Bugfix for procps

[Arthur Demchenkov]

Hello everyone.

Here is the patch that fixes two bugs in watch program from procps-3.2.7: 1) watch -n doesn't handle large integers properly 2) watch ignores multibyte characters (was pointed in locale related issues chapter of BLFS book earlier)

It's adapted from Fedora Core procps source rpm to LFS.

[alexander@…]

As far as I can see:

• The part about closing the fd 0, reopening it from /dev/null and asserting that the new fd is really 0 doesn't have any relation to the two abovementioned bugs, it just makes sure that the executed program receives nothing on stdin.

• The rest of the patch is a straightforward rewrite of the old logic using wide characters - i.e., the fix for (2). It is definitely good, but should be rechecked for corner cases like a Chinese character at the end of line, and correct display of the title.

• I don't see how issue (1) is fixed, or a testcase.

[Arthur Demchenkov]

You are right. The first patch is some kind of atavism. I removed it.

I'm reattaching the properly patch to make watch to work with unicode.

[alexander@…]

Still, you didn't explain why you dropped the "close fd 0" part. I didn't say that it is bad - I only noted the inaccurate patch description.

As for the unicode patch, it is definitely better than nothing, but I still haven't tested it. From looking at the code, it seems that it mishandles zero-width characters.

[Arthur Demchenkov]

Hello, Alexander.

As I said before the first patch is compilation of two patches from Fedora Core 6 src.rpm

The first one is called procps-3.1.15-misc.patch (that's about fd0)

The second on is called procps-3.2.7-watch-unicode.patch

But the problem was the second patch needed the first one. So I checked rpm's spec file for occurence of 'watch' and found 2 bugs (the first one was in procps-3.2.6 so my description was incorrect as this bug was fixed in procps-3.2.7):

* Thu Apr 26 2007 Tomas Smetana <tsmetana@redhat.com> 3.2.7-9
- fix #183029 - watch ignores multibyte characters
* Thu Jun  1 2006 Karel Zak <kzak@redhat.com> -> 3.2.6-4 <-
- fix #191493 - watch -n doesn't handle large integers properly

So after your message I checked that watch -n 12345678901234 <some command> works as expected without the first patch and I removed it. But if you know why watch needs it you are free to apply it. Looking at its name it comes from ancient version of procps-3.1.15.

I'm attaching original source converted to cpio.

[Arthur Demchenkov]

Replaced with tar.bz2 without procps-3.2.7.tar.gz because of limited size of attachments

[robert@…]

I can't reproduce the 'watch -n' bug at ​https://bugzilla.redhat.com/show_bug.cgi?id=191493.
Anyone else?

[alexander@…]

The "watch -n" bug was relevant for versions 3.2.6 and below. procps 3.2.7 silently caps the interval to 4294 seconds. So we only have the unicode-related bug.

[Arthur Demchenkov]

Replying to alexander@linuxfromscratch.org:

The "watch -n" bug was relevant for versions 3.2.6 and below. procps 3.2.7 silently caps the interval to 4294 seconds. So we only have the unicode-related bug.

Yes, unicode-related bug is fixed with procps-3.2.7-watch-unicode.patch. Do you have any contraindication on applying it?

[alexander@…]

Please apply it, as a better patch doesn't exist yet.

[randy@…]

This is marked for the next release, but a consensus has not even been reached on this issue. We need to revisit this as a 6.4 ticket.

[randy@…]

Added the watch_unicode patch to the Procps instructions in r8636.