$PostgreSQL: pgsql/src/backend/utils/misc/README,v 1.10 2008/03/20 17:55:15 momjian Exp $

Guc Implementation Notes
========================

The GUC (Grand Unified Configuration) module implements configuration
variables of multiple types (currently boolean, enum, int, float, and string).
Variable settings can come from various places, with a priority ordering
determining which setting is used.


Per-Variable Hooks
------------------

Each variable known to GUC can optionally have an assign_hook and/or
a show_hook to provide customized behavior.  Assign hooks are used to
perform validity checking on variable values (above and beyond what
GUC can do).  They are also used to update any derived state that needs
to change when a GUC variable is set.  Show hooks are used to modify
the default SHOW display for a variable.

If an assign_hook is provided, it points to a function of the signature
	bool assign_hook(newvalue, bool doit, GucSource source)
where the type of "newvalue" matches the kind of variable.  This function
is called immediately before actually setting the variable's value (so it
can look at the actual variable to determine the old value).  If the
function returns "true" then the assignment is completed; if it returns
"false" then newvalue is considered invalid and the assignment is not
performed.  If "doit" is false then the function should simply check
validity of newvalue and not change any derived state.  The "source" parameter
indicates where the new value came from.  If it is >= PGC_S_INTERACTIVE,
then we are performing an interactive assignment (e.g., a SET command), and
ereport(ERROR) is safe to do.  But when source < PGC_S_INTERACTIVE, we are
reading a non-interactive option source, such as postgresql.conf.  In this
case the assign_hook should *not* ereport but should just return false if it
doesn't like the newvalue.

If an assign_hook returns false then guc.c will report a generic "invalid
value for option FOO" error message.  If you feel the need to provide a more
specific error message, ereport() it using "GUC_complaint_elevel(source)"
as the error level.  Note that this might return either ERROR or a lower level
such as LOG, so the ereport call might or might not return.  If it does
return, return false out of the assign_hook.

For string variables, the signature for assign hooks is a bit different:
	const char *assign_hook(const char *newvalue,
				bool doit,
				GucSource source)
The meanings of the parameters are the same as for the other types of GUC
variables, but the return value is handled differently:
	NULL --- assignment fails (like returning false for other datatypes)
	newvalue --- assignment succeeds, assign the newvalue as-is
	malloc'd (not palloc'd!!!) string --- assign that value instead
The third choice is allowed in case the assign_hook wants to return a
"canonical" version of the new value.  For example, the assign_hook for
datestyle always returns a string that includes both output and input
datestyle options, although the input might have specified only one.

Note that a string variable's assign_hook will NEVER be called with a NULL
value for newvalue, since there would be no way to distinguish success
and failure returns.  If the boot_val or reset_val for a string variable
is NULL, it will just be assigned without calling the assign_hook.
Therefore, a NULL boot_val should never be used in combination with an
assign_hook that has side-effects, as the side-effects wouldn't happen
during a RESET that re-institutes the boot-time setting.

If a show_hook is provided, it points to a function of the signature
	const char *show_hook(void)
This hook allows variable-specific computation of the value displayed
by SHOW.


Saving/Restoring Guc Variable Values
------------------------------------

Prior values of configuration variables must be remembered in order to deal
with several special cases: RESET (a/k/a SET TO DEFAULT), rollback of SET
on transaction abort, rollback of SET LOCAL at transaction end (either
commit or abort), and save/restore around a function that has a SET option.
RESET is defined as selecting the value that would be effective had there
never been any SET commands in the current session.

To handle these cases we must keep track of many distinct values for each
variable.  The primary values are:

* actual variable contents	always the current effective value

* reset_val			the value to use for RESET

(Each GUC entry also has a boot_val which is the wired-in default value.
This is assigned to the reset_val and the actual variable during
InitializeGUCOptions().  The boot_val is also consulted to restore the
correct reset_val if SIGHUP processing discovers that a variable formerly
specified in postgresql.conf is no longer set there.)

In addition to the primary values, there is a stack of former effective
values that might need to be restored in future.  Stacking and unstacking
is controlled by the GUC "nest level", which is zero when outside any
transaction, one at top transaction level, and incremented for each
open subtransaction or function call with a SET option.  A stack entry
is made whenever a GUC variable is first modified at a given nesting level.
(Note: the reset_val need not be stacked because it is only changed by
non-transactional operations.)

A stack entry has a state, a prior value of the GUC variable, a remembered
source of that prior value, and depending on the state may also have a
"masked" value.  The masked value is needed when SET followed by SET LOCAL
occur at the same nest level: the SET's value is masked but must be
remembered to restore after transaction commit.

During initialization we set the actual value and reset_val based on
whichever non-interactive source has the highest priority.  They will
have the same value.

The possible transactional operations on a GUC value are:

Entry to a function with a SET option:

	Push a stack entry with the prior variable value and state SAVE,
	then set the variable.

Plain SET command:

	If no stack entry of current level:
		Push new stack entry w/prior value and state SET
	else if stack entry's state is SAVE, SET, or LOCAL:
		change stack state to SET, don't change saved value
		(here we are forgetting effects of prior set action)
	else (entry must have state SET+LOCAL):
		discard its masked value, change state to SET
		(here we are forgetting effects of prior SET and SET LOCAL)
	Now set new value.

SET LOCAL command:

	If no stack entry of current level:
		Push new stack entry w/prior value and state LOCAL
	else if stack entry's state is SAVE or LOCAL or SET+LOCAL:
		no change to stack entry
		(in SAVE case, SET LOCAL will be forgotten at func exit)
	else (entry must have state SET):
		put current active into its masked slot, set state SET+LOCAL
	Now set new value.

Transaction or subtransaction abort:

	Pop stack entries, restoring prior value, until top < subxact depth

Transaction or subtransaction commit (incl. successful function exit):

	While stack entry level >= subxact depth

		if entry's state is SAVE:
			pop, restoring prior value
		else if level is 1 and entry's state is SET+LOCAL:
			pop, restoring *masked* value
		else if level is 1 and entry's state is SET:
			pop, discarding old value
		else if level is 1 and entry's state is LOCAL:
			pop, restoring prior value
		else if there is no entry of exactly level N-1:
			decrement entry's level, no other state change
		else
			merge entries of level N-1 and N as specified below

The merged entry will have level N-1 and prior = older prior, so easiest
to keep older entry and free newer.  There are 12 possibilities since
we already handled level N state = SAVE:

N-1		N

SAVE		SET		discard top prior, set state SET
SAVE		LOCAL		discard top prior, no change to stack entry
SAVE		SET+LOCAL	discard top prior, copy masked, state S+L

SET		SET		discard top prior, no change to stack entry
SET		LOCAL		copy top prior to masked, state S+L
SET		SET+LOCAL	discard top prior, copy masked, state S+L

LOCAL		SET		discard top prior, set state SET
LOCAL		LOCAL		discard top prior, no change to stack entry
LOCAL		SET+LOCAL	discard top prior, copy masked, state S+L

SET+LOCAL	SET		discard top prior and second masked, state SET
SET+LOCAL	LOCAL		discard top prior, no change to stack entry
SET+LOCAL	SET+LOCAL	discard top prior, copy masked, state S+L


RESET is executed like a SET, but using the reset_val as the desired new
value.  (We do not provide a RESET LOCAL command, but SET LOCAL TO DEFAULT
has the same behavior that RESET LOCAL would.)  The source associated with
the reset_val also becomes associated with the actual value.

If SIGHUP is received, the GUC code rereads the postgresql.conf
configuration file (this does not happen in the signal handler, but at
next return to main loop; note that it can be executed while within a
transaction).  New values from postgresql.conf are assigned to actual
variable, reset_val, and stacked actual values, but only if each of
these has a current source priority <= PGC_S_FILE.  (It is thus possible
for reset_val to track the config-file setting even if there is
currently a different interactive value of the actual variable.)

The assign_hook and show_hook routines work only with the actual variable,
and are not directly aware of the additional values maintained by GUC.
This is not a problem for normal usage, since we can assign first to the
actual variable and then (if that succeeds) to the additional values as
needed.  However, for SIGHUP rereads we may not want to assign to the
actual variable.  Our procedure in that case is to call the assign_hook
with doit = false so that the value is validated, but no derived state is
changed.


String Memory Handling
----------------------

String option values are allocated with strdup, not with the
pstrdup/palloc mechanisms.  We would need to keep them in a permanent
context anyway, and strdup gives us more control over handling
out-of-memory failures.

We allow a string variable's actual value, reset_val, boot_val, and stacked
values to point at the same storage.  This makes it slightly harder to free
space (we must test whether a value to be freed isn't equal to any of the
other pointers in the GUC entry or associated stack items).  The main
advantage is that we never need to strdup during transaction commit/abort,
so cannot cause an out-of-memory failure there.