Opposite to other bin utilities such as ls(1), ps(1) always respects terminal settings even if it's stdout is not a terminal (pipe, file). You should manually specify terminal settings ignorance to ps, but this behavior is counter-intuitive. Fix: Workaround is to add "ww" to ps cmd line args in every ps call pipelined or file redirected. How-To-Repeat: resize xterm window to small number of columns (i.e, 10) and try to do: ps aux | grep something you'll won't get anything matched even if you'll get it matched on a wider terminal
What you describe is standard BSD ps(1) behaviour, as is documented in the manual page. Basically, the intent is that, when you redirect ps(1) output somewhere (to a file or pipe), you get exactly the same that is displayed on the screen. The rules are: If ps(1) is running in a terminal, it's width is used. If it's not running in a terminal (e.g. via a cron job) or the width cannot be determined, the default is 80 columns. To determine the terminal width, any one of stdout, stderr or stdin is used -- therefore, to let ps(1) ignore the terminal width, you have to redirect stderr and stdin. If you specify the -w option once, the limit is expanded to 132 columns (unless your terminal width is already larger than that). Finally, if you specify -w twice, the width is assumed to be unlimited. When you use ps(1) in a pipe to grep for things, you should always use the -ww options. Also note that newer versions of FreeBSD (> 4.x) have a pgrep(1) command which can be used as a convenient replacement for ps | grep pipes. Best regards Oliver -- Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing Dienstleistungen mit Schwerpunkt FreeBSD: http://www.secnetix.de/bsd Any opinions expressed in this message may be personal to the author and may not necessarily reflect the opinions of secnetix in any way. I suggested holding a "Python Object Oriented Programming Seminar", but the acronym was unpopular. -- Joseph Strout
Hello Oliver, Wednesday, December 21, 2005, 11:00:06 PM, you wrote: OF> What you describe is standard BSD ps(1) behaviour, as is OF> documented in the manual page. It's not clarified there in the case of redirection. OF> Basically, the intent is that, when you redirect ps(1) output OF> somewhere (to a file or pipe), you get exactly the same that OF> is displayed on the screen. This is counter-intuitive. If I am redirecting output to file, I expect to see result somewhat later. It's somewhat strange that result will be not stable constant, but unpredictable, depending on terminal settings at the moment of running. OF> The rules are: If ps(1) is running in a terminal, it's width OF> is used. If it's not running in a terminal (e.g. via a cron OF> job) or the width cannot be determined, the default is 80 OF> columns. To determine the terminal width, any one of stdout, OF> stderr or stdin is used -- therefore, to let ps(1) ignore the OF> terminal width, you have to redirect stderr and stdin. OF> If you specify the -w option once, the limit is expanded to OF> 132 columns (unless your terminal width is already larger OF> than that). Finally, if you specify -w twice, the width is OF> assumed to be unlimited. Compare it with ls(1) behavior. If output is not to a terminal, ls behaves as expected, not using any terminal crap. And this is clearly documented in ls man page. OF> When you use ps(1) in a pipe to grep for things, you should OF> always use the -ww options. Also note that newer versions of FreeBSD (>> 4.x) have a pgrep(1) command which can be used OF> as a convenient replacement for ps | grep pipes. Yes, but grep is not the only thing used with redirection. Moreover, I've tried to run "ps aux > myprocs.txt" on Linux (Slackware 10) and it worked as I expected - full lines in file while on terminal they were truncated (and linux have pgrep(1) also). So, I insist that ps(1) behavior should be corrected. OK, developers may have another opinion, but in that case at least ps man page should be fixed, adding a paragraph clarifying this, somewhat like what you explained earlier in this post. -- WBR, Vadim Goncharov FidoNet 2:5005/106.426 ICQ#166852181
Vadim Goncharov wrote: > Oliver Fromme wrote: > OF> What you describe is standard BSD ps(1) behaviour, as is > OF> documented in the manual page. > > It's not clarified there in the case of redirection. Redirection of stdout does not change the behaviour of ps, so it need not be mentioned. Almost all tools in the base system don't care what their stdout is, and there is no reason to document that fact in all of the manual pages. Note that ls(1) is an exception, not ps(1). That's why it is documented in the ls manpage. > OF> Basically, the intent is that, when you redirect ps(1) output > OF> somewhere (to a file or pipe), you get exactly the same that > OF> is displayed on the screen. > > This is counter-intuitive. For me, it is intuitive. When I type "ps" in the shell, look at the output and then decided to use grep, I expect it to work on the same kind of data. In my opinion, tools that behave differently depending on their stdout are counter-intuitive. > If I am redirecting output to file, I expect > to see result somewhat later. It's somewhat strange that result will be > not stable constant, but unpredictable, depending on terminal settings > at the moment of running. As I pointed out, that's the reason why the -ww options exist. You cannot change that without breaking existing scripts. And without breaking existing admins. ;-) > OF> The rules are: If ps(1) is running in a terminal, it's width > OF> is used. If it's not running in a terminal (e.g. via a cron > OF> job) or the width cannot be determined, the default is 80 > OF> columns. To determine the terminal width, any one of stdout, > OF> stderr or stdin is used -- therefore, to let ps(1) ignore the > OF> terminal width, you have to redirect stderr and stdin. > OF> If you specify the -w option once, the limit is expanded to > OF> 132 columns (unless your terminal width is already larger > OF> than that). Finally, if you specify -w twice, the width is > OF> assumed to be unlimited. > > Compare it with ls(1) behavior. If output is not to a terminal, ls > behaves as expected, not using any terminal crap. Unless you specify -C. However, if you ask me, the behaviour of ls(1) is counter-intuitive and should be fixed. Problem is that it would break existing scripts (and admins) and violate SUSv3/POSIX2001. > OF> When you use ps(1) in a pipe to grep for things, you should > OF> always use the -ww options. Also note that newer versions > of FreeBSD (>> 4.x) have a pgrep(1) command which can be used > OF> as a convenient replacement for ps | grep pipes. > > Yes, but grep is not the only thing used with redirection. Moreover, > I've tried to run "ps aux > myprocs.txt" on Linux (Slackware 10) and it > worked as I expected - full lines in file while on terminal they were > truncated (and linux have pgrep(1) also). The ps of Linux behaves differently because it's a different ps. Historically, there are two different families of ps implementations: SysV and BSD. Both are incompatible and have different options. Linux chose to adapt the SysV behaviour, with some compatibility hacks for BSD. You'll notice the "bad syntax" warning when you type "ps -aux". Therefore you cannot take Linux' ps for comparison. > So, I insist that ps(1) behavior should be corrected. OK, developers may > have another opinion, but in that case at least ps man page should be > fixed, adding a paragraph clarifying this, somewhat like what you > explained earlier in this post. Personally I think that the manpage explains the behaviour sufficiently. But adding a little clarification probably won't hurt. Especially beginners often forget to use -ww in scripts. For example, a sentence could be added to the paragraph at the beginning which explains the default output format: "Note that the command is truncated to the terminal width; see the -w option to change that." Best regards Oliver -- Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing Dienstleistungen mit Schwerpunkt FreeBSD: http://www.secnetix.de/bsd Any opinions expressed in this message may be personal to the author and may not necessarily reflect the opinions of secnetix in any way. "One of the main causes of the fall of the Roman Empire was that, lacking zero, they had no way to indicate successful termination of their C programs." -- Robert Firth
Hello Oliver, Thursday, December 22, 2005, 7:41:53 PM, you wrote: >> OF> What you describe is standard BSD ps(1) behaviour, as is >> OF> documented in the manual page. >> >> It's not clarified there in the case of redirection. OF> Redirection of stdout does not change the behaviour of ps, OF> so it need not be mentioned. Almost all tools in the base OF> system don't care what their stdout is, and there is no OF> reason to document that fact in all of the manual pages. Because these tools (almost all of base system) don't care about terminal at all, always assuming "dumb wide printer" stdout. For example, try "df -i" - you'll get garbled output on standard 80-column terminal with wrapped lines. Because they're always care about redirection giving full output, and not your terminal. OF> Note that ls(1) is an exception, not ps(1). That's why it OF> is documented in the ls manpage. What about top(1) ? >> OF> Basically, the intent is that, when you redirect ps(1) output >> OF> somewhere (to a file or pipe), you get exactly the same that >> OF> is displayed on the screen. >> >> This is counter-intuitive. OF> For me, it is intuitive. When I type "ps" in the shell, OF> look at the output and then decided to use grep, I expect OF> it to work on the same kind of data. Yeap, but I expect work on _full_ data, as with all other tools not respecting terminal. OF> In my opinion, tools that behave differently depending on OF> their stdout are counter-intuitive. If so, all tools which at least see at what stdout is (terminal or not) are breaking unix-way because they must simply stream data out, not preparing it for terminal at all. >> If I am redirecting output to file, I expect >> to see result somewhat later. It's somewhat strange that result will be >> not stable constant, but unpredictable, depending on terminal settings >> at the moment of running. OF> As I pointed out, that's the reason why the -ww options OF> exist. You cannot change that without breaking existing OF> scripts. And without breaking existing admins. ;-) What existing scripts and admins would break situation of not needing to add -ww to cmd line in case of redirect? Any real facts? >> Yes, but grep is not the only thing used with redirection. Moreover, >> I've tried to run "ps aux > myprocs.txt" on Linux (Slackware 10) and it >> worked as I expected - full lines in file while on terminal they were >> truncated (and linux have pgrep(1) also). OF> The ps of Linux behaves differently because it's a different OF> ps. Historically, there are two different families of ps OF> implementations: SysV and BSD. Both are incompatible and OF> have different options. Linux chose to adapt the SysV OF> behaviour, with some compatibility hacks for BSD. You'll OF> notice the "bad syntax" warning when you type "ps -aux". OF> Therefore you cannot take Linux' ps for comparison. I can, because stdout problem is absolutely unrelated to other implementation functionality. >> So, I insist that ps(1) behavior should be corrected. OK, developers may >> have another opinion, but in that case at least ps man page should be >> fixed, adding a paragraph clarifying this, somewhat like what you >> explained earlier in this post. OF> Personally I think that the manpage explains the behaviour OF> sufficiently. But adding a little clarification probably OF> won't hurt. Especially beginners often forget to use -ww OF> in scripts. Yes, I preferably meant beginners at first when did send-pr. OF> For example, a sentence could be added to the paragraph at OF> the beginning which explains the default output format: OF> "Note that the command is truncated to the terminal width; OF> see the -w option to change that." No, this is too short. It should be more detailed one- or two-paragraph explanation, like the one you gave in your first response to PR. -- WBR, Vadim Goncharov FidoNet 2:5005/106.426 ICQ#166852181 mailto:vadim_nuclight@mail.ru
My thoughts on issue: I cannot think of an example where including more information on the *same line * will break any scripts. Attached low-impact patch to implement "correct" behaviour. cd /usr/src patch < ps.wide-stdout.patch
For bugs matching the following criteria: Status: In Progress Changed: (is less than) 2014-06-01 Reset to default assignee and clear in-progress tags. Mail being skipped
Fixed in 2017 (r314685) in FreeBSD 12. See also Bug 217159.