Patrick Welche <prlw1@cam.ac.uk>

[netbsd-mini2440.git] / libexec / cron / MAIL

[ this is really old mail that came to me in response to my 1986 posting
  to usenet asking for feature suggestions before releasing the first 
  version of cron.  it is presented here for its entertainment value.
  --vix ]

$Id: MAIL,v 1.1 1994/01/05 20:40:11 jtc Exp $

From ptsfa!lll-crg!ames!acornrc!bob Wed Dec 31 10:07:08 1986
Date: Wed, 31 Dec 86 08:59:31 pst
From: lll-crg!ames!acornrc!bob (Bob Weissman)
To: ptsfa!vixie!paul
Status: RO

Sure, here's a suggestion: I'd like to be able to run a program, say,
every two hours.  Current cron requires me to write
0,2,4,6,8,10,12,14,16,18,20,22 in the hours field.  How about a notation
to handle this more elegantly?

<<      Okay, I've allowed 0-22/2 as a means of handling this.
        The time specification for my cron is as follows:
                specification = range {"," range}
                range = (start "-" finish ["/" step]) | single-unit
        This allows "1,3,5-7", which the current cron doesn't (it won't
        do a range inside a list), and handles your specific need.      >>

From drw@mit-eddie Wed Dec 31 18:25:27 1986
Date: Wed, 31 Dec 86 14:28:19 est
From: drw@mit-eddie (Dale Worley)
To: mit-eddie!vixie!paul
Status: RO

We have a lot of lines in our crontab of the form

        00 12 * * * su user < /usr/users/user/script.file

This barfs (silently!) on our system (Dec Ultrix 1.2 == 4.2bsd) if
user's shell is csh.  This, I am told, is because csh requires that
the environment be set up in certain ways, which cron doesn't do.
(Actually, I believe, it is because /etc/rc, which runs cron, doesn't
set up the environment enough for csh to run, and cron just inherits
the situation.)  Anyway, the point is that if you find out what csh
really needs in its environment, you might want to set up cron to
provide some reasonable defaults (if it isn't supplied by cron's
parent).  Also, could you tell me what csh needs, if you find out, so
we can hack our /etc/rc?

<<      well, the environment IS a problem. processes that cron forks
        will inherit the environment of the person who ran the cron
        daemon... I plan to edit out such useless things as TERMCAP,
        TERM, and the like; supply correct values for HOME, USER, CWD,
        and whatever else comes to mind.  I'll make sure csh works...   >>
From ptsfa!ames!seismo!dgis!generous Thu Jan  1 07:33:17 1987
Date: Thu Jan 1 10:29:20 1987
From: ames!seismo!dgis!generous (Curtis Generous)
To: nike!ptsfa!vixie!paul
Status: RO

Paul:

One of the limitations of the present versions of cron is the lack
of the capability of specifying a way to execute a command every
n units of time.

Here is a good example:

# Present method to start up uucico
02,12,22,32,42,52 * * * *       exec /usr/lib/uucp/uucico -r1

# New method ?? (the ':' here is just one possibility for syntax)
02:10 * * * *                   exec /usr/lib/uucp/uucico -r1

This method would prove very helpful for those programs that get started
every few minutes, making the entry long and not easily readable.  The first
number would specify the base time, and the second number the repetition
interval.

<<      Good idea, but bob@acornrc beat you to it.  I used '/' instead of
        ':'.  This is my personal preference, and seems intuitive when you
        think of the divide operator in C... Does anyone have a preference? >>

From ptsfa!lll-lcc!seismo!decuac!c3pe!c3engr!charles Thu Jan  1 17:04:24 1987
From: lll-lcc!seismo!c3pe!c3engr!charles (Charles Green)
To: c3pe!decuac!dolqci!vrdxhq!seismo!lll-lcc!ptsfa!vixie!paul
Date: Thu Jan  1 19:22:47 1987
Status: RO

Well, this isn't a compatible extension, but I have in times past wondered
about a facility to let you start a process at intervals of, say, 17 minutes,
instead of particular minutes out of each hour.

<<      This was a popular request!     >>

From seismo!uwvax!astroatc!nicmad!norvax!mann Sun Jan  4 13:04:01 1987
Date: Fri, 2 Jan 87 09:23:53 cst
From: lll-lcc!seismo!uwvax!astroatc!nicmad!norvax!mann (Tom Mann)
To: ptsfa!vixie!paul
Status: RO

I'm not sure if it is in cron (either SysV or BSD ... if it is, I haven't
figured it out ) but a comment feature would SURE BE NICE!.
There are times when I want to comment out an entry
for a period of time; it might also make it a lot more legible.

<<      My cron allows blank lines and standard #-type comments.  I know
        that one BSD4.2 cron I've used had it.  I don't know about SysV.  >>

From ptsfa!hoptoad!hugh Mon Jan  5 10:26:46 1987
Date: Mon, 5 Jan 87 01:22:17 PST
From: hoptoad!hugh (Hugh Daniel)
To: ptsfa!vixie!paul
Status: RO

  Hi, I do have a BIG one that I would like.  I want to log ALL output
from command lines into a file for each line.  Thus I might have a chance
of finding out why my crontab entry did not work.
  This would seem to work best if done by cron, as it is now I have a google
of shell scripts laying about just to put the error output where I can see
it.

<<      My cron (and the SysV cron) will send mail to the owner of the
        particular crontab file if a command generates any output on stdout
        or stderr.  This can be irritating, but if you write a script such
        that any output means a problem occurred, you can avoid most logfile
        needs, and not generate mail except in unforeseen circumstances.   >>

From ptsfa!dual!ucbvax!ihnp4!anvil!es!Robert_Toxen Mon Jan  5 13:08:46 1987
From: dual!ucbvax!ihnp4!anvil!es!Robert_Toxen
Date: Fri,  2 Jan 87 14:25:29 EST
To: anvil!ihnp4!ucbvax!dual!ptsfa!vixie!paul
Status: RO

Here are some suggestions:
1. Run it through the C preprocessor via "/lib/<whatever>".

<<      hmmm. this seems of limited utility, and if you really wanted
        to do it that way, you could do it yourself (since users can
        write to their own crontab files).  I'll add '-' (read stdin)
        to the crontab installer program to facilitate this.            >>

2. Allow specifying every Nth day of week, i.e., every second Wednesday.
   I did this to calendar by separating the day of week (Wed=4, which one
   to start on and N with slashes).  I took modulo the day of year as a
   starting point so that someone with a desk calendar documenting such
   things can easily determine the offset (second number).  I did this
   while at SGI; alas I don't have a copy of the code.

<<      I can see how this could be useful, but I'm not sure how I'd
        implement it.  Cron currently doesn't keep track of the last time
        a given command was run; whether the current Wednesday is the first
        or second since the command was last run would be pretty hard to
        figure out.  I'd have to keep a database of commands and their
        execution around, and purge it when the crontab was overwritten.
        This is too much work for me, but if someone adds it, let me know.  >>

From ptsfa!ames!seismo!cbmvax!devon!paul Tue Jan  6 05:50:17 1987
From: ames!seismo!cbmvax!devon!paul
To: cbmvax!seismo!nike!ptsfa!vixie!paul
Date: Mon Jan  5 09:29:57 1987
Status: RO

One problem that has always plagued me with cron is the assumed ORing.
I'd like to see some type of ANDing implemented.  I guess I can best
describe this by example.  Say I have the following line in my crontab
file:

*  *  4-31  *  1-6      /usr/bin/command

What this does is run 'command' on the 4th thru 31st days of the
month, AND on Monday thru Saturday; which probably means running it
every day of the month (unless Sunday falls on days 1-3).  This
happens because cron runs the command if the day-of-month OR the
day-of-week is true.

What I'd like to happen with the above line is to run the command ONLY
on Monday thru Saturday any time after the 3rd of the month, e.g. if
the day-of-month AND the day-of-week are true.

My proposal to you is to implement some special chars for the first
five fields.  Examples:

*  *  !1-3  *  1-6      /usr/bin/command

(run command Mon-Sat, but NOT [!] on the first 3 days of the month)

*  *  &4-31 *  &1-6     /usr/bin/command

(run command if day-of-month AND day-of-week are true)

Get the picture?  This would be compatable with existing versions of
cron (which wouldn't currently be using any special characters, so
that old crontabs would be handled correctly).

<<      This message made me aware of the actual boolean expression involved
        in a crontab entry.  I'd assumed that it was
                (minute && hour && DoM && month && DoW)
        But it's really
                (minute && hour && month && (DoM || DoW))

        I can see some value in changing this, but with a fixed order of
        fields, operators get to be kindof unary, which && and || really
        aren't.  If someone has an idea on a syntax that allows useful
        variations to the standard (&& && && (||)) default, please suggest. >>

From bobkat!pedz Tue Jan  6 20:02:10 1987
From: pedz@bobkat.UUCP (Pedz Thing)
Date: 2 Jan 87 17:34:44 GMT
Status: RO

Log files!  It would be nice to be able to specify a log for cron
itself and also a log for each program's stdout and stderr to go to.
The latter can of course be done with > and 2> but it would be nice if
there could be a single line with some sort of pattern like
`> /usr/spool/log/%' and the command would be substituted for the %.
Another thing which would be nice is to be able to specify which shell
to call to give the command to.

<<      Log files are done with mail.  The '%' idea could be useful if
        a different character were used (% is special to cron, see man
        page); a different directory would have to be chosen, since each
        user has their own crontab file; and something intelligent would
        have to be done in the file naming, since the first word of the
        command might be ambiguous (with other commands).  In short, it's
        too much work.  Sorry.                                            >>

From guy%gorodish@sun Tue Jan  6 20:03:13 1987
From: guy%gorodish@sun (Guy Harris)
Message-ID: <10944@sun.uucp>
Date: 5 Jan 87 12:09:09 GMT
References: <429@vixie.UUCP> <359@bobkat.UUCP>
Sender: news@sun.uucp
Status: RO

> Another thing which would be nice is to be able to specify which shell
> to call to give the command to.

Well, the obvious choice would be the user's shell, but this wouldn't work
for accounts like "uucico".

<<      I use the owning user's shell, and to handle "uucico" I check a
        list of "acceptable shells" (currently compiled in, does anybody
        mind?), substituting a default (compiled in) shell if the user's
        shell isn't on the list.

        BTW, "compiled in" means that it's in a .h file, easily changed
        during installation, but requiring recompilation to modify.  You
        don't have to go digging through the code to find it...           >>

From qantel!hplabs!ucbvax!mwm@violet.berkeley.edu Tue Jan  6 21:24:48 1987
To: hplabs!qantel!vixie!paul (Paul Vixie Esq)
Date: 04 Jan 87 00:42:35 PST (Sun)
From: Mike Meyer <mwm@violet.berkeley.edu>
Status: RO

<<[Discussion of RMS/FSF, and mwm's GNU Cron deleted]>>

Oh, yeah - here are the extensions on my cron:

1) Sunday is both day 0 and day 7, so it complies with both SysV and
BSD cron.

<<      Good idea. I did it too, thanks for informing me.       >>

2) At is integrated into the cron. Instead of atrun to scan the
/usr/spool/at directory, at files are put into the /usr/lib/cron
directory along with users cron files, and cron fabricates a line from
a crontab file to run them. This is considered a major win by all who
use it.

<<      I don't use 'at', and my cron doesn't do anything with it.  To run
        'at', I use 'atrun' the same way the current BSD cron does.  My
        crontab files are in /usr/spool/cron/crontabs, in the SysV
        tradition -- not in /usr/lib/cron.  This is a configuration
        parameter, of course.                                           >>

There are two known restrictions:

1) I don't support any of the SysV security hooks. I don't have a use
for them, and RMS didn't like the idea at all :-).

<<      This means cron.allow and cron.deny.  I plan to support them, as
        they've been quite helpful at various HPUX sites I've administered. >>

2) Cron expects to be able to create files with names longer than 14
characters, which makes it hard to run on SysV. At least one person
was working on a port, but I don't know how it's going. That might
make for a good reason for releasing yours, right there.

<<      If someone has SysV (with the 14-character limit), they probably
        won't want my cron, since it doesn't add much to the standard
        version (which they may have support for).  My cron is not currently
        portable to non-BSD systems, since it relies on interval timers (I
        needed to sleep for intervals more granular than seconds alone would
        allow).  The port would be trivial, and I will do it if a lot of
        people ask for it...                                            >>

Oh, yeah - I'm going to see about getting this cron integrated into
the next 4BSD release.

<<      How does one go about this?  I have a few nifty gadgets I'd like
        to contribute, this cron being one of them...                   >>

<<[more FSF/GNU discussion deleted]>>

From qantel!hplabs!ames!ut-sally!ut-ngp!melpad!bigtex!james Tue Jan  6 21:24:57 1987
Posted-Date: Fri, 2 Jan 87 19:26:16 est
Date: Fri, 2 Jan 87 19:26:16 est
From: hplabs!ames!ut-sally!ut-ngp!bigtex!james
To: vixie!paul
Status: RO

Yes!!!  There are several critical failures in System V cron...

1. Pass all variables in cron's environment into the environment of things
   cron starts up, or at least into the crontab entries started up (at jobs
   will inherit the environment of the user).  If nothing else it is critically
   important that the TZ variable be passed on.  PATH should be passed on too.
   Basically, passing environment values allows one to design a standard
   environment with TZ and PATH and have that run by everything.  If anyone
   tells you this is no big deal, consider what happens when uucico is
   started by cron in CA to make a long distance phone link...  Unless the
   administrator is really on his/her toes, calls scheduled at 5pm will really
   go at two in the afternoon, needlessly incurring huge phone bills, all
   because System V refuses to pass the TZ from its environment down.  There
   are work arounds, but only putting it in cron will really work.  This is
   not a security hole.

<<      delete TERM and TERMCAP; modify HOME, USER, and CWD; pass TZ and
        PATH through undisturbed.  any other requests out there?

        BSD doesn't have this problem -- TZ is passed right on through if
        you define it in the shell before starting my cron daemon.  However,
        the BSD I'm running this on doesn't need TZ to be defined anyway...
        The default in the kernel has been just fine so far...  But just the
        same, if/when I port to SysV (I guess I really should), I'll make
        sure this works right.

        I guess I've been spoiled.  HPUX is SysV-based, and I never had a
        problem with cron and TZ when I used it.                          >>

2. A way to avoid logging stuff in /usr/lib/cron/log.  I have a cron entry
   run uudemon.hr every 10 minutes.  This is 144 times/day.  Each run generates
   three lines of text, for a total of 432 lines of text I don't want to see.
   Obviously this should be optional, but it would be nice if there were a
   way to flag an entry so that it wasn't logged at all unless there was an
   error.

<<      I don't know nothin' 'bout no /usr/lib/cron/log.  What is this file?
        I don't see any reason to create log entries, given the mail-the-
        output behaviour.  Opinions, anyone?                            >>

I will come up with other ideas no doubt, but I can always implement them
myself.

<<      That's what I like about PD software.  Please send me the diffs!  >>

The other problem you have is making sure you can run standard
crontabs.  I would suggest something like this: if the command part of the
entry starts with an unescaped -, then flags and options follow immediately
thereafter.  As in:

2,12,22,32,42,52 * * * * -q /usr/lib/uucp/uudemon.hr

This could mean do not log the uudemon.hr run unless there is a problem of
some kind.  This is probably safe as not many filenames start with "-", and
those that do are already a problem for people.

<<      Since I don't plan on supporting /usr/lib/cron/log in ANY form unless
        many people request it, I won't be needing -q as you've defined it.
        I could use something like this to avoid sending mail on errors, for
        the occasional script that you don't want to bullet-proof.

        The compatibility issue is CRITICAL.  The 4.3BSD crontab format is
        a crime against the whole philosophy of Unix(TM), in my opinion.   >>

One other minor thing to consider is the ulimit: can different users get
different ulimits for their crontab entries?

<<      Boy I'm ignorant today.  What's a ulimit, and what should I do with
        it in a crontab?  Suggestions, enlightenment, etc ??               >>

From qantel!lll-crg!ames!uw-beaver!uw-nsr!john Tue Jan  6 23:32:44 1987
Date: Thu, 1 Jan 87 10:53:05 pst
From: lll-crg!ames!uw-beaver!uw-nsr!john (John Sambrook 5-7433)
To: vixie!paul
Status: RO

How about not hardwiring the default environment that cron builds for its
children in the cron program itself?  Our cron does this and it's the pits
because we are TZ=PST8PDT not TZ=EST5EDT !

<<      yeachk.  I assure you, I will not hardwire the TZ!              >>
From ptsfa!well!dv Fri Jan  9 04:01:50 1987
Date: Thu, 8 Jan 87 23:50:40 pst
From: well!dv (David W. Vezie)
To: ptsfa!vixie!paul
Status: RO

6, have a special notation called 'H' which would expand to weekends
   and holidays (you'd have to keep a database somewhere of real
   holidays), and also 'W' for workdays (neither weekend or holiday).

<<      Too much work.  There should be a standard way to define and
        detect holidays under Unix(TM); if there were, I'd use it.  As
        it is, I'll leave this for someone else to add.

        I can see the usefulness; it just doesn't quite seem worth it.    >>
From qantel!gatech!akgua!blnt1!jat Wed Jan 14 20:00:40 1987
Date: Tue, 13 Jan 87 16:39:38 EST
From: gatech!akgua!blnt1!jat
Status: RO

1) Add some way to tell cron to reread the files, say kill -1 <pid>

<<      whenever the 'crontab' program is run and updates a crontab file,
        a file /usr/spool/cron/POKECRON is created; next time the cron
        daemon wakes up, it sees it, and re-reads the crontab files.

        I thought of handling the signal; even implemented it.  Then this
        clever idea hit me and I ripped it all out and added a single
        IF-statement to handle the POKECRON file.                       >>

2) Have some kind of retry time so that if a command fails, cron will try to
   execute it again after a certain period.  This is useful if you have some
   type of cleanup program that can run at the scheduled time for some reason
   (such as locked device, unmounted filesystem, etc).

<<      Hmmm, sounds useful.  I could do this by submitting an 'at' job...
        I'll think about it.                                            >>
From ptsfa!dual!ucbvax!ihnp4!mtuxo!ender Sat Jan  3 16:54:00 1987
From: dual!ucbvax!ihnp4!mtuxo!ender
Date: Sat, 3 Jan 87 14:05:13 PST
To: ucbvax!dual!ptsfa!vixie!paul
Status: RO

It would be nice if nonprivileged users can setup personal crontab files
(~/.cronrc, say) and be able to run personal jobs at regular intervals.
        
<<      this is done, but in the SysV style: the 'crontab' program installs
        a new crontab file for the executing user (can be overridden by root
        for setup of uucp and news).  the advantage of this is that (1) when
        a crontab is changed, the daemon can be informed automatically; and
        (2) the file can be syntax-checked before installation.         >>
From ptsfa!ames!seismo!ihnp4!lcc!richard Fri Jan 16 04:47:33 1987
Date: Fri, 16 Jan 87 07:44:57 EST
To: nike!ptsfa!vixie!paul
Status: RO

The System V cron is nice, but it has a few annoying features.  One is that
its mail files will say that the previous message is the output of "one of your
cron commands."  I wish it would say WHICH cron commmand.

<<      Done.  Also which shell, which user (useful when the mail gets
        forwarded), which home directory, and other useful crud.        >>

Another problem is with timezones.  It is necessary to specify TZ=PST8PDT (or
whatever) when you invoke cron (from inittab, or /etc/rc) and it is also
necessary to add TZ=PST8PDT to each crontab line which might need it.  Cron
should automatically export its idea of the "TZ" to each invoked command, and
it should be possible to put a line in the crontab file which overrides that
for every command in the file (e.g., most users are on EST, so cron is run
with TZ=EST5EDT; but one user is usually on PST and wants all of his cron
commands to run with TZ=PST8PDT).  This might be extended to allow any
environment variable to be specified once for the whole crontab file (e.g.,
PATH).

<<      Well, since I run the user's shell, you could put this into .cshrc.
        generic environment-variable setting could be useful, though.  Since
        I have to modify the environment anyway, I'll consider this.      >>

A log file might be a nice idea, but the System V cron log is too verbose.
I seem to remember that cron keeps it open, too; so you can't even have
something go and periodically clean it out.

<<      I don't do /usr/lib/cron/log.  I wasn't aware of this file until I
        got all these suggestions.  Do people want this file?  Tell me!    >>