Service Manuals, User Guides, Schematic Diagrams or docs for : xerox sdd memos_1977 19771107_Adding_New_Devices_To_The_Channel_Interface
<< Back | HomeMost service manuals and schematics are PDF files, so You will need Adobre Acrobat Reader to view : Acrobat Download Some of the files are DjVu format. Readers and resources available here : DjVu Resources
For the compressed files, most common are zip and rar. Please, extract files with Your favorite compression software ( WinZip, WinRAR ... ) before viewing. If a document has multiple parts, You should download all, before extracting.
Good luck. Repair on Your own risk. Make sure You know what You are doing.
Image preview - the first page of the document
>> Download 19771107_Adding_New_Devices_To_The_Channel_Interface documenatation <<
Text preview - extract from the document
Inter-Office Memorandum
DRAFT - DRAFT - DRAFT - DRAFT
To Pilot interest Date Novem ber 7, 1977
From Paul McJones Location Palo Alto
Subject Adding new devices to the Channel interface
Organization SO~/SO
XEROX XEROX SDD ARCR1VES _
I have .roaCt and undo1"::;tood.
Filed on: [lfs]Channel.memo rag es __,_.---'fo - , - - - - -
'f'I
_ _ __
1h~e
Reviewer_-----
# of Pages _ __ Ed. -t1SDQ-.:2 feJ4
The problem
The Channel interface in the second draft of the Pilot Functional Specification has the
property that adding a new device requires editing and recompiling the Channel DEFINITIONS
module, and thus all the modules which include it. This is in conflict with the desire not to
recompile the system except for the semiannual releases (assuming new devices are to be
added between releases). Thus we propose a revised Channel interface, which, it is claimed,
solves the recompilation problem.
The problem with the current interface stems from its dependence on the enumerated type
Channel.DeviceType, which has a value for each implementation of the interface. Adding a
new implementation requires adding a new value to DeviceType, and also adding
corresponding variants to the definitions of the DeviceType-tagged variant record types
DeviceSelection, PhysicalRecordStatus, DeviceCommand. DeviceStatus, and "Rep" (the
private record to which a Handle points). (It is worth noting that these variant records
constitute a nuisance for both client and implementor. Most values crossing the interface
must be wrapped in a variant record constructor, and then "unwrapped" with a one-armed
WITH statement essentially doing at runtime what Mesa is meant to do at compile time.)
A solution
Half of the operations in the current interface do not depend on DeviceType. and so are
not a problem:
Delete
Suspend
Restart
GetPhysicalRecord
PutPhysicalRecord
Those which do depend on DeviceType are:
Create (accepts DeviceSelection)
WaitPhysicalRecord (retu rns PhysicalRecordStatus)
CommandDevice (accepts DeviceCommand)
GetDeviceStatus (returns DeviceStatus)
WaitDeviceStatusChange (accepts and returns DeviceStatus)
DRAFT - DRAFT - DRAFT - DRAFT
Adding new devices to the Channel interface 2
A simple way to remove this dependency is to remove these operations from the Channel
interface, and to add device-specific version of each of them into each specific device
interface. Taking the floppy disk for example we would have:
FloppyDrive.Create: PROCEDURE[Spindle: CARDINAL ~ FloppyDrive.defaultSpindle]
RETURNs[ChanneI.Handle];
FloppyDrive.defaultSpindle: CARDINAL = LAST[CARDINAL];
FloppyDrive.WaitPhysicaIRecord: PROCEDURE[Channel.Handle, Channel.lOEvent]
RETURNS[ FloppyDrive.CompletionCode];
FloppyDrive.CompletionCode: TYPE = RECORD[ ... ]; --as currently
FloppyDrive.CommandDevice: PROCEDURE[ Channel.Handle, FloppyDrive .Command]
RETURNS[ Channel.Handle];
FloppyDrive.Command: TYPE = RECORD[ ... ]; --as currently
FloppyDrive.GetDeviceStatus: PROCEDURE[ Channel.Handle ]
RETURNS[ FloppyDrive.Status];
FloppyDrive.Status: TYPE = MACHINE DEPENDENT RECORD[ ... ]; --as currently
FloppyDrive. WaitDev iceStatusChange: PROCEDURE[ Channel.Handle, FloppyDrive.Status]
RETURNS[ FloppyDrive.Status];
Discussion
It is interesting to note that in all cases, each variant of the records specified in the Channel
interface consist of a single field whose type had already been defined in a separate device
interface; the above device-dependent operations just pass that object directly. The
argument that forcing all device operations to go through the Channel interface will assure
uniformity from the diverse community of device interfacers seems weak in this light;
providing models in the form of actual device interfaces produced here at SD/Palo Alto
would seem to go a long way towards achieving useful consistency.
Implementation comments
Both the current Channel interface and the above variation presuppose a "streamlike"
implementation in which every Channel.Handle points to a record which has in standard
positions implementation-specific procedure descriptors for the generic operations Get, Put,
etc. (Since Create doesn't take a Channel.Handle as parameter in the current interface. it
would have to dispatch to an appropriate routine using a global table.) In the variation
proposed above, the "DeviceType dependent" operations would be IMPORTed directly by the
client from a device interface. To maintain type-checking, each of these operations
accepting a Channel.Handle parameter should logically check the "device type" of the channel.
This could be done with a numeric type field and type codes assigned by the Pilot
maintainers, but a simpler way is available. The procedure descriptors already in the record
pointed to by the Handle constitute a type code! Thus the floppy disk CommandDevice
routine might begin with something like
IF channeI.GetPhysicaIRecord-=FloppyDrive.GetPhysicaIRecord THEN ERROR ...
◦ Jabse Service Manual Search 2024 ◦ Jabse Pravopis ◦ onTap.bg ◦ Other service manual resources online : Fixya ◦ eServiceinfo