topic Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump) in Operating System - HP-UX

sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

Wayne Green — Wed, 15 May 2002 09:34:51 GMT

Trying to sed a file with long lines, over 2000 chars a line. Get a core dump trying to get rid of the "\" char.

Any other char is OK - sed 's/\|//g' or sed 's/4//g' or chars enclosing the "\" chars - sed 's/4*05775//g'.

Also get a core dump on occasion searching file for "\" char in more. Try to vi the file but get the line too long message. od -c doesn't show any dodgy chars.

Created a dummy file with "\" chars and sed can find the "\" file and remove or replace them.

Anyone with any ideas

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

S.K. Chan — Wed, 15 May 2002 09:41:03 GMT

It more of a limit thing. I thing if you got more than 8k in one line it'll core dump on you. Try GNU sed .. it's more robust ..

http://hpux.cs.utah.edu/hppd/hpux/Gnu/sed-3.02/

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

John Strang — Wed, 15 May 2002 09:41:57 GMT

Hi Wayne,

Would it be possible for you to attach a couple of lines of the file as an example? Is it possible that the \ is showing as part of a newline character, \n or a tab \t ?

John

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

Wodisch — Wed, 15 May 2002 09:50:06 GMT

Hello,

why don't you simply use "tr" for that:

tr -d "\\" < input.file > output.file

this would delete all backslashes from the input.file - without any problem about line-lenght limits...

Just my $0.02,
Wodisch

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

Wayne Green — Wed, 15 May 2002 10:07:51 GMT

I've localised the problem to a few different lines, and put this into a separate file to work on. Its wc output is 1 131 2666 /tmp/wtg1.
The previous 2 lines are OK wc output is 2 247 5306 /tmp/wtg2 so it doesn't seem to be a size limit.

od -c output of the lines with \ in below, dunno if it'll be readable though

0004640 | 1 | p r z e l - 1 1 1 0 0 0 4
0004660 5 2 8 5 4 \ 0 1 \ 0 5 7 7 5
0004700 |
0004720 3 3 8 1 5 | 0 |
0004740 0 | 0 | 0 | 0 |
0004760 | 1 | 5 1 2 6
0005000 | 1 | 0 | 0 | 5 2 4 0 3 0
0005020 3 3 7 5 | R 0 1 1 5 6
0005040 6 0 1 0 | 1 0 1 0 1 0
0005060 2 3 | 2 6 - 1 3 9 - | | N | 0
0005100 0 2 7 3 9 6 2 0 4 5 5 | 1 1 4 0
0005120 1 0 1 0 0 0 2 7 3 9 6 2 0 4 5 5
0005140 2 5 2 1 0 0 | 6 8 \n
0005152

wodisch tried your suggestion, no core dump but it didn't find the "\" char. No change to output.

Thanks for your interest.

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

H.Merijn Brand (procura — Wed, 15 May 2002 10:15:23 GMT

Unless you want to translate \010 to the corresponding value, and just realy just want to drop the \, Wodish' tr should work, as would

# perl -pe 'tr/\\//d' infile >outfile

if you want to translate octals, you could use

# perl -pe 's/\\([0-7]+)/oct$1/ge' infile >outfile

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

Wayne Green — Wed, 15 May 2002 11:12:13 GMT

Ran this by HP aswell and they suggested loading a patch to sed

Symptoms:
PHCO_22760:
1. sed(1) dumps core when the length of input string is
larger than 2K bytes.

once loaded it fixed the problem though dont quite see the logic here when it worked ok for a larger line. Ran other tests with files I created over 2666 chars long on other system built from same ignite tape and sed doesn't produce a core dump. Beats me so I'll just load the path on 'em all.

Re: sed 's/\\//g' < /tmp/wtg1 gives Memory fault(coredump)

Wayne Green — Wed, 15 May 2002 11:15:35 GMT