From /G=Owen/S=Smith/O=SJ-Research/ADMD=INTERSPAN/C=GB/@mhs-relay.ac.uk Wed Feb 17 12:44:26 1993
Newsgroups: comp.sys.acorn.tech
From: /G=Owen/S=Smith/O=SJ-Research/ADMD=INTERSPAN/C=GB/@mhs-relay.ac.uk
Subject: RMTidy (was Monty Python)
Organization: Yale CS Mail/News Gateway
Date: Tue, 16 Feb 1993 14:33:35 GMT

tytgat@esat.kuleuven.ac.be (John Tytgat wrote:

>But this also means that if an *application* claimed an RMA block, this
>block will probably being overwritten during an RMTidy command (in the case
>RMTidy would work in Desktop).

Are you sure of this? I thought that RMTidy left allocated blocks alone.
Indeed if it doesn't then some of my code is in serious trouble. My
understanding is as follows:

Each module in the module chain is issued a non-fatal finalisation (this is
in fact the ONLY time modules get a non-fatal finalisation). Modules can
free various bits of RMA up if they want, and indeed this is desirable. If
they free their main workspace, then they should set their private word to
zero rather than leaving it pointing at the freed block (a common bug).

Compaction of the RMA occurs. Two sorts of blocks are regarded as movable:
modules, and RMA blocks pointed to by the private words of modules. Any
other allocated RMA blocks are not movable and are left where they are.

Each module in the module chain is initialised. If they left their workspace
allocated, then their private word will now point at the new location of the
workspace. A frequent bug at this point is to ignore this and allocate some
more workspace. Another bug is to assume that the workspace is in the same
place it used to be. Thus if you have pointers in your workspace that point
to other bits of your workspace, they must now be relocated. Hence using
offsets is much easier. Also the fact that you get your workspace contents
back means you can preserve state across RMTidy, plus of course you could
leave some state in extra RMA blocks but then the compaction of the RMA
won't be as good.

Some frequent bugs/misunderstandings come out of all this:

 1) Non-fatal finalisation of modules is ONLY for RMTidy. RMReinit does a
    fatal finalisation.

 2) You must take to care avoid losing RMA blocks across RMTidy. You should
    either a) free your workspace yourself and set your private word to
    0 in your finalisation, or b) check your private word in initialisation
    and if it's contents are none zero don't allocate more workspace.

Owen.