diff options
Diffstat (limited to 'Bugs/substleak')
| -rw-r--r-- | Bugs/substleak | 98 |
1 files changed, 98 insertions, 0 deletions
diff --git a/Bugs/substleak b/Bugs/substleak new file mode 100644 index 0000000000..ff14446488 --- /dev/null +++ b/Bugs/substleak @@ -0,0 +1,98 @@ +Return-Path: Martin.Ward@durham.ac.uk +Return-Path: <Martin.Ward@durham.ac.uk> +Received: from scalpel.netlabs.com by netlabs.com (4.1/SMI-4.1) + id AA01931; Thu, 20 Jan 94 03:56:39 PST +Received: from netlabs.com (vaccine-eng1.netlabs.com) by scalpel.netlabs.com (4.1/SMI-4.1) + id AA09639; Thu, 20 Jan 94 03:56:36 PST +Received: from sun2.nsfnet-relay.ac.uk by netlabs.com (4.1/SMI-4.1) + id AA01923; Thu, 20 Jan 94 03:56:01 PST +Via: uk.ac.durham; Thu, 20 Jan 1994 11:47:16 +0000 +Received: from easby.dur.ac.uk by durham.ac.uk; Thu, 20 Jan 94 11:47:05 GMT +Received: from ws-csm3.durham.ac.uk (ws-csm3.dur) by uk.ac.durham.easby; + Thu, 20 Jan 94 11:46:29 GMT +From: Martin.Ward@durham.ac.uk (Martin Ward) +Date: Thu, 20 Jan 94 11:46:27 GMT +Message-Id: <AA00871.9401201146.ws-csm3@uk.ac.durham> +To: des0mpw@easby.durham.ac.uk, lwall@scalpel.netlabs.com +Subject: Re: My last message + +>: After saying I was stumped, I managed to track down the problem! +>: The problem was caused by a line much higher up: +>: +>: $seqpat = "$bs[s]\000e\000q\000\{\000"; # } hack +>: +>: Changing this by adding {} gives: +>: +>: $seqpat = "${bs}[s]\000e\000q\000\{\000"; # } hack +>: +>: which worked! No idea why :-) +> +>It apparently intuited $bs[s] to be an array reference. + +Aha! I think the interpretation is: +"$bs[ s] ..... +^^^^^ ^^ +array ref Therefore this is an expression, so "s]" is the start of + a pattern match/replace, so it scans for ...]....] + +I found the line by repeatedly chopping away everything after (and including) +the line where perl _thought_ the error started. This gradually worked back +through a nasty cascade of errors! + +>: No speed improvement this time (the improvement in user time was swamped +>: by an increase in system time. This may be because it uses 5048k of +>: data/stack space, compared with 985k for perl4). +> +>That doesn't sound good. I hope it's a bug. Does it grow continuously? +>Hopefully it's just a memory leak. + +For perl4 the memory size (shown by top) grows by about 100-150k per 2 seconds, +for perl5 it grows by about 1 - 1.5 Meg per 2 seconds. I don't use the script +very often (and I have over 100 Meg of swap space) so its not a big problem. +Still, with a 124k input file, the size for perl5 went up to 40 Meg! + +I have tracked down a memory leak, which is basically one line from the +texqed script. Store this script in "tmp" and run "perl tmp /vmunix" +(or some other large random file). Monitor the process using "top" in +another window. + +Perl4 is OK but perl5 leaks like a leaky cistern. + + +#!/usr/local/bin/perl + +# print a "." every $interval lines: +$interval = 10; +open (PAIRS, "$ARGV[0]"); +open (OUT, "> /dev/null"); +$bs = "\\\\" . "\000"; +for (;;) { + $line++; + if (($line % $interval) == 0) { + print STDERR "."; + } + read(PAIRS, $_, 20); + last if ($_ eq ""); + + s/$bs([_^\\])\000/\\\377$1\377/g; + + print OUT ; +} + +print STDERR "\n"; + + + +I hope you find this useful! + +NB Changing the "s/.../.../g" to an "m/.../" (with the same pattern) +makes the leak go away even on input files where the pattern NEVER matches!! + + Martin. + +JANET: Martin.Ward@uk.ac.durham Internet (eg US): Martin.Ward@durham.ac.uk +or if that fails: Martin.Ward%uk.ac.durham@nsfnet-relay.ac.uk +or even: Martin.Ward%DURHAM.AC.UK@CUNYVM.CUNY.EDU +BITNET: Martin.Ward%durham.ac.uk@UKACRL UUCP:...!uknet!durham!Martin.Ward +[Last acked 0.7 days ago--not acked] + |