jbr0871@fedeva.UUCP (Blaine Robertson) (12/07/88)
Can someone please tell me how 'tail -f' works?. Obviously, it does not do a busy loop while doing a stat on the file. I know that it has got to be blocked, waiting for the length to change, but how does it get notified of a change? Thanks in advance for any responses. -- Blaine Robertson Federal Express, Memphis, TN [..!mirror!premise, ..gatech!emcard]!fedeva!jbr0871
gwyn@smoke.BRL.MIL (Doug Gwyn ) (12/07/88)
In article <412@fedeva.UUCP> jbr0871@fedeva.UUCP (Blaine Robertson) writes: >Can someone please tell me how 'tail -f' works?. Obviously, it does not >do a busy loop while doing a stat on the file. All it does is to sleep for a short time (around one second), then copy out everything past the last high water mark via read/write. There is no statting, no asynchronous notification, no nonblocking I/O, etc. It IS a "busy loop" (except for the short sleep).
jds@mimsy.UUCP (James da Silva) (12/08/88)
In article <412@fedeva.UUCP> jbr0871@fedeva.UUCP (Blaine Robertson) writes: >Can someone please tell me how 'tail -f' works?. Obviously, it does not >do a busy loop while doing a stat on the file. I know that it has got >to be blocked, waiting for the length to change, but how does it get >notified of a change? Thanks in advance for any responses. Actually, tail doesn't get notified of a change. It simply sleeps for a second, then tries a read(). Probably something like this: while(1) { sleep(1); while(size=read(file,buffer,BSIZE)) write(1,buffer,size); } Nothing fancy, I'm afraid. Jaime ........................................................................... : domain: jds@mimsy.umd.edu James da Silva : path: uunet!mimsy!jds
rick@kimbal.UUCP (Rick Kimball) (12/08/88)
From article <412@fedeva.UUCP>, by jbr0871@fedeva.UUCP (Blaine Robertson): > Can someone please tell me how 'tail -f' works?. Obviously, it does not > do a busy loop while doing a stat on the file. I know that it has got > to be blocked, waiting for the length to change, but how does it get > notified of a change? Thanks in advance for any responses. My guess is that it sleeps for a second and then tries to read more data. -- ____________________________________________________________________________ Rick Kimball | Mac Source BBS, Altamonte Springs, FL DATA (407) 862-6214 | VOICE (407) 788-6875 UUCP: rick@kimbal ..!gatech!fabscal!kimbal!rick ..!ucf-cs!sdgsun!kimbal!rick
guy@auspex.UUCP (Guy Harris) (12/09/88)
>Can someone please tell me how 'tail -f' works?. Obviously, it does not >do a busy loop while doing a stat on the file. I know that it has got >to be blocked, waiting for the length to change, but how does it get >notified of a change? Thanks in advance for any responses. Oh, dear, he used the "O" word. One should be careful using that word.... In fact, it *does* do a busy loop while doing "read"s (not "stat"s) of the file; it just sleeps 1 second between attempts to "read", so it doesn't burn quite so much CPU doing it. If there were some way for it to block waiting for the length to change, it would; however, there isn't any such way.
decot@hpisod2.HP.COM (Dave Decot) (12/10/88)
> >Can someone please tell me how 'tail -f' works?. Obviously, it does not > >do a busy loop while doing a stat on the file. I know that it has got > >to be blocked, waiting for the length to change, but how does it get > >notified of a change? Thanks in advance for any responses. > > Oh, dear, he used the "O" word. One should be careful using that > word.... > > In fact, it *does* do a busy loop while doing "read"s (not "stat"s) of > the file; it just sleeps 1 second between attempts to "read", so it > doesn't burn quite so much CPU doing it. If there were some way for it > to block waiting for the length to change, it would; however, there > isn't any such way. I suppose it could use select(2)... Dave
guy@auspex.UUCP (Guy Harris) (12/11/88)
>I suppose it could use select(2)...
You suppose incorrectly. I know of no UNIX system where "select" will
permit you to wait for a regular file (as in S_IFREG) file to be
modified....mirk@warwick.UUCP (Mike Taylor) (12/12/88)
>>>Can someone please tell me how 'tail -f' works? ... >>In fact, it *does* do a busy loop while doing "read"s (not "stat"s) of >>the file; it just sleeps 1 second between attempts to "read" ... >I suppose it could use select(2)... Not necessarily - select(2) is a Berkeleyism, I believe, so code written using it is not portable to older Unices - Vanilla V7, for instance. I must agree that it would be a far more elegant solution, though. I am surprised that it hasn't been re-written for Berkeley systems. ______________________________________________________________________________ Mike Taylor - {Christ,M{athemat,us}ic}ian ... Email to: mirk@uk.ac.warwick.cs *** Unkle Mirk sez: "Gm F#7 Bb Cm->Gm->Eb->Gm F; Gm F#7 Bb C Eb C Bb Dm D" *** ------------------------------------------------------------------------------
guy@auspex.UUCP (Guy Harris) (12/13/88)
>Not necessarily - select(2) is a Berkeleyism, I believe, so code written >using it is not portable to older Unices - Vanilla V7, for instance. >I must agree that it would be a far more elegant solution, though. >I am surprised that it hasn't been re-written for Berkeley systems. I'm not at all surprised, since even on a Berkeley system you can't use "select()" to wait for somebody to write something to a regular file.
moss@BRL.MIL (VLD/VMB) (12/13/88)
>>I suppose it could use select(2)... > >You suppose incorrectly. I know of no UNIX system where "select" will >permit you to wait for a regular file (as in S_IFREG) file to be >modified.... Right, he was probably referring to a non-blocking read which won't help since blocking is really what you want. Anyway, I haven't looked at the source for 'tail', but I wrote an asynchronous log file monitoring subroutine to allow the contents of an interactive program's log file to be displayed in a scrolling window. A subroutine gets called from an alarm signal handler which loops on fgets() until one fails, then does a clearerr() (only necessary on some systems like 4.2 BSD) to clear the EOF condition. Then, the signal handler resets the alarm for N seconds. Of course you can use sleep() instead of the alarm() call/handler if you want. -moss
dce@mips.COM (David Elliott) (12/14/88)
No, you can't use select, but tail can be made a little smarter.
In Tektronix's UTek, tail -f has a "back-off" for the timing of the
loop. After 10 failed reads, the sleep time is increased by some
amount (with some upper limit, of course). Tim Knox, who was at Intel
last I saw, might remember the details, as he was the person that
wrote the code.
In any event, it made a tail -f that was running for no reason not
take up so much time.
--
David Elliott dce@mips.com or {ames,prls,pyramid,decwrl}!mips!dce
"Did you see his eyes? Did you see his crazy eyes?" -- Iggy (who else?)guy@auspex.UUCP (Guy Harris) (12/15/88)
>Right, he was probably referring to a non-blocking read which won't help >since blocking is really what you want. *Ahem* a non-blocking "read" on a regular file is no different from a blocking "read" on the same file under most versions of UNIX. Folks, there really, truly, honestly *is* no special magic feature of e.g. 4.xBSD that obviates the need for the "sleep and try again" stuff....