From: utzoo!decvax!harpo!seismo!hao!hplabs!sri-unix!gwyn@Brl
Newsgroups: net.unix-wizards
Title: Re:  compaction using realloc
Article-I.D.: sri-unix.4941
Posted: Tue Dec 28 02:23:06 1982
Received: Wed Dec 29 04:32:17 1982

From:     Doug Gwyn <gwyn@Brl>
Date:     18 Dec 82 17:00:58-EST (Sat)
Some years ago I implemented the first-fit allocation method using a
"roving" search-start pointer as you mentioned, along with combining
adjacent freed regions (identifiable because of "boundary tags")
during allocation searching.  The application was display processor
real-time segment management, and it worked well.  One drawback to
the "boundary-tag" approach is that it assumes there is no memory
in the pool other than that allocated through this package.  This is
NOT the case with the "malloc" scheme, whereby a user can increase
the program break directly without breaking malloc's operation.  I
have tried to think of ways of permitting fake "busy block" boundary
tags for uncontrolled heap space, but I couldn't think of anything
reasonable to do about it.  So malloc is still a sensible choice.