Name Last Update
..
extensions Loading commit data...
.gitignore Loading commit data...
Makefile.am Loading commit data...
README Loading commit data...
actions.c Loading commit data...
comparator.c Loading commit data...
conf.c Loading commit data...
encoded.c Loading commit data...
load.c Loading commit data...
mem.c Loading commit data...
prog.c Loading commit data...
register.c Loading commit data...
relational.c Loading commit data...
require.c Loading commit data...
runtime.c Loading commit data...
sieve-priv.h Loading commit data...
sieve.l Loading commit data...
sieve.y Loading commit data...
strexp.c Loading commit data...
tests.c Loading commit data...
util.c Loading commit data...
* Overview

A compiled sieve program consists of a sequence of cells. Each cell
is a pointer to sieve_op_t data, i.e. it points to an instruction handler
(sieve_instr_t type) or to a memory location containing sieve_value_t
structure. The exact interpretation of this pointer depends on the
value of program counter. The very first entry in a sieve program
(pc = 0) is always a NULL pointer. The next entry (pc = 1) is always
expected to contain a valid instruction handler. When executing a
program, the sieve runtime evaluator starts from pc = 1. It reads
the cell contents, interprets it as address of an instruction handler,
increments the program counter and executes the handler. The evaluator
stops executing the program when it encounters a NULL pointer.

When invoked, the instruction handler receives a single argument: a pointer
to the sieve_machine_t structure, describing current runtime environment.
If the handler needs any surplus arguments, it is its responsibility
to retrieve them from the program cells immediately following the
handler address and increment the pc value accordingly.

* Existing handlers

** Nop

Name: instr_nop
Arguments: None

Does nothing

** Push

Name: instr_push
Arguments: None

Pushes current numeric register on stack.

** Pop

Name: instr_pop
Arguments: None

Pops the top of stack value into the numeric register.

** Unconditional branch

Name: instr_branch
Arguments: [pc  ]  (number) Offset of the new cell.

** Branch if zero

Name: instr_brz
Arguments: [pc  ]  (number) Offset of the new cell.

** Branch if not zero

Name: instr_brnz
Arguments: [pc  ]  (number) Offset of the new cell.

** Logical NOT

Name: instr_not
Arguments: none

** Logical AND

Name: instr_allof
Arguments: [pc  ]  (number) Number of items to be popped from stack


** Logical OR

Name: instr_anyof
Arguments: [pc  ]  (number) Number of items to be popped from stack

** Action handler

Name: instr_action
Arguments: [pc  ]  (sieve_handler_t*) Pointer to the action handling function.
           [pc+1]  (list_t of sieve_value_t) A list of required arguments.  
           [pc+2]  (list_t of sieve_runtime_tag_t) A list of tags.
	   [pc+3]  (string) Name of the action (for debugging purposes).

** Test handler

Name: instr_test
Arguments: [pc  ]  (sieve_handler_t*) Pointer to the test handling function.
           [pc+1]  (list_t of sieve_value_t) A list of required arguments.  
           [pc+2]  (list_t of sieve_runtime_tag_t) A list of tags.
	   [pc+3]  (string) Name of the test (for debugging purposes).

* Conditional statement branching

A simple statement

	IF cond list1 ELSE list2

is translated into the following program

# Compute the condition
	.	
	.
	.
	brz	L_ELSE
# Evaluate list1
	.
	.
	.
	br	L_END
L_ELSE:
# Evaluate list2
	.
	.
	.
L_END:
	
A more complex statement in the form:

	IF cond1 list1 ELSIF cond2 list2 ELSE list3

is translated into the following program

# Compute the condition 1
	.	
	.
	.
	brz	L_ELSIF
# Evaluate list1
	.
	.
	.
	br	L_END
L_ELSIF:
# Compute the condition 2
	.
	.
	.	
	brz	L_ELSE
# Evaluate list 2
	.
	.
	.
	br	L_END
L_ELSE:
# Evaluate list 3
	.
	.
	.
L_END:

Generally speaking, each ELSIF branch generates a code, similar to usual
IF ... ELSE sequence.

* Boolean shortcuts

Boolean shortcuts are implemented for coding ALLOF and ANYOF conditions.
The code these produce is shown in the next two subsections. Notice the
insertion of the two Nop statement after the last condition. These replace
the two slots used by the Brz or Brnz instruction, which would be useless,
since the next statement after ALLOF or ANYOF is guaranteed to be a branch
statement. This replacement speeds up the execution a little.

** ALLOF

ALLOF(cond1,cond2,cond3)

# Evaluate cond1
	.
	.
	.
	brz	L_end
# Evaluate cond2
	.
	.
	.
	brz	L_end
# Evaluate cond3
	.
	.
	.
	nop
	nop
L_end:
	.

** ANYOF

ALLOF(cond1,cond2,cond3)

# Evaluate cond1
	.
	.
	.
	brnz	L_end
# Evaluate cond2
	.
	.
	.
	brnz	L_end
# Evaluate cond3
	.
	.
	.
	nop
	nop
L_end:
	.


Local variables:
mode: outline
paragraph-separate: "[ 	]*$"
end: