[Cyan]<Cedar6.0>Tioga>FileOps.mesa!3

FileOps.mesa

written by Bill Paxton December 1980

last written by Paxton; 4-Mar-82 8:59:40

Doug Wyatt, March 2, 1985 3:50:09 pm PST

Michael Plass, March 14, 1985 11:56:00 am PST

FileOps: CEDAR DEFINITIONS

= BEGIN

Op: TYPE = CHAR;

endOfFile: Op = 0C;

startNode: Op = endOfFile+1;

length of formatName in next byte. text for format name follows that.

enter in formatName table

and assign it the next number

startNodeFirst: Op = startNode+1;

startNodeLast: Op = startNodeFirst+numFormats;

these opcodes encode previously seen formatName so don't need to repeat it

find formatName in op-startNodeFirst of format table

other information follows same as for startNode

terminalTextNode: Op = startNodeLast+1;

terminalTextNodeFirst: Op = terminalTextNode+1;

terminalTextNodeLast: Op = terminalTextNodeFirst+numFormats;

these opcodes are for nodes without children so can skip endNode opcode

identical to startNodeFirst..startNodeLast, except implies no children

find format name in op-startNodeFirst of format table

other information follows same as for startNode

otherNode: Op = terminalTextNodeLast+1;

for "other" format of nodes

length of formatName in next byte. text for format name follows that.

enter in formatName table

and assign it the next number

otherNodeShort: Op = otherNode+1;

like otherNode, but followed by a formatName code number instead of length+text

otherNodeSpecs: Op = otherNodeShort+1;

gives "variety" and specifications for immediately previous "other" format node

length of variety name follows in next byte(s). then text of name

enter in property name table and assign next number

length of specs text in next byte(s)

specs follow that

otherNodeSpecsShort: Op = otherNodeSpecs+1;

like otherNodeSpecs, but followed by a propname code number instead of length+text

prop: Op = otherNodeSpecsShort+1;

specifies property for current node

property specs are stored on file as a rope

length of property name follows in next byte(s). then text of name

enter in property name table and assign next number

length of property specs rope in next byte(s)

followed by text for property specs rope

propShort: Op = prop+1;

like prop, but followed by a propname code number instead of length+text

endNode: Op = propShort+1;

end current node and go back to adding to its parent.

rope: Op = endNode+1;

this op declares rope for most recently started node

length of text for the node in next byte(s).

actual text comes from text block

followed by a CR which is not included in the length

comment: Op = rope+1;

identical to rope except implies text stored in comment area of file

runs: Op = comment+1;

this op preceeds definition of looks for most recently started node

number of runs in following byte(s)

have at most 1 runs op per node. if omit, then no special looks.

looks: Op = runs+1;

looks vector in following 4 bytes.

enter vector in looks table.

and assign it the next number

length of run in next byte(s).

looksFirst: Op = looks+1;

looksLast: Op = looksFirst+numLooks;

these ops encode previously encountered looks so don't need to repeat

find looks in looks table[op-looksFirst].

length of run in the next byte(s).

look1: Op = looksLast+1;

like looks op, except has single look char instead of 4 byte looks vector

look2: Op = look1+1;

like looks1, except has two looks chars

look3: Op = look2+1;

like look2, except has three looks chars

numFormats: CARDINAL = 70;

max number of entries in the format table for the file

others must be given in long form

FormatIndex: TYPE = [0..numFormats);

numLooks: CARDINAL = 50;

max number of entries in a look table

others must be given in long form

each entry in a look table is a distinct vector of 32 bits

LooksIndex: TYPE = [0..numLooks);

numProps: CARDINAL = 50;

max number of entries in a prop table

PropIndex: TYPE = [0..numProps);

lengths are given in as few bytes as possible according to following scheme

first bit of each byte tells if other bytes follow (up to 4 bytes max)

concat other 7 bits of bytes with leading 0's to form length

thus can represent lengths up to 2**28-1 if use 4 bytes

lengths of 127 or less fit in 1 byte

lengths in [128..4095] fit in 2 bytes

IntBytes: TYPE = MACHINE DEPENDENT RECORD [

first (0:9..15): [0..127] ← 0, -- first byte written on file

second (0:2..8): [0..127] ← 0, -- second byte written

thirdBottom (0:0..1): [0..3] ← 0, -- bottom bits of third byte written

thirdTop (1:11..15): [0..31] ← 0, -- top bits of third byte written

fourth (1:4..10): [0..127] ← 0, -- fourth byte written

unused (1:0..3): [0..15] ← 0 -- these are always 0

];

LengthByte: TYPE = MACHINE DEPENDENT RECORD [

unused (0:0..7): [0..255] ← 0,

others (0:8..8): BOOL ← FALSE, -- tells if other bytes are present

data (0:9..15): [0..127] ← 0

];

ThirdByte: TYPE = MACHINE DEPENDENT RECORD [

unused (0:0..7): [0..255] ← 0,

others (0:8..8): BOOL ← FALSE,

dataTop (0:9..13): [0..31] ← 0, -- top 5 bits of data

dataBottom (0:14..15): [0..3] ← 0 -- bottom 2 bits of data

];

The following defines the trailer bytes that form the Tioga password

commentHeaderId: ARRAY [0..fileIdSize) OF CHAR = [0C,0C];

controlHeaderId: ARRAY [0..fileIdSize) OF CHAR = [235C,312C];

controlTrailerId: ARRAY [0..fileIdSize) OF CHAR = [205C,227C];

fileIdSize: NAT = 2;

numTrailerLengths: NAT = 3; -- <file-props-length> <data-length> <file-length>

endSize: NAT = fileIdSize+numTrailerLengths*4; -- trailer plus three lengths

END.

Tioga file format is as follows:

<data> == characters from (non-comment) nodes in display order with a CR after each node

<comments-header> == <commentHeaderId> <comments-length>

<comments-length> == byte length of <comments>

<comments-text> == characters from comment nodes in display order with a CR after each node

<control-header> == <controlHeaderId> <control-length>

<commentHeaderId> == <controlHeaderId> == [..bytes..] as defined above

Thus <comments> start with double 0's which can serve to mark end of <data>.

<control-length> == byte length of <control>

<control-trailer> == <file-props> <controlTrailerId>

<file-props-length> <data-length> <file-length>

<file-props-length> == gives length of <file-props>

<file-prop> == <propIdByte> <valueLengthByte> <propValueBytes>

<controlTrailerId> == [..bytes..] as defined above

<comments-start> == byte address in file for start of comments

<control-info> == sequence of control bytes and control text.

The last thing in the file is the <control-trailer>. Thus to read a file, first

check the <controlTrailerId> and read the byte address for the <data-length>.

We also read the saved <file-length> as a double check.

The saved <data-stamp> is a 4 byte checksum of the data part of the file.

This lets us read the <comments-header> where we double check to make sure we

have a valid file by reading the <commentHeaderId> and the <comments-length>.

Then we go to the <control-header>, checking the <controlHeaderId> and reading

the <control-length>. Finally, we check that

<control-length>+<comments-length>+<data-length> = <file-length>.

The <data>, <comments>, and <control> are written node by node starting at the root.

For each node the following steps are taken:

1. Output a start node control byte of the appropriate sort.

either startNode, terminalTextNode, otherNode, otherNodeShort, or

something in the startNodeFirst..startNodeLast range or

in the terminalTextNodeFirst..terminalTextNodeLast range.

2. Output the properties, if any, of the node.

if the value is a rope, output ropeProp or ropePropShort

else if the value is a node, output nodeProp or nodePropShort.

then output the value itself

text for props is put in <control> rather than <data>

3. Output the contents of the node.

for a text node, first write the runs info, if any, and then the rope.

if it is a comment node, then the text goes in comments,

else it goes in data.

for other varieties of node, write otherNodeSpecs or otherNodeSpecsShort

followed by the text of the specifications.

4. Output the children of the node, if any, in the same manner

5. Output an endNode byte, unless this is a terminal text node

Following the end of the root node, there is an endOfFile byte.