it be more portable to use nftw(), which is POSIX compatible? (Yes, with fts it seems more elegant, just one while()).

the manpage says "These functions are available in Linux since glibc2"

@karelzak I looked at nftw(). It's a cool API, but it has a big disadvantage, there isn't any way to pass an argument to the handler, and we need both fincore_control and fincore_state to pass.
Sure we could use a global variable, but it's not an approach I endorse usually.

@karelzak I looked at nftw(). It's a cool API, but it has a big disadvantage, there isn't any way to pass an argument to the handler, and we need both fincore_control and fincore_state to pass. Sure we could use a global variable, but it's not an approach I endorse usually.

Using fincore_state as an argument for fincore_name() doesn't make sense since it's only used for the file name. It would be better to move fincore_state into the function and use just name as the argument. This would also reduce the use of switch() in main(). See the patch below.

Anyway, I agree that FTS looks better (no callbacks). We can try to use it since fincore is a Linux-only tool, but it will require checking for fts_openinconfigure.acandmeson.buildIt's also an opportunity to encourage the use of the FTS interface., adding #ifdef HAVE_FTS_OPENto the code, and printingerrx()if not supported and--recursive` is requested. It's also an opportunity to encourage the use of the FTS interface :-)

From fded72bd8269206ad4208e9a2372898a35070eea Mon Sep 17 00:00:00 2001
From: Karel Zak <kzak@redhat.com>
Date: Wed, 16 Apr 2025 09:40:46 +0200
Subject: [PATCH] fincore: fincore_state refactoring
 
---
 misc-utils/fincore.c | 34 ++++++++++++----------------------
 1 file changed, 12 insertions(+), 22 deletions(-)
 
diff --git a/misc-utils/fincore.c b/misc-utils/fincore.c
index a2f240052..7a4b5185b 100644
--- a/misc-utils/fincore.c
+++ b/misc-utils/fincore.c
@@ -356,12 +356,12 @@ static int fincore_fd (struct fincore_control *ctl,
 /*
  * Returns: <0 on error, 0 success, 1 ignore.
  */
-static int fincore_name(struct fincore_control *ctl,
-                       struct fincore_state *st)
+static int fincore_name(struct fincore_control *ctl, const char *name)
 {
        int fd;
        int rc = 0;
        struct stat sb;
+       struct fincore_state _st = { .name = name }, *st = &_st;
 
        if ((fd = open (st->name, O_RDONLY)) < 0) {
                warn(_("failed to open: %s"), st->name);
@@ -381,15 +381,15 @@ static int fincore_name(struct fincore_control *ctl,
                        warn(_("failed ioctl to get size: %s"), st->name);
        } else if (S_ISREG(sb.st_mode)) {
                st->file_size = sb.st_size;
-       } else {
-               rc = 1;                 /* ignore things like symlinks
-                                        * and directories*/
        }
 
        if (!rc)
                rc = fincore_fd(ctl, fd, st);
 
-       close (fd);
+       close(fd);
+
+       if (!rc)
+               rc = add_output_data(ctl, st);
        return rc;
 }
 
@@ -538,24 +538,14 @@ int main(int argc, char ** argv)
        }
 
        for(; optind < argc; optind++) {
-               struct fincore_state st = {
-                       .name = argv[optind],
-               };
-
-               switch (fincore_name(&ctl, &st)) {
-               case 0:
-                       add_output_data(&ctl, &st);
+               rc = fincore_name(&ctl, argv[optind]);
+               if (rc)
                        break;
-               case 1:
-                       break; /* ignore */
-               default:
-                       rc = EXIT_FAILURE;
-                       break;
-               }
        }
 
-       scols_print_table(ctl.tb);
-       scols_unref_table(ctl.tb);
+       if (rc == 0)
+               scols_print_table(ctl.tb);
 
-       return rc;
+       scols_unref_table(ctl.tb);
+       return rc ? EXIT_FAILURE :  EXIT_SUCCESS;
 }
-- 
2.49.0

Anyway, I agree that FTS looks better (no callbacks).

Unfortunately musl does not implement FTS.

Anyway, I agree that FTS looks better (no callbacks).

Unfortunately musl does not implement FTS.

Yes, but if we all ignore FTS, it will never be adopted by other libcs. :-)

I have no strong opinion about it. Both ways (nftw with global ctl strcut; or FTS with errx on poor libcs) work for me.

Yes, but if we all ignore FTS, it will never be adopted by other libcs. :-)

For musl to adopt it I guess it would need to make it into POSIX first.
OTOH there are standalone implementations of FTS which can be used with musl.
If this is the util-linux policy, it's fine for me. I also have a WIP branch somewhere which got blocked on this.

@karelzak I like it, if you merge it I will rebase and fix my PR

bash-completion/fincore and misc-utils/fincore.1.adoc should be updated.

@karelzak I had to refactor your change because it broke the test by changing the behavour in a substantial way.
On file error, fincore prints a warning, handles the other files, and then exit with error.
With that change it stopped on first unreadable or non existing file.

@t-8ch the problem with the recursive test is that the scan order is not consistent among runs, making the output diff unreliable

It looks good (after the meson update). @t-8ch, is there any objection?

@glaubitz hi!

The repo is: https://github.com/teknoraver/util-linux.git

and the branch fincore_mu

I've fixed the test blindly, let me know if the pass on a 64k page size machine

The test fincore/count fails with the following diff:

(sid_ppc64-dchroot)glaubitz@perotto:~/util-linux-64k$ cat ./tests/diff/fincore/count 
--- /home/glaubitz/util-linux-64k/tests/output/fincore/count.aggregate  2025-04-25 07:15:05.073991097 +0000
+++ /home/glaubitz/util-linux-64k/tests/output/fincore/count    2025-04-25 07:15:05.841998281 +0000
@@ -53,13 +53,13 @@
 return value: 1
 0     0 i_dir/EMPTY_FILE
 return value: 0
-1  65535 i_dir/PAGESIZE_-1__incore_
+1 65535 i_dir/PAGESIZE_-1__incore_
 return value: 0
-1  65536 i_dir/JUST_PAGESIZE_incore_
+1 65536 i_dir/JUST_PAGESIZE_incore_
 return value: 0
 [ RECURSIVE SCAN ]
-    0      0 i_dir/EMPTY_FILE
-    1  65535 i_dir/PAGESIZE_-1__incore_
-    1  65536 i_dir/JUST_PAGESIZE_incore_
-PAGES   SIZE FILE
+    0     0 i_dir/EMPTY_FILE
+    1 65535 i_dir/PAGESIZE_-1__incore_
+    1 65536 i_dir/JUST_PAGESIZE_incore_
+PAGES  SIZE FILE
 return value: 0
(sid_ppc64-dchroot)glaubitz@perotto:~/util-linux-64k$

@teknoraver I guess --raw option may help you.

@glaubitz so it's just a whitespace corruption.
I used --raw as suggested by @masatake, it should pass now

@glaubitz so it's just a whitespace corruption. I used --raw as suggested by @masatake, it should pass now

There is still some diff left, unfortunately:

(sid_ppc64-dchroot)glaubitz@perotto:~/util-linux-64k$ cat ./tests/diff/fincore/count 
--- /home/glaubitz/util-linux-64k/tests/output/fincore/count.aggregate  2025-04-25 12:17:51.153632799 +0000
+++ /home/glaubitz/util-linux-64k/tests/output/fincore/count    2025-04-25 12:17:52.421647956 +0000
@@ -53,9 +53,9 @@
 return value: 1
 0     0 i_dir/EMPTY_FILE
 return value: 0
-1  65535 i_dir/PAGESIZE_-1__incore_
+1 65535 i_dir/PAGESIZE_-1__incore_
 return value: 0
-1  65536 i_dir/JUST_PAGESIZE_incore_
+1 65536 i_dir/JUST_PAGESIZE_incore_
 return value: 0
 [ RECURSIVE SCAN ]
 0 0 i_dir/EMPTY_FILE
(sid_ppc64-dchroot)glaubitz@perotto:~/util-linux-64k$

How about appending sed?

ts_check_prog "sed"
...
... | sort | sed -e 's/ \{2,\}/ /g'

@masatake I'd rather add --raw everywhere and make the test more stable.
I have an ARM Macbook here which is using 16k page table, and that will require more work and more padding calculation if we want to support it

@glaubitz I removed all spaces from the expected files and used --raw.
It should work now.

@glaubitz I removed all spaces from the expected files and used --raw. It should work now.

Yes, it works now:

---------------------------------------------------------------------
  All 340 tests PASSED
---------------------------------------------------------------------
``

I have turned 16k pages on on my ARM machine, and I've extended the test to work with such size too:

teknoraver@fedorarm:~/src/util-linux$ uname -a
Linux fedorarm 6.14.2-401.asahi.fc42.aarch64+16k #1 SMP PREEMPT_DYNAMIC Wed Apr 16 12:06:15 UTC 2025 aarch64 GNU/Linux

teknoraver@fedorarm:~/src/util-linux$ ./test_sysinfo pagesize
16384

teknoraver@fedorarm:~/src/util-linux$ tests/run.sh --show-diff fincore/count

-------------------- util-linux regression tests --------------------

                    For development purpose only.
                 Don't execute on production system!

       kernel: 6.14.2-401.asahi.fc42.aarch64+16k

      options: --show-diff \
               --srcdir=/home/teknoraver/src/util-linux/tests/.. \
               --builddir=/home/teknoraver/src/util-linux/tests/..

      fincore: count file contents in core    ... OK

---------------------------------------------------------------------
  All 1 tests PASSED
---------------------------------------------------------------------

This is the default on Apple ARM machines running Linux

The test failure in the distcheck CI job looks genuine.

I saw it but I can't understand:

/home/runner/work/util-linux/util-linux/util-linux-2.41/_build/sub/../../tests/ts/fincore/count: line 124: /home/runner/work/util-linux/util-linux/util-linux-2.41/_build/sub/tests/output/fincore/count.err.aggregate: Permission denied
I create output/fincore/count.err.aggregate as temporary file to compare the result, is this path read-only?
Other outputs are placed in output/fincore so it must be writable

Maybe replace cp "$TS_EXPECTED_ERR" "$AGG_ERR" with cat "$TS_EXPECTED_ERR" > "$AGG_ERR". But that is just a guess.

Hmm qemu-user armv7 is the only one failing now.
That build supports fts_open(), but the output with --recursive is empty, and the return value is 0.

--- /home/runner/work/util-linux/util-linux/tests/output/fincore/count.aggregate	2025-04-28 22:41:38.297718592 +0000
+++ /home/runner/work/util-linux/util-linux/tests/output/fincore/count	2025-04-28 22:41:47.448702913 +0000
@@ -58,8 +58,4 @@
 1 4096 i_dir/JUST_PAGESIZE_incore_
 return value: 0
 [ RECURSIVE SCAN ]
-0 0 i_dir/EMPTY_FILE
-1 4095 i_dir/PAGESIZE_-1__incore_
-1 4096 i_dir/JUST_PAGESIZE_incore_
-PAGES SIZE FILE
 return value: 0

This is weird, how can I reproduce such test locally?

You don't have to insist on reproducing the issue locally.
It's painful, but can be debugged in a CI/CD environment; you can run any code from ts/fincore/count for debugging.

To shorten the debugging cycle, disable all CI/CD environments other than qemu-user armv7.
e.g. #3205

I think that the problem is a bad behaviour of fts_read() with the combination {linux, glibc 2.39, armhf).
I added a configure test which checks that fts_read() really works, and disables recursive otherwise.

OK. I think the configure.ac test is overkill :-) (don't forget that the user can update libc, etc.)

It would be better to always compile if FTS is available, add the check to tests/helpers/test_sysinfo.c, and use it in the tests/ts/fin core/count to define TS_KNOWN_FAIL="yes".

@teknoraver sorry for all the extra work, it's more complicated than I had originally expected ;-)

@karelzak it works, the armv7 test just skipped it as

fincore: count file contents in core    ... SKIPPED (fts_open() is unreliable on this setup)