[Info-vax] RMS record metadata, was: Re: Re; Spiralog, RMS Journaling (was

[lawrencedo99 at gmail.com]

On Wednesday, June 22, 2016 at 7:04:22 AM UTC+12, David Froble wrote:
> Lawrence D’Oliveiro wrote:
>> * I/O buffering, because the underlying ACP/XQP layer only allows reading
>> and writing whole multiples of 512 bytes (how dumb is that?).
> 
> You might THINK you're doing something else, but when the data hits the rust, 
> it's in chunks defined by the disk, usually in multiples of 512 bytes.

Newer disks have 4kiB physical sector sizes. And then there’s shingle magnetic recording, where writing anything less than a whole track becomes an involved process (remember track buffers on floppy disks?). How well does your 512-byte-multiple restriction fit into this world? It starts to look pretty anachronistic, doesn’t it?

> Something has got to do filespec parsing.  Better done once than custom code
> in every application.

Then there’s the sheer complexity of a VMS filespec. Does the idea of exposing disk device names seem so wise nowadays? On POSIX, you have a pathspec with components separated by simple slashes, and disk device names are completely hidden.