garyb@abekrd.UUCP (Gary Bartlett) (01/30/91)
Can someone explain to me what is happening with the following Bourne shell script and more importantly how I can get around it: #!/bin/sh cat file | ( head -200 echo "Line 201 follows" head -200 echo "Line 401 follows" cat ) I am trying to use 'head' as a quick way to split up an input stream. I originally used 'read' and an 'expr' counter but this was too slow. This script loses lines after each 'head'. eg if file contained a stream of numbers, the output would be missing lots of numbers! It looks like 'head' initially reads in a whole buffer of data from file (stdin), prints out the requisite number of lines and then dumps the rest of the buffer. The next 'head' then reads the NEXT buffer. Is this right/normal? How can I get around this (preferably whilst still using 'head'). Is it possible to change the buffering within this script? Pointers, anyone? Thanks, Gary -- --------------------------------------------------------------------------- Gary C. Bartlett NET: garyb@abekrd.co.uk Abekas Video Systems Ltd. UUCP: ...!uunet!mcsun!ukc!pyrltd!abekrd!garyb 12 Portman Rd, Reading, PHONE: +44 734 585421 Berkshire. RG3 1EA. FAX: +44 734 567904 United Kingdom. TELEX: 847579
mcgrew@ichthous.Eng.Sun.COM (Darin McGrew) (01/30/91)
In article <1671@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes:
->Can someone explain to me what is happening with the following Bourne shell
->script and more importantly how I can get around it:
->
-> #!/bin/sh
-> cat file | (
-> head -200
-> echo "Line 201 follows"
-> head -200
-> echo "Line 401 follows"
-> cat
-> )
->
->...
->It looks like 'head' initially reads in a whole buffer of data from file
->(stdin), prints out the requisite number of lines and then dumps the rest
->of the buffer. The next 'head' then reads the NEXT buffer....
Yes, head reads a bufferful at a time. I'd use awk:
awk ' NR==201 {print "Line 201 follows"}
NR==401 {print "Line 401 follows"}
{print}' < file
->Thanks,
->Gary
You're welcome.
Darin McGrew "The Beginning will make all things new,
mcgrew@Eng.Sun.COM New life belongs to Him.
Affiliation stated for He hands us each new moment saying,
identification purposes only. 'My child, begin again....
You're free to start again.'"
]) (01/31/91)
In article <6925@exodus.Eng.Sun.COM> mcgrew@ichthous.Eng.Sun.COM (Darin McGrew) writes: >In article <1671@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes: >->Can someone explain to me what is happening with the following Bourne shell >->script and more importantly how I can get around it: >-> >-> #!/bin/sh >-> cat file | ( >-> head -200 >-> echo "Line 201 follows" >-> head -200 >-> echo "Line 401 follows" >-> cat >-> ) >-> >->... >->It looks like 'head' initially reads in a whole buffer of data from file >->(stdin), prints out the requisite number of lines and then dumps the rest >->of the buffer. The next 'head' then reads the NEXT buffer.... > >Yes, head reads a bufferful at a time. I'd use awk: > > awk ' NR==201 {print "Line 201 follows"} > NR==401 {print "Line 401 follows"} > {print}' < file And this might be faster (note, though, that I'll need to left-justify this to avoid inserting leading white-space). -- start fragment -- sed \ -e '200a\ LIne 201 follows' \ -e '400a\ Line 401 follows' < file -- end fragment -- I'm making no statement here that the sed call is better than the awk call, just that if performance is significant, you might want to try this approach too and compare execution times. If, however, the echo "Line ?01 follows" in the original example was a place holder for "I want to do other stuff here, then pick up processing with the next set of lines", neither the awk nor the sed calls will allow it, as both simply insert the line-counting messages into the stream of data from file. Dog slow though it be, the following will do it: #!/bin/sh ( i=1 while [ $i -lt 201 ] do read line; echo "$line" i=`expr $i + 1` done : process some stuff here while [ $i -lt 401 ] do read line; echo "$line" i=`expr $i + 1` done : process some more stuff here cat - ) < file It's only slightly better in ksh, by replacng the i=1 assignment with typeset -i i=1 and replacing the expr call to increment $i with ((i += 1)) . In either case, mayhem will result if file isn't at least 400 lines long. You may be forced into multiple reads of the file to get something resembling good performance: #!/bin/sh ( sed 200q file echo "Line 201 follows" sed -e '1,200d' -e '400q' file echo "Line 401 follows" sed '1,400d' file ) The saving graces here are that, even though the file is opened three times, (1) only the first 200 lines are read thrice and the second 200 twice, and (2) one avoids the nearly nightmarish performance of the while loops in the example preceeding this one. It doesn't hurt, too, that sed is pretty quick. Now, let's take it one step further and generalize it into a function... #!/bin/sh # # A function to get $2 lines from file $1 starting at $3 # Only the file ($1) is required # getlines() { file=$1 count=$2 start=${3:-1} # default start at line 1 if [ ! -r "$file" ] then echo "getlines: file '$1' not readable" 1>&2 return 1 fi # Whole file? if [ $start -eq 1 -a "$count" = "" ] then cat $file return $? fi # From start to EOF? if [ "$count" = "" ] then sed -n "$start,\$p" $file return $? fi # Start at line 1 for count lines? if [ $start -eq 1 ] then sed "${count}q" $file return $? fi # We have a start other than 1 and a count cut=`expr $start - 1` # Don't print through $cut end=`expr $cut + $count` # $end is last to print if [ $end -le $cut ] then echo "getlines: bad count($count)/start($start)" 1>&2 return 1 fi sed -e "1,${cut}d" -e "${end}q" $file return $? } # # Mainline code # file=${1:-file} # If there's an arg, it's the filename wc=`wc -l < $file` count=200 current=1 while [ $current -le $wc ] do if [ $current -ne 1 ] then echo "Next line is $current" fi if getlines $file $current $count then current=`expr $current + $count` else saverc=$? echo "$0: getlines returned $saverc" 1>&2 exit $saverc fi done All that's left is to have flags for count and maybe the initial "current". >->Thanks, >->Gary > >You're welcome. I'll second that! ...Kris -- Kristopher Stephens, | (408-746-6047) | krs@uts.amdahl.com | KC6DFS Amdahl Corporation | | | [The opinions expressed above are mine, solely, and do not ] [necessarily reflect the opinions or policies of Amdahl Corp. ]
tchrist@convex.COM (Tom Christiansen) (01/31/91)
From the keyboard of krs@amdahl.uts.amdahl.com (Kris Stephens [Hail Eris!]): :In article <6925@exodus.Eng.Sun.COM> mcgrew@ichthous.Eng.Sun.COM (Darin McGrew) writes: :>In article <1671@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes: :>->Can someone explain to me what is happening with the following Bourne shell :>->script and more importantly how I can get around it: :>-> :>-> #!/bin/sh :>-> cat file | ( :>-> head -200 :>-> echo "Line 201 follows" :>-> head -200 :>-> echo "Line 401 follows" :>-> cat :>-> ) :>-> :>->... :If, however, the echo "Line ?01 follows" in the original example :was a place holder for "I want to do other stuff here, then pick up :processing with the next set of lines", neither the awk nor the sed :calls will allow it, as both simply insert the line-counting messages :into the stream of data from file. : :Dog slow though it be, the following will do it: I really wasn't going to do this, but once I saw things like "dog slow" and "shell functions" (which many of us don't have) and "mayhem will result", I just couldn't not give a perl solution. while (<>) { # fetch line into pattern space if ($. < 201) { s/foo/bar/g; # do some other stuff } elsif ($. < 401) { tr/a-z/A-Z/; # do other stuff } else { print "line number is $., continuing...\n"; # do final stuff } } Basically, perl keeps track of your current input line number in the $. variable (mnemonic: think of dot in editors) just as awk does with NR. The advantage to using perl is that you can do much more without having to call other program, and instead of asking yourself a load of questions like "does their awk have functions?", "does their sh have functions?", "did I exceed awk/sh's field limits?", you only have the one question of whether perl is on the system, and unlike nawk and ksh (although like gawk and bash), you can put it on your system without shelling out money if the answer should be no. You'll also find that perl will be faster than the sed/sh/awk combo, and often faster than even just one of them. Please, save the flames for alt.religion.computers. I'm just trying to present another possibility of which the original poster may not have been aware. --tom -- "Hey, did you hear Stallman has replaced /vmunix with /vmunix.el? Now he can finally have the whole O/S built-in to his editor like he always wanted!" --me (Tom Christiansen <tchrist@convex.com>)
garyb@abekrd.UUCP (Gary Bartlett) (01/31/91)
In <fcPl016n13RO00@amdahl.uts.amdahl.com> krs@uts.amdahl.com (Kris Stephens [Hail Eris!]) writes: >In article <6925@exodus.Eng.Sun.COM> mcgrew@ichthous.Eng.Sun.COM (Darin McGrew) writes: >>In article <1671@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes: >>->... >>->It looks like 'head' initially reads in a whole buffer of data from file >>->(stdin), prints out the requisite number of lines and then dumps the rest >>->of the buffer. The next 'head' then reads the NEXT buffer.... >> >If, however, the echo "Line ?01 follows" in the original example >was a place holder for "I want to do other stuff here, then pick up >processing with the next set of lines", neither the awk nor the sed >calls will allow it, as both simply insert the line-counting messages >into the stream of data from file. This is indeed what I intended - see my last piece of news on the subject. >Dog slow though it be, the following will do it: > #!/bin/sh > ( > i=1 > while [ $i -lt 201 ] > do > read line; echo "$line" > i=`expr $i + 1` > done > : process some more stuff here > cat - > ) < file This is effectively what I started out using - a 'while' loop, an 'expr' counter, and a couple of 'read's. Hideously slow! >You may be forced into multiple reads of the file to get something >resembling good performance: >The saving graces here are that, even though the file is opened three >times, (1) only the first 200 lines are read thrice and the second >200 twice, and (2) one avoids the nearly nightmarish performance of >the while loops in the example preceeding this one. It doesn't hurt, >too, that sed is pretty quick. The thing is, the file I'm merging from may be very long (ie very many sed passes). >Now, let's take it one step further and generalize it into a function... I DO like the function idea though. I did actually write my own 'head' (C) program which turned off all buffering of the stdin before doing any reading. This did the trick and worked in the shell script. It was faster but not greatly so - I guess it had to read every character individually. I did try using line-buffering but this did not work. It still lost data (although not as much as when using the full-buffering of head). I'm not overly happy with that solution though - I but it's not at all portable. *** FLASH OF INSPIRATION *** I have an idea: - Process the original file by putting the line number at the beginning of each line, - Process the file to be merged so that the merge points are at the beginning of each of these lines, - Cat the two processed files together and pass through 'sort', - Remove line numbers from beginning of resulting file, QED This doesn't matter how big either file is. Thoughts? Thanks again for some very useful input, Gary -- --------------------------------------------------------------------------- Gary C. Bartlett NET: garyb@abekrd.co.uk Abekas Video Systems Ltd. UUCP: ...!uunet!mcsun!ukc!pyrltd!abekrd!garyb 12 Portman Rd, Reading, PHONE: +44 734 585421 Berkshire. RG3 1EA. FAX: +44 734 567904 United Kingdom. TELEX: 847579
tchrist@convex.COM (Tom Christiansen) (01/31/91)
From the keyboard of garyb@abekrd.UUCP (Gary Bartlett): [various approaches including special versions of head and other nasties deleted, many complaining of being dog slow.] :Thoughts? I think this all a lot of unnecessary complexity; it's not that I think it impossible using tools from V6 UNIX, but it's more cumbersome and slow than it really needs to be. Consequently, I offer to translate Gary's sed/sh/head/sort/blah script to perl for him if he wants. I promise it'll be faster, smaller, simpler, and a lot more portable than writing a new version of head to do the job. Sorry, one per customer please. :-) --tom -- "Hey, did you hear Stallman has replaced /vmunix with /vmunix.el? Now he can finally have the whole O/S built-in to his editor like he always wanted!" --me (Tom Christiansen <tchrist@convex.com>)
martin@mwtech.UUCP (Martin Weitzel) (02/01/91)
In article <1678@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes:
[...]
:*** FLASH OF INSPIRATION ***
:
:I have an idea:
:- Process the original file by putting the line number at the beginning of
: each line,
Good.
:- Process the file to be merged so that the merge points are at the beginning
: of each of these lines,
Good.
:- Cat the two processed files together and pass through 'sort',
Depending on how the lines look like it may also be possible to use
join(1).
:- Remove line numbers from beginning of resulting file, QED
and split the joined lines.
--
Martin Weitzel, email: martin@mwtech.UUCP, voice: 49-(0)6151-6 56 83
]) (02/02/91)
In article <1678@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes: >*** FLASH OF INSPIRATION *** > >I have an idea: >- Process the original file by putting the line number at the beginning of > each line, >- Process the file to be merged so that the merge points are at the beginning > of each of these lines, >- Cat the two processed files together and pass through 'sort', >- Remove line numbers from beginning of resulting file, QED > >This doesn't matter how big either file is. > >Thoughts? Hmmm.... If you've got the blocks available on your disk to handle a second copy of the file, read it once and break it into (n) pieces of your required length, each. This would use pretty much the same number of blocks of storage (plus at most a block-per-chunk for fractional last blocks) and some number of inodes (number of lines / lines per chunk). --- multi-piece processing -- : file=${1:-file} lines=${2:-200} base=Base # awk will return the number of subsets it created subsets=`awk ' NR == 1 { i++ outfile = base "" i print > outfile next } NR % lines == 1 { close outfile i++ outfile = base "" i print > outfile next } { print >> outfile } END { close outfile; print i + 0 }' base=$base lines=$lines $file` i=1 while [ $i -le $subsets ] do subset="$base$i" cat $subset rm $subset # Clean up now echo "# End of block $i" i=`expr $i + 1` done > report # or pipe it to something else (lp?) --- multi-piece processing --- The perl-literate will undoubtedly say "If you're going to do this, use perl instead", but so it goes... :-) Anyhow, this reads the data twice (once to break it into pieces of $lines; once to cat them into the report file). >Thanks again for some very useful input, >Gary You're very welcome -- helps me stretch out my thinking, too. ...Kris -- Kristopher Stephens, | (408-746-6047) | krs@uts.amdahl.com | KC6DFS Amdahl Corporation | | | [The opinions expressed above are mine, solely, and do not ] [necessarily reflect the opinions or policies of Amdahl Corp. ]
rbp@investor.pgh.pa.us (Bob Peirce #305) (02/04/91)
In article <1671@abekrd.UUCP> garyb@abekrd.UUCP (Gary Bartlett) writes: >Can someone explain to me what is happening with the following Bourne shell >script and more importantly how I can get around it: > > cat file | ( > head -200 > echo "Line 201 follows" > head -200 > echo "Line 401 follows" > cat > ) > >I am trying to use 'head' as a quick way to split up an input stream. I >originally used 'read' and an 'expr' counter but this was too slow. > >This script loses lines after each 'head'. eg if file contained a stream of >numbers, the output would be missing lots of numbers! > Why not use cat file | split -200 commands to process x?? files -- Bob Peirce, Pittsburgh, PA 412-471-5320 ...!uunet!pitt!investor!rbp rbp@investor.pgh.pa.us