mac-rom/Tidbits/PictWhap.a

;
;	File:		PictWhap.a
;
;	Contains:	This is an FKEY that will save the screen as a PICT file, for use by all other programs
;				as data files. 
;
;	Written by:	Bo3b Johnson
;
;	Copyright:	<09> 1988-1991 by Apple Computer, Inc., all rights reserved.
;
;	Change History (most recent first):
;
;		 <5>	 8/23/91	JSM		Remove benign redefinition of FALSE, which is now defined by the
;									build script.
;		 <3>	 1/21/91	csd			& PKE: Use SANE and the Script Manager to build the file name.
;									This will let us localize the file name so that it works for all of the
;									various script cases, i.e. right to left text in Hebrew.
;		 <2>	 9/25/90	dba			Check in Bo3b<33>s change to use D7 instead of D2. This caused a crash in
;									32-bit mode.
;		 <1>	 1/17/90	fjs			Recover from stupid mistakes
;

; PICTWhap.a;
;
;	For best viewing, use Palatino 12.
;
;	Copyright <20> 1988, 1989, 1990, 1991 Apple Computer, Inc.
;	All rights reserved.
;
;	This is an FKEY that will save the screen as a PICT file, for use by all other programs
;	as data files.   The PICT file is the Apple standard for transferring graphics between
;	programs.  
;
;	Bo3b Johnson: 9/21/88
;
;
; Build using the PictWhap.make file, and the standard build menu.
;
;---------------------------------------------------------------
;	*** bugs list:
;		Bus Error during ClosePicture, DisposCCursor, if Heap Scramble is on.
;			Unlikely to be in this code.
;---------------------------------------------------------------
;
;	9/26/88:
;		Fixed the problem with overwriting the header of the PICT, now leaving space
;		for the header before starting the spooling.
;		Killed the SetGDevice/GetGDevice stuff as Darin noted it isn<73>t necessary.
;	9/27/88:
;		Skip the buffer finding operation, the spooling is sufficiently fast to make it
;		unnecessary, even on floppies.  Timing results with Kong:  about 5 seconds
;		with floppy using bottlenecks, 4 seconds straight buffer.  On Mac+ with old
;		HD20, 2 seconds with BN, 1.5 seconds without.  On Mac II HD, 1.5 second with
;		BN, 0.5 without.  There is thus little incentive to make the code more complicated.
;	9/28/88:
;		Add the jShieldCursor hack to make the cursor appear in the saved image.
;		As per a Darin suggestion to avoid having to go find the cursor data and CopyBits.
;	3/29/89:
;		About to release it for D1 BigBang.  Change the creator of the PICT files to
;		'View' to match the PAT/Viewer program.  Added the asm/link lines in the
;		docs below so it can be done the same each time.  Set the place to create the
;		files as the root directory of the startup volume, instead of default.  Also
;		change the other file system calls to find the file in the root.  This makes it
;		HFS only.
;	5/1/89:
;		Change the use of globals in the FKEY code block to a StackFrame instead.  The
;		PictWriter globals are still stored in the code block however.  Killed the use of
; 		MyNum2String in deference to the ROM based Pack for _NumToString.
;		Changed the use of BootDrive into the sysVRefNum field from SysEnvirons, since 
;		we need to call SysEnvirons to determine color QD or not anyway.
;		Change the creator to 'TTxt' for TeachText to be the owner instead.  
;	5/7/89:
;		Added the code to make it do all available screens, not just main screen.  Uses the
;		GrayRgn to get to all the devices in the world. 
;		Change the CrsrCouple low memory global to keep mouse from moving while we
;		snapshot.  Added code to save the registers we trash around both the FKEY proc
;		itself, as well as the PictWriter routine.  
;	5/10/89:
;		Clearing the incredibly mindless ioFFlType for HFS since it is a swine and
;		sets us up for the big fall.  If this isn't cleared too, then Standard file will barf on
;		the file, and not display it in the list.  Notably there is also ioFileType which must
;		be cleared too.  Neither are used by the file system, or the Mac, but we still have
;		to do this.  Thank you HFS. 
;	5/11/89:
;		Finalizing the code.   Changed the init code to clear all the variables in a big loop,
;		added a few comment lines in PB blocks to show all parameters expected for each
;		call.  Verified each call to be sure each field is cleared in advance, or explicitly set.
;		Removed the code that would handle a missing STR resource, check for error instead.
;	5/19/89:
;		Calling it 1.0B2 now, since it is essentially complete.  Change creator one more time,
;		to 'ttxt' to have TeachText open the files.  (lower case is it?, Hah!  tricked myself.)
;		Added change to have it use A5 as the pointer to QD globals, as per a Darin<69> concept.
;		This also makes it possible to remove the global in the code.
;	10/26/89:
;		Version 1.0B3, and changed to DC.L for the 'FKEY' string.  Also made big changes 
;		to make it do banding in low RAM/Big Ass screen situations.  The banding is done
; 		in a simple fashion to make it easy to write/debug, and relies on QD returning from
; 		the CopyBits with no change in the picture size, or setting QDError.
;	10/30/89:
;		Numerous minor fixes that were suggested in code reviews.  Things like MOVEQ
;		instead of CLR and so on.  Up the version number to 1.0B4 since I half released B3.
;	11/13/89:
;		More fixes related to code review.  Set up string to be 'Picture ^0' so we can Munger
;		in the file number, killed MyConCat.  Changed the cursor protection stuff to set
;		CrsrVis to false, and decrement the cursor state.  This freezes the mouse, as well
;		as keeping it on the screen.  No more ShieldCursor hack, nor CrsrCouple.  
;	11/14/89:
;		Set up a hard coded string to handle the case where the STR resource may get lost.
;		The 'Picture ^0' is thus in the code, turned into a StringHandle in case we can't
;		get the STR.  This is unlikely, but may happen, and we can handle it easily.
;	8/20/90:
;		Added the code to SndPlay a camera noise when we start up.  This gives immediate
;		feedback that it is doing something, then goes on to do the saving.  If we get an error,
;		you get a beep as well as the camera click.  Also fixed a bug where it would bail
;		out to DeathCity in the middle of the device loop, causing the mouse to remain
;		frozen in place.  Moved the mousefreeze and state save to early on, and restore
;		the state in CloseNClean.
;	9/3/90:
;		Fixed a bug introduced by system build, where the file name as a string was bigger
;		than the length byte, so GetHandleSize was adding that extra pad byte to names.
;		Now it uses the length bytes and does all the math to get string length.  Change
;		build file so that the resources don't have any names, as desired by build guys.
;	9/25/90:
;		Fixed stupid bug where SndPlay would trash register D2 in 32-bit mode.  I was being
;		dumb using a scary register.  I moved it to use D7 instead, and save and restore D7.
;	1/21/91:
;		Changed the way that the picture file name is generated by making the number
;		calculation and appending code use the Script Manager. This required: 1) Using
;		SANE to increment an extended number, 2) using the script manager routine
;		FormatX2Str to convert it to a string, and 3) using the script manager routine
;		ReplaceText instead of Munger to merge the base name and the calculated number.
;
; Strategy:
;	The FKEY 3 is intended to take a snapshot of the screen when invoked, saving the image
;	into a file for editing/printing by an appropriate graphic editor.
;	Codewise the basic idea behind the FKEY is that when invoked it will do a CopyBits call using
;	the screen as the source and the destination, while an OpenPicture is under way.  The screen
;	is obviously the source, but it is also the destination to avoid having to create some huge old
;	buffer to copy the bits into.  This approach will still save the bits as long as a Picture is
;	Open.  The bits get saved, and we can write the bits out as a file with the right header in it to 
;	make it a PICT file.  The sequence it will use is:
;	1) Create and open the file to use.  This will use a string out of the system file, to allow
;	for localization of the generated files.  The numbers will be appended to the name, starting
;	at 1, and going up.
;	2) We will use the bottlenecks in QuickDraw to spool the data out line by line.  This is 
;	sort of slow but takes no RAM.
;	3) We create a new port that we can play with, using OpenPort.  This port will be changed
;	in size, as well as installing bottlenecks in the PutPicProc.
;	4) After the bottlenecks are set up we can do the OpenPicture to create a new picture.
;	5) Do the CopyBits (main, main, ...) to write the bits through our proc with the bottlenecks.
;	6) As CopyBits runs, it will call our proc to have the bits written to the disk as it goes along.
;	7) Do each monitor in turn, if we are on colorQD machines.
;	8) Close the file.
;	9) Close the port we created.
;	10) Split.
;
;  Some overall observations:  
;	There appeared to be a dependancy on A0 or A1, causing crashes, so just to be extra safe
; 	we will save all registers we use, even though we shouldn't have to save A0-A1/D0-D2.  
;	For a long, disk writing routine like this, it won't hurt to be extra robust.  All registers
;	are thus preserved around both the FKEY itself, and the PictWriter routine, since it is
;	not clear what can be trashed as part of a bottleneck routine.  The philosophy here is to
;	be extra safe, to save development time.
;
;	I looked into setting up the parameter block we use in A2, then using that throughout
;	the FKEY, but it didn't appear to be worth it.  It would save only two lines of code, at the
;	expense of making it harder to figure out, and more prone to bugs when modified.  The
;	current approach is to load A0 with the address of the block each time it is required.
;
;	In keeping with the goal of trying to keep it small, the spurious lines of initialization
;	for parameter blocks have been commented out.  They are left there as a notice that
; 	those lines are required for an unknown parameter block, but that we know they were
;	cleared or set immediately before.  Thus, it should be clear what is required for the 
;	call, without taking up code space to actually do it.  This is in contrast to my normal
;	approach of explicity setting each field used, to minimize copy/paste errors.
;
;	We don't preflight for space on the disk to write the file.  We have to handle any write
;	errors later in the code anyway, and DiskFull is just another form of a write error,
;	so there seems to be little point in taking the extra code to preflight.  Since the files are
;	of non-obvious lengths (they are packed with PackBits), we couldn't do a good job
;	preflighting, even if we wanted to.
;
;	Error checking and handling on all possible calls. The DeathCity error handler 
;	can be called at anytime, and it will do the right thing in terms of closing the port,
;	killing the picture, closing the file and so on, as long as the variables are cleared
;	before the first possible error.
;
;	If I get an error during creating, naming or preflighting the file I can<61>t finish
;	the job, so I need to clean up and split.  A SysBeep is wimpy user feedback on 
;	errors, but there isn't a whole lot better.  It doesn't really seem appropriate to
;	put up a dialog from an FKEY either.  The only Expected error is DiskFull.
;
;	Any errors during the Write or during the bottleneck Writes will be handled
;	similarly.  This makes it possible to error out in the bottleneck, and avoid getting
; 	recurring errors.
;
; 	The Lone Ranger in the form of Darin showed up to get around the problem of
;	needing to access our globals from the PictWriter bottleneck routine.  The solution
;	is to set A5 to point into our global space like usual, and to set the contents
;	of that first location to point to the current QD globals.  This way we use the
;	A5 world of whoever we are invoked within, and thus the PictWriter can get
;	to our globals via A5 (the only register we can rely on there, other than PC).
;	This is preferable to storing a pointer in the code block and doing a PC relative
;	move during the bottleneck routine.  The setup for the A5 is a little funky but
;	it should be possible to understand it without too much work.  Of course, if I
;	had pictures I'd show you stack frames and A5's pointing around and so on,
;	but until then, we'll just have to make do with words.  I can't bring myself to 
; 	use NoRes graphics after all this time.
;
;	HFS is a pig.  It is poorly documented and the interface is clunky and hard to use.
;	When doing all HFS calls, I would expect things to be more lucid.  There are two
;	file version number parameters, one for Create, and one for SetFileInfo it would
;	appear.  These are not documented, with the exception of two technotes, one talking
;	about one flag, the other talking about the second.  HGetFileInfo will return me a
;	bogus number in ioFFlType, thank you very much.  This number if then applied via
;	SetFileInfo will then make the file not show up in StdFile.  Another particularly fine
;	approach.  If these numbers are used by anybody, then why does StdFile filter them?
;	Should StdFile be fixed to not care?  What this also points out is that the documentation
;	is not sufficient to use HFS.  A better approach will always be to clear the full 
;	parameter block in a loop, and not try to be accurate about what needs to be set and
;	what doesn't.  In fact, I'm going to change this code to do that instead.
;
;	Given the poor shape of the HFS interface and documentation, we will now adopt the
;	philosophy of clearing the parameter block before we start, and not explicitly set 
;	any of the fields in the block, knowing they are zeroed.  This will buy us a little 
;	code savings, and should be a more robust approach.  HFS seems to be pretty consistent
;	about expecting zero for unused, or unknown, fields.  This is less than optimal, since
;	it would be better to know what exactly is necessary, and set those specific fields. 
;	But, since we are in doubt, it is better to play it safe and overkill.  It is more 
;	important to be robust than accurate.  Thus all our globals on the stack will be 
;	cleared in a single loop, where we will walk through the entire stack frame.  This will
;	set all our variables to zero, as well as the parameter block.  (Notably, this is what
;	Andy Hertzfeld did in RMaker...  are you scared yet?)  In keeping with HFS paranoia,
;	we will set all fields in the parameter block, whether it is an HFS version of the call
;	or not.  The documentation is not clear if this is necessary, but The Debugger indicates
;	it is, so better safe than spending hours tracking down something bogus.
;
;	We are going to save and restore A5 around the FKEY, and will modify it from inside
;	the FKEY.  This will allow us to have no globals stored in the code, and PictWriter as
; 	the bottleneck will still be able to get to our globals.  This way we avoid the self-modifying
;	code in order to store PC relative variables.  The QD globals still need to be accessed
;	by QD however, so we still have to provide that path.  The way to do that here,
; 	is to copy the 0(A5) pointer that QD uses to hunt down the globals, and store it in
;	our variables block, keeping the same link.
;
;	Another change here that seems to be dictated by 32-bit QD, and old systems.  We want
;	to band the image in the cases where there is not enough RAM.  Under 32-bit QD, a 
;	big buffer might need to be made to copy the bits into before saving them in the picture.
;	For a two-page screen this may be megs of data, causing us to create an empty picture,
;	which is lame.  Instead, we will look for the failing condition, and chop the image
;	up into small pieces (bands).  The technique we will use here will just keep shrinking
;	the rect that we use as the source and destination.  If it gets too small we'll bail out and
;	signal an error.  Otherwise we will keep a copyHeight that will be used to size each band
;	vertically.  Each time we fail during a CopyBits, we will divide the copyHeight by two,
; 	until it either succeeds or we get too small (like a single horizontal row of pixels).
;	Part of the trouble is that a color table gets saved for each band, but we don't
;	really expect to have to do that many bands.  Also, it seems better to make a bigger 
;	banded picture (on disk), than to fail from too little RAM.   In system 6.0 there is a patch
; 	that handles the source and destination both being the screen, and if so it will not
;	create a buffer.  It appears that we cannot necessarily rely on that, due to the problem 
;	of needing to access 32 bit screens.  32 bit QD handles the drawing, but the OpenPicture
;	can't do it directly, and needs a buffer.  It may be fixed in the future, and this approach
;	will cleanly default to a no buffer/single copybits situation.  This banding will be done
; 	on a per device basis.  The only slightly skanky aspect of doing it this way means that
;	we need to look at the picture size after the CopyBits call, and if it didn't change size,
;	we know it failed.  We'll also check QDError to be safer, but it is a implementation
;	detail of the picture mechanism.  Another possible approach would be to do the 
;	buffer allocation ourselves, knowing about the various depths and so on, which seems
;	excessively complicated, and requires even more knowledge of the QD model, like
;	pixelSize.  The algorithmic pseudo-code is like:
;
;		copyHeight = screenRect.b-screenRect.t
;		copyRect = screenRect
;		copyTop = screenRect.t
;	Copy:
;		copyRect.t = copyTop
;		copyRect.b = copyTop + copyHeight
;		If copyRect.b > screenRect.b Then copyRect.b = screenRect.b
;		gWroteBits = false
;		CopyBits (screen, screen, copyRect, copyRect, scrCopy, nil)
;		If NOT gWroteBits or QDError Then
;			copyHeight = copyHeight / 2
;			If copyHeight < 2 Then Failure
;			Goto Copy
;		copyTop = copyTop + copyHeight
;		If copyTop < screenRect.b Then Goto Copy
;
;	Also remember that the banding loop is surrounded by the device loop for color QD,
;	so they are nested.  
;
; A little info about the sound.  I looked into doing a 3:1 compression on the sound, to save
;	disk space, but it doesn't seem to be worth it.  The sound is already sampled at 11K, 
;	which seems the best compromise.  At 3:1, that is 3:1 off of the 22K sample, so it is
;	still 2/3 the size of the 11K sound.  It is not a big savings, and the sound is worsened.
;	In addition, the compression jazz doesn't work in 6x systems, in case this gets copied
;	over there.  At 6:1 compression, it is totally gross.  The compression seems really lame,
;	and not an advantage over just going down in sample speed.   Trying some more
;	tests, the 3:1 compression actually does sound better than 11K sample, so it would
;	be preferred, but it isn't a huge win, so the lack of 6x compatibility wins out.
;
;
; Some changes suggested via Code Review:
; 	Another minor change, the string that defines the file name is now 'Picture ^0' to start
; 	with, and can be changed at will for internationalization reasons.  The ^0 is replaced
;	at run time with the current file number string (from Num2String).  This is easy to 
;	do using Munger and took about the same amount of code as my old ConCat routine,
;	but is better for localizing.
;
;	Rather than setting the CrsrVis flag specifically when we are done, I will save 
;	and restore the old value, in case it is more sophisticated in the future.
;
;	I changed the code to decrement the CrsrState, set the CrsrVis to false,
;	then do the CopyBits.  This freezes the mouse, keeps it on the screen, and
;	allows it to be restarted after the CopyBits, so we capture the bits of the cursor in the
;	picture as well.  Previously I was doing a fake jShieldCursor, which was too sick, and
;	this approach seems to have the same effect, whilst merely futzing with low memory
;	globals.  Unfortunately, I have slightly more confidence that the vector will behave 
;	the same way in the future, but am unconvinced that the low memory globals even 
;	work as they are described.  Just how screwed up is the cursor code anyway?
;	This appears to work fine, but if there is any compatibility risk in this code, that's it.
;
;
;	Yes, I realize some of the comments get a little redundant, but you know, I'd rather say
;	it twice than to not say it once...
;
;
; 	Open Questions:
;		1) Should we do the menu down snapshots?  This is pretty tough to do, will make
;		it bigger, and risk more compatibility.  Is it worth the effort?  The proposed approach
;		is to create our own heap block when invoked, turn it into a heap only we use, set
;		up our own A5 world and QD world in this heap, start up a Time Manager task to
;		call us back later, wait 3 seconds or so, when called by TM, change to our heap which
;		will be valid at interrupt time, swap out QD globals that need it (unknown list, to date),
;		do the CopyBits with bottleneck spooling, close file, set up a Deferred Task Manager
;		task to call us back, at DTM time (which is GNE time) dispose the heap block we
;		created.  So... Technically challenging, but feasible.  Again, is it worth it?
;		-- answer: not.  Do it in the system.
;		2) Is it too sick to use GrayRgn?  To dangerous?  Any better way to do it?
;		3) This only runs on the latest system.  It will only be shipped with that system,
;		but if anyone copies and pastes it around, it won't work.  System 6.0.3 is required
;		in order to have the multiple monitor part work.  The CopyBits bug is fixed in
;		that version.  Is it bad to require the newest (I prefer to make it live with anything).
;		Requires HFS too.
;		-- answer: no biggy.  works now on system 4.2 with banding, HFS is OK.  
;		4) This saving of the picture will create a PICT 2 type picture, even if the current
;		color depth is only 1 bit deep.  Is this bad?  Will it make pictures that are hard for
;		Mac+ to read?  If I remember correctly, the Mac+ system patches allow it to read
;		PICT 2, and extract the picts if they are B&W...
;		5) If the string for the file name isn't available in the system we will bail out with
;		an error.  Is this reasonable?  Should we try to be more robust here with a default
;		string?
;		-- answer: added the change to make it default to 'Pic ' as the name.


	PRINT		PUSH,OFF						; don't print any of the headers.
	INCLUDE 		'Traps.a'
	INCLUDE 		'PackMacs.a'
	INCLUDE 		'ToolEqu.a'
	INCLUDE 		'QuickEqu.a'
	INCLUDE 		'SaneMacs.a'
	INCLUDE 		'ScriptEqu.a'
	INCLUDE 		'SysEqu.a'
	INCLUDE 		'SysErr.a'
	PRINT		POP							; restore the PRINT options


; Constants that are defined here instead of putting them in the code.

headerSize		EQU		512					; 512 bytes in PICT files for header info.
fileNameId		EQU		-16385				; resource ID number for 'Picture' string.
pFileType		EQU		'PICT'				; file type we create.
pFileCreator		EQU		'ttxt'				; file creator for teachtext.
pSoundType	EQU		'snd '				; resource type for sound resources.
cameraId		EQU		-16504				; resource Id of camera noise.

PICTWhap		PROC


; The StackFrame we are using here is just for local variables for the FKEY.  They
; would be globals, but FKEY's can't have globals, and this puts them on the stack,
; instead of in the code.  The QDglobalPtr is to be the 0th location in the list, so that
; 0(A5) will be that pointer.  This is the way QD expects it.  Thus baseName will be
; -4(A5), baseNum is -8(A5) and so on.

FkeyGlobals		RECORD	{QDglobalPtr},DECR	; build a stack frame record

;------------------------------------------------------------------
;------------------------------------------------------------------

; Here I define all the variables that are used.  These are
; used in a global variable fashion, even though they are allocated on the stack.
; FKEYs can<61>t have global vars though, so this is a reasonable way out.
; RA5 will still be used to point to our globals, so we set up the first element of 
; our globals to be the pointer to the QD variables as QD expects.  This is relatively
; complicated, so peruse the Segment loader and Memory manager chapters to see
; how all the variables are to be set up, or take my word for it.
; Our globals will still be referenced as some negative number off of A5.
; A6 is not used as such, except to give us a handy frame easily.
; It is possible to use a straight Link A5, but the sequence of code is thus critical,
; since we need to pull out the old 0(A5) before we change A5.  Rather than make it
; extra tricky, hard to read, hard to modify, and hard to maintain, in order to save 2 
; lines of code; I left it as LINK A6.

QDglobalPtr		DS.L		1					; pointer to QD globals, so QD can still work.

baseName		DS.L		1					; This will become a StringHandle to the name.
copyBaseName	DS.L		1					; copy of the handle, will be modified.
baseNum		DS.B		10					; an 80-bit extended, SANE compatible
baseNumString	DS.B		32					; smaller than Str255, to save unused space.
hBaseNumString	DS.L	1					; handle to copy of the number string.
fileName		DS.B		32
filePB			DS.B		ioFQElSize			; big enough for the SetFileInfo too.
newGrafs		DS.B		cqdProcsRec
picPort			DS.B		portRec
portIsValid		DS.W		1
oldCrsrVis		DS.W		1					; saved value from CrsrVis.
currPort			DS.L		1
sysInfo			DS			SysEnvRec			; for colorQD test, and boot volume refnum. 
maxRect		DS.W		4					; big rect containing all monitors.
nextGD			DS.L		1					; handle for current GDevice when doing Mac II

; The next few variables are for the banding operation.  These are used on a per device basis,
; and only really change in really low RAM situations, or Big Ass monitors.  All these variables
; are defined on a per band basis.  One band can be the entire device though.

copyHeight		DS.W		1					; the height used for each CopyBits band.
copyRect		DS.W		4					; the rectangle used for each band.
copyTop		DS.W		1					; the starting location of a given band.

; These next few are used by the PictWriter routine directly.

gPictRefNum	DS.W		1
gPictSize		DS.W		1
gPictError		DS.W		1
gPictHandle		DS.L		1
gWroteBits		DS.W		1					; for memory full checking.

; Here we have some variables which are used while doing file name construction in an
; international-friendly manner.

itlHandle		DS.L		1					; handle to itl resource with the NumberPartsTable
numPartsOffset	DS.L		1					; offset into the itl of the NumberPartsTable
itlState			DS.W		1					; so we can save/restore the state of the handle
tableLength		DS.L		1					; length of the table (not used)

; The end of the Stack Frame as our globals.  This LocalSize will thus be subtracted from the 
; stack to store the variables.

				ALIGN	4						; make sure we stop at a nice 4 byte boundary.
												; This is so we can clear the whole block easily.
GlobalSize		EQU 		*-FkeyGlobals		; size of all the local variables
				ENDR


	WITH		FkeyGlobals						; use our 'global' stack frame

;------------------------------------------------------------------
;------------------------------------------------------------------

	
; The start of the FKEY, with the conventional FKEY header on it to identify it and the
; current version of this little wag.

Start
	BRA.S		@0								; skip over the text, obviously.
	
	STRING	ASIS							; turn off length byte. 
	DC.L		'FKEY'							; a hunk of data in the code.
	DC.W		$0003							; FKey 3 for the hex viewer.
	DC.W		'1.0B6'							; version number.
	
; Set up some of our variables to initial values.   And save off the current gDevice
; and port.  In case of errors, we can restore the environment to what it was.  
; Call SysEnvirons to determine if color QD is here or not, since that determines
; the use of gDevices or not.

@0
	LINK		A6,#GlobalSize					; allocate our local stack frame 'globals'.
	MOVEM.L	A0-A2/A5/D0-D2/D7,-(SP)		; save registers we trash.   (9/25/90)
	
; Since we don't trust HFS, or anybody else, including ourselves, we will clear all of the
; global data space to zero.  This starts us in a nice known condition, a welcome change of
; pace.  Do all the variables in a big loop, 4 bytes at a time.  The number stowed in D0 is
; a positive DBRA count of the number of longwords in our global area.  GlobalSize is a 
; negative number, so we invert it before dividing by 4, then subtracting 1 to make it DBRA.

	MOVEQ	#(-GlobalSize/4)-1,D0			; total size of our space, in longwords, DBRA count.
	MOVE.L	A6,A0							; the start of the global space.
@1
	CLR.L		-(A0)							; clear the next 4 bytes.  A6 offsets are negative.
	DBRA		D0,@1							; until we got them all.

; Now set up the globals pointer in A5.  This will point to our block on the stack, but we will
; use A5 to get there.  We also save off the QD link, so we can use some one else's QD world for
; our picture creation.
	
	MOVE.L	(A5),A0							; save off the link.
	MOVE.L	A6,A5							; swap globals over to our space,
	SUBQ		#4,A5							; make it so 0(A5) is QDGlobalPtr.
	MOVE.L	A0,QDGlobalPtr(A5)				; set the link back so QD is all set.

; 8/17/90: B3B:
; Before we start the show, make the noise of a camera so that they know
; that something happened.  If we get an error somewhere inside, we'll get a beep as well,
; but it doesn't come out too bad.  We handle the case of the sound being missing, by
; just skipping the attemp.  It won't cause a problem.  We lock the sound handle just to
; be extra safe.  We shouldn't have to, but the sound manager has a sordid history, and
; it is preferable to be safe, rather than explicitly correct.  We want to
; do it async, but can't really do that since we might error out somewhere in the middle,
; giving a weird noise effect.  Also, we can't leave until the sound is done, and don't 
; want to screw around trying to find out if it is done or not.

	CLR.L		-(SP)							; room for handle result.
	MOVE.L	#pSoundType,-(SP)				; the type of resource, sound.
	MOVE.W	#cameraID,-(SP)					; the resource ID of the camera noise.	
	_GetResource								; get the sound as a resource.
	MOVE.L	(SP)+,D7						; retrieve the handle to the sound. (9/25/90)
	BEQ.S		GetSysInfo						; if it is gone, skip making a sound.
	
	MOVE.L	D7,A0							; the handle to the sound. (9/25/90)
	_HLock										; lock it down while we play.
	
	CLR.W		-(SP)							; room for function result.
	CLR.L		-(SP)							; no sound channel from us,
	MOVE.L	D7,-(SP)							; the handle to the noise, locked. (9/25/90)
	MOVE.W	#False,-(SP)						; and do it is synchronously.
	_SndPlay									; play that camera noise.
	MOVE.W	(SP)+,D0						; get the result code, but ignore it.
	
	MOVE.L	D7,A0							; retrieve the handle to the sound, (9/25/90)
	_HUnLock									; and unlock it, so it can be purged.
	
; GetSysInfo:
;	We need to determine if colorQD exists on the machine or not, so call SysEnvirons
; to find out.  This also gets the boot volume vRefNum, which is more robust than the
; BootDrive low memory global.  We don<6F>t link with the glue here since we know that
; the trap has to exist on machines where this FKEY will run.  We can always expect
; Version 1 to exist, and thus don't check the error code on return.

GetSysInfo
	LEA		sysInfo(A5),A0					; address of the record.
	MOVEQ	#1,D0							; version 1 record, and info desired.
	_SysEnvirons								; fill the environment record on stack.

; Save the current port so we can restore it at the end.

	PEA		currPort(A5)						; the var to set,
	_GetPort									; save off the current port

; Now we need to do a sick thing.  This is to make the cursor visible in the pict, once it
; is saved off.  The problem is that CopyBits will ShieldCursor in order to make it not 
; appear in an offscreen picture.  We need it there in order to make it a real screen
; snapshot.  We will mark the cursor as Invisible so that QD will think that it doesn't
; have to ShieldCursor.  We do this by setting it invisible, and decrementing the CrsrState.
; We also want to freeze the mouse while this goes on, since we don't want the mouse to
; change around during the save.  It takes long enough that we can get multiple smashed
; copies of the cursor.  The mouse is frozen by the CrsrState change.  This has to be done
; before the first possible DeathCity, so that we can be sure to restore the state properly.

	MOVE.B	CrsrVis,oldCrsrVis(A5)			; save the old value.
	SF			CrsrVis							; Pretend cursor is invisible.
	SUB.B		#1,CrsrState						; Turn cursor off.
	
; Now try to get the STR resource for the name of the file we want to use.

	CLR.W		-(SP)							; returned refnum, will be left here.
	_CurResFile									; where are we currently?  (save this place.)
	
	CLR.W		-(SP)							; switch to the system file.
	_UseResFile									; use the system file, to avoid getting wrong string.
	
	CLR.L		-(SP)							; the handle result.
	MOVE.W	#fileNameId,-(SP)				; the rez id for it.
	_GetString									; get it if we can.
	MOVE.L	(SP)+,baseName(A5)				; save handle to the string.
	
												; the refnum is still on the stack.
	_UseResFile									; return resource chain to how it was.
	
; Get the international table for number parts so we can use it when building
; the file name.

	MOVE.W	#iuSystemScript, -(SP)			; desktop items are in the System script
	MOVE.W	#iuNumberPartsTable, -(SP)		; we need the number parts table
	PEA		itlHandle(A5)					; place to store the itl4 handle
	PEA		numPartsOffset(A5)				; offset into the itl4 of the NumberPartsTable
	PEA		tableLength(A5)					; somewhere to put this, not used by us
	_IUGetItlTable								; try to get the number parts table
	MOVE.L	itlHandle(A5), D0				; did we get the table?
	BEQ		DeathCity						; if not, probably low on memory
	MOVE.L	D0, A0							; the itl4 resource handle
	_HGetState									; remember its present state
	MOVE.B	D0, itlState(A5)					; save it for a later restore
	_HLock										; don<6F>t let it go anywhere (or get purged)
	
; We have the handle to the STR resource.  Make sure it is valid.

	TST.L		baseName(A5)					; is it non-nil?
	BNE.S		TryFileName					; if it is valid, go ahead and use it.
	
; For the case where we couldn't find the string, a silly case arguably, but one we want to
; allow them the option of making, let's hard code a file name that will be used instead
; of the string.  Turn this hardcoded string into a handle, and store the handle.

	CLR.L		-(SP)							; returned handle result.
	PEA		hardName(PC)					; address of string in code,
	_NewString									; turn it into a handle,
	MOVE.L	(SP)+,baseName(A5)				; save off this handle instead.
	BEQ		DeathCity						; if it is nil, bail out, no ram.

; If there are files there already, I want to keep upping the number to
; find an unused file name.  There is not much point in stopping at 9,
; so I<>ll just keep trying until it fails, or I get  an OK file name. 
; We have a valid handle to the string, which starts as 'Picture ^0'.  Make a copy of it so we
; can change the handle with ReplaceText.  We do this each time through the TryFileName
; loop so we can start with a fresh 'Picture ^0' each time.  We copy the handle instead
; of using the resource, to avoid having to hit the disk each time we try a new name.

TryFileName
	MOVEQ	#0, D0							; start with a clean register
	MOVE.L	baseName(A5),A0				; the handle to current string,
	MOVE.L	(A0), A1							; point to the current string
	MOVE.B	(A1), D0							; get the length of the string
	_NewHandle								; try to make a handle of the right size
	BNE		DeathCity						; get out if we run out of memory
	MOVE.L	A0, copyBaseName(A5)			; save this handle so we can nuke it later
	MOVE.L	(A0), A0							; point to the handle data
	MOVEQ	#0, D0							; start with a clean register
	MOVE.B	(A1)+, D0						; get the length to copy & point to first charater
	EXG		A0, A1							; make source & dest registers correct
	_BlockMove
	
; Increment our number by one. We use SANE because we don<6F>t want this code to
; know about the extended type implementation. Also, since we<77>re going to use the
; international utilities to convert this to a string, and they call SANE, there<72>s no
; extra hit because we call SANE here. Finally, SANE in ROM is likely to be used,
; saving us from loading it in any case.

	PEA		extendedOne					; the constant value 1
	PEA		baseNum(A5)					; our local counter
	FADDX										; baseNum := baseNum + 1

; Now convert the number to a string so we can hook it into the picture string.
; We use the script manager to be completely international-friendly. This will almost
; certainly cause a RAM hit, unfortunately.

	PEA		baseNum(A5)					; the number to convert
	PEA		numFormat						; how to format the number
	MOVE.L	itlHandle(A5), A0				; handle to itl4 resource
	MOVE.L	(A0), A0							; pointer to the (locked) itl4 resource
	ADDA.L	numPartsOffset(A5), A0			; adjust pointer to point at table
	MOVE.L	A0, -(SP)						; push the NumberPartsTablePtr
	PEA		baseNumString(A5)				; where to put the formatted string
	_FormatX2Str								; script manager-friendly NumToString

; Now combine the two strings, the name, plus the number as a string to build a
; file name to try.  We will build the string using ReplaceText, to replace the ^0 in the 
; string with the current baseNumString. Since ReplaceText requires two handles, we have
; to first copy the number string into a handle.

	MOVEQ	#0, D0							; start with a clean register
	MOVE.B	baseNumString(A5), D0			; get the length byte
	_NewHandle								; try to make a handle to hold it
	BNE		DeathCity						; if we get an error, just leave
	MOVE.L	A0, hBaseNumString(A5)		; save the handle for later
	MOVE.L	(A0), A1							; point to the new handle data
	MOVEQ	#0, D0							; start with a clean register
	LEA		baseNumString(A5), A0			; point at our copy of the string
	MOVE.B	(A0)+, D0						; get the length & point at the first character
	_BlockMove								; copy the string into the handle
	
	CLR.W		-(SP)							; return result from ReplaceText
	MOVE.L	copyBaseName(A5), -(SP)			; the handle to the copy of the file name
	MOVE.L	hBaseNumString(A5), -(SP)		; the handle to the number string
	PEA		upSnick0Str(PC)					; the key to replace
	_ReplaceText								; substitute the number for ^0
												; function result removed below
												
; Throw away the handle which contained a copy of the number string

	MOVE.L	hBaseNumString(A5), A0
	_DisposHandle								; kill it
	
; Set the length byte for this new string, using the length of the handle resulting from
; the ReplaceText call. Then copy the string data from the handle into our local file
; name string.

	MOVE.L	copyBaseName(A5), A0			; handle to complete file name
	_GetHandleSize								; how many bytes?
	MOVE.L	(A0), A0							; point at data
	LEA		fileName(A5), A1				; point at local file name string
	MOVE.B	D0, (A1)+						; move the length byte & point past it
	_BlockMove								; copy the rest of the file name
	
; Now we have a copy of the next name to try, kill the string handle, since we have to start
; with the resource in order  to get the ^0 in the string.

	MOVE.L	copyBaseName(A5),A0			; the handle to string,
	_DisposHandle								; kill it.
	
; Now we have cleaned up the handle, see if ReplaceText worked or not.  If it didn't, it
; couldn't find the ^0 in the string which would imply something very bad happened.
; We do this sort of late here, so that we can blow away the StringHandle without
; having to worry about it in the error handler.

	MOVE.W	(SP)+,D0						; get result from ReplaceText,
	BMI		DeathCity						; if couldn't find it, something is wrong.

; Now try to create the file based on that file name, on the current volume.
; The parameter block is set up for both Create and Open. 
; Create the file in the root of the startup volume.  The rest of the parameter block
; is zeroed, from above, just in case.

	LEA		filePB(A5),A0					; the file parameter block,
	LEA		fileName(A5),A1				; the address of the file name string,
	MOVE.L	A1,ioFileName(A0)				; set up parameter,
	MOVE.W	sysInfo.sysVRefNum(A5),ioVRefNum(A0) ; use startup volume.
;	CLR.B		ioFVersNum(A0)				; clear file type, of course.
	MOVE.L	#fsRtDirId,ioDirID(A0)			; create in the root directory.
	_HCreate									; try to create that file.

; If we have an error other than dupFNErr, then we are toast and have to exit.
; If it is a dupFNErr, then we need to up the file name/number by one and 
; try again.

	CMP.W		#dupFNErr,D0					; is it a duplicate file name?
	BEQ		TryFileName					; if so, try next one.
	TST.W		D0								; was result OK?
	BNE		DeathCity						; If not, we must die.

; The file is valid, set the file information so that it is marked as a PICT file.  Start
; with a GetFileInfo to set up all the junk we don't need to fool with.  Recall that we
; comment out the lines that aren't specifically required in this location in the code,
; but leave the lines there to remind you that they are required for the call.  They
; are already set up by the above HCreate call, or by the clearing process at the start
; of the code.

;	LEA		filePB(A5),A0					; the file parameter block,
;	LEA		fileName(A5),A1				; the address of the file name string,
;	MOVE.L	A1,ioFileName(A0)				; set up parameter,
;	MOVE.W	sysInfo.sysVRefNum(A5),ioVRefNum(A0) ; use startup volume,
;	CLR.W		ioFileType(A0)					; no version number, any permission.
;	CLR.W		ioFDirIndex(A0)					; use name and vRefNum.
;	MOVE.L	#fsRtDirId,ioDirID(A0)			; get in the root directory.
	_HGetFileInfo								; get the info on the slime.
	BNE		DeathCity						; everything OK?
	
; Now change the file type and creator and Set the info.

;	LEA		filePB(A5),A0					; the file parameter block,
;	LEA		fileName(A5),A1				; the address of the file name string,
;	MOVE.L	A1,ioFileName(A0)				; set up parameter,
;	MOVE.W	sysInfo.sysVRefNum(A5),ioVRefNum(A0) ; use startup volume,
;	CLR.W		ioFileType(A0)					; no version number, any permission.
;	CLR.B		ioFFlType(A0)					; and the other file type Must be cleared too.
	MOVE.L	#pFileType,ioFLUsrWds+fdType(A0) ; change file type to PICT.
	MOVE.L	#pFileCreator,ioFLUsrWds+fdCreator(A0) ; and the file creator to 'ttxt' for TeachText
												; Creation and ModDates are set in block,
												; from the GetFileInfo.
	MOVE.L	#fsRtDirId,ioDirID(A0)			; set in the root directory. (HGetFileInfo pounds
												; the dirId on the way back.)
	_HSetFileInfo								; Set the info on the slime.
	BNE		DeathCity						; everything OK?
	
; We have a valid file name, and the file has been created on the default volume.
; So... it is time to open the slime dog.  We use the 'write' permission, but there are
; a couple of other stupid things in the file system, like not reporting an error here if
; the disk is write protected.  We get that later in the Write call, so we'll handle a
; no write access error there too.  If everything is OK, we'll just get Rd/Wr permission
; back by default.  This is fsWrPerm though, just to be explicit.

;	LEA		filePB(A5),A0					; the file parameter block,
;	LEA		fileName(A5),A1				; the address of the file name string,
;	MOVE.L	A1,ioFileName(A0)				; set up parameter,
;	MOVE.W	sysInfo.sysVRefNum(A5),ioVRefNum(A0) ; use startup volume,
	MOVE.B	#fsWrPerm,ioPermssn(A0)		; desire Write permission to file.
;	CLR.B		ioFileType(A0)					; clear the, worse than useless, version #.
;	CLR.L		ioOwnBuf(A0)					; use standard buffer.
;	MOVE.L	#fsRtDirId,ioDirID(A0)			; open in the root directory.
	_HOpen									; open the file for writing.
	BNE		DeathCity						; if it cannot be opened, we cannot continue.

	MOVE.W	ioRefNum(A0),gPictRefNum(A5)	; saved off for future use.

; Now write the header information into the file, so that it is a PICT file.   It is actually
; 512+10 bytes long, since we throw in the picture header as well.  The space needs to
; be allocated in the file, although it will be updated after the picture is done.
; This is kind of a cheater approach to writing an empty header, but serves two purposes.
; It will get clear block we can write at one shot to clear the header on the file, and it also
; has to get 522 bytes of storage or we die here.  Since QD will crash on no memory for 
; the upcoming OpenPort, this is a minor preflight to be sure we won't die when we try
; to open the port, or open the picture.  OK, so I'm a slacker, but I really just wanted to
; use the Clear bit.

	MOVE.L	#headerSize+picData,D0			; size of the block.
	_NewHandle	CLEAR						; make a block of zeros.
	BNE		DeathCity						; are we so low we can't get 512 bytes?

	MOVE.L	A0,A2							; save off the handle for later disposal.
	LEA		filePB(A5),A0					; the parameter block again,
;	MOVE.W	gPictRefNum(A5),ioRefNum(A0)	; the refnum to use.
	MOVE.L	(A2),ioBuffer(A0)				; pointer to the handle of data.
	MOVE.L	#headerSize+picData,ioReqCount(A0) ; number of bytes to write.
	MOVE.W	#fsFromStart,ioPosMode(A0)		; start at start in file,
	CLR.L		ioPosOffset(A0)					; no offset from start.
	_Write										; Write the empty header to the file.
	
; Now dispose the handle we used to write the empty header to the file.  Disposed
; before the error check to avoid having to dispose it in the error handler.

	MOVE.L	A2,A0							; handle to the data,
	_DisposHandle								; freed back to system.

; Check the result of the write.

	TST.W		filePB+ioResult(A5)				; retrieve the error code from Write.
	BNE		DeathCity						; if we got an error, leave.

; Now open a port for our
; use to allow us to easily copy the screen to itself, and to avoid having to worry if
; a port is valid currently.  We have preflighted at least 500 bytes out of the heap,
; which should be enough for a port, so we won<6F>t blow up on this call at least.
; In addition to the Open, this becomes the current port. (as: thePort)  The
; same record is used whether it is color or not.

	PEA		picPort(A5)						; address of record to fill
	TST.B		sysInfo.hasColorQD(A5)			; do we have color QD?
	BEQ.S		@0								; if not, open a regular port.
	_OpenCPort									; fill it with color port junk.
	BRA.S		@1
@0
	_OpenPort									; open a standard port.
	
@1
	ST			portIsValid(A5)					; Set bool to Yes, for the picPort being in use.

; Now that we have a new port to draw into, we want to be sure to get all the monitors/
; devices in the system.  So... use the GrayRgn from low memory to Union with the current
; visRgn we have in our new port.  We want to do the Union since the new Port's visRgn will 
; include the menu bar area, and GrayRgn doesn't.  But of course, the new port doesn't have
; the other devices, so we need the union.  This is also considered a little sick, that of 
; changing the visRgn of a port, but it needs to be done.  Once we've changed the visRgn,
; we'll get the whole system when we CopyBits. 

	MOVE.L	picPort+visRgn(A5),-(SP)			; the handle to the visRgn for the new port.
	MOVE.L	GrayRgn,-(SP)					; the handle to GrayRgn out of low mem.
	MOVE.L	picPort+visRgn(A5),-(SP)			; the visRgn as the destination as well.
	_UnionRgn									; modify the visRgn to include all monitors.

; Now save the bounding rect that encompasses all the monitors.

	MOVE.L	picPort+visRgn(A5),A0			; the handle to the visRgn.
	MOVE.L	(A0),A0							; pointer to the region.
	LEA		maxRect(A5),A1					; the address of our Rect,
	MOVE.L	rgnBBox+topLeft(A0),(A1)+		; copy over the topLeft,
	MOVE.L	rgnBBox+botRight(A0),(A1)+		; and the botRight
	
; Set the new GrafProc record up to have the standard pieces.  Here again,
; use the same record for both color and not, but use the larger color record,
; not filling all the space in the non-color case.

	PEA		newGrafs(A5)					; the new array to use.
	TST.B		sysInfo.hasColorQD(A5)			; do we have color QD?
	BEQ.S		@2								; if not, set std procs, not color.
	_SetStdCProcs								; fill the array with procs.
	BRA.S		@3								; go on to change procs.
@2
	_SetStdProcs								; fill the array with B&W procs.
@3

; Change the port to use those GrafProcs instead. 
; thePort^.grafProcs := @newGrafs

	LEA		picPort(A5),A0					; get address of thePort.
	LEA		newGrafs(A5),A1				; address of new procs record,
	MOVE.L	A1,grafProcs(A0)					; set the port to the new record.

; We are in our offscreen port.  Change the GrafProc pointer for picture saving
; to our routine that actually writes the bytes to the disk. Works for both color and
; regular.

	LEA		PictWriter(PC),A0				; the address of the routine to use,
	MOVE.L	A0,putPicProc(A1)				; change the newGraf record.

; Now when the putPicProc gets called, it will run to our PictWriter routine.  This
; way we get to save the data off to the disk instead of use up huge hunks of RAM to
; build the picture.  PictWriter will use the gPictRefNum as the file to write to, but we
; need to init a couple of other global variables to set up for the picture.  The two other
; variables were already cleared by the top loop that cleared all the globals.

	MOVE.W	#picData,gPictSize(A5)			; start off as just header size.
;	CLR.W		gPictError(A5)					; starting as no error.
;	CLR.L		gPictHandle(A5)					; since we have no picture handle yet.

; Actually open the picture and do the CopyBits in order to process the picture.
; The data will be written by PictWriter as it is called by QuickDraw. 

	CLR.L		-(SP)							; the handle result comes back on stack.
	PEA		maxRect(A5)					; the rect to use, full screen.
	_OpenPicture								; start up the process.
	MOVE.L	(SP)+,gPictHandle(A5)			; save the handle created.

; Now the picture is open and ready to cruise.  Start by setting the clip region down,
; to make it a more mellow picture (clipping weirdnesses abound).  This clip rect is the
; rect that surrounds all monitors in the system.

	PEA		maxRect(A5)					; the rect to use again,
	_ClipRect									; clip to the screen bounds.

; Loop through each monitor in the device list.  With each monitor, do a CopyBits of the 
; full device, just the bounds rectangle. 

	TST.B		sysInfo.hasColorQD(A5)			; do we have color QD?
	BEQ.S		CopyLips						; if not, skip the GDevice stuff.
	
; Otherwise, we need to drive through the device list, so get the first one.

	CLR.L		-(SP)							; returned GDHandle here.
	_GetDeviceList								; get start of list.
	MOVE.L	(SP)+,nextGD(A5)				; save the handle.
	BEQ		DeathCity						; if no devices, get out of here.
	
; Loop back here to CopyBits the next device in the list.  The maxRect as we use it here is
; for a given gDevice/screen, and isn't the total area of all monitors, as it was when we
; setup the clipping.

DoNextDevice
	MOVE.L	nextGD(A5),A0					; get the handle to gDevice,
	MOVE.L	(A0),A0							; deref the handle,
	MOVE.L	gdPMap(A0),A0					; get PixMapHandle,
	MOVE.L	(A0),A0							; deref to pointer of PixMap.
	LEA		maxRect(A5),A1					; the address of rect to save,
	MOVE.L	bounds+topLeft(A0),(A1)+		; save off the TopLeft of the rect,
	MOVE.L	bounds+botRight(A0),(A1)+		; save the botRight too.
	
; For each gDevice that we copy, we want to start off the banding parameters as being the
; full screen size, basically the maxRect parameters.  If this is a single monitor old system,
; the concept still holds.  Start the copyTop at the top of the current screen, start the copyHeight
; as the full screen size, and start the copyRect as the full screen.  These will all get changed
; if there isn't enough ram to do a full screen copybits.  If there is enough ram, we'll get 
; a single CopyBits saved in the picture for this screen/gDevice.

CopyLips
	MOVE.W	maxRect+top(A5),copyTop(A5)	; start first band at top of screen.
	MOVE.W	maxRect+bottom(A5),D0			; get the bottom of screen,
	SUB.W		maxRect+top(A5),D0				; giving height of screen,
	MOVE.W	D0,copyHeight(A5)				; save it off as starting copyHeight.  
	LEA		copyRect(A5),A1					; address of our copyRect,
	MOVE.L	maxRect+topLeft(A5),(A1)+		; save off the TopLeft of the rect,
	MOVE.L	maxRect+botRight(A5),(A1)+		; save the botRight too.

; Set up the banding parameters for this time through the band loop.   This may be a one
; time shot, but we can't be sure, and don't care, so just do it generally.

BandLoop
	MOVE.W	copyTop(A5),D0					; get the top of the next band to do,
	MOVE.W	D0,copyRect+top(A5)				; set the top of the copyRect to new top.
	ADD.W		copyHeight(A5),D0				; add in current band height, giving bottom.
	
	CMP.W		maxRect+bottom(A5),D0			; are we off bottom of full screen rect?
	BLT.S		@0								; if not, stow the value.
	MOVE.W	maxRect+bottom(A5),D0			; if we are, just stop at the max bottom.
@0
	MOVE.W	D0,copyRect+bottom(A5)			; the bottom of the next rectangle (band).
	
; The parameters are all setup for this band (maybe full screen).  
; The picture is ready to roll.  CopyBits the screen to itself.  While inside of an open
; picture this will save all the bits off into the picture handle.  Since the handle is
; being written to disk as it is created by the PictWriter, this CopyBits will automatically
; spool the data out to the disk.  We use the copyRect since we may only be doing a band
; of the full screen right now. 

	SF			gWroteBits(A5)					; set up to handle full memory case.
	
	PEA		picPort+portBits(A5)				; the source bitMap,
	PEA		picPort+portBits(A5)				; the destination bitMap too.
	PEA		copyRect(A5)					; the source rectangle, max screensize.
	PEA		copyRect(A5)					; the dest rect too.
	MOVE.W	#srcCopy,-(SP)					; use the copy mode.
	CLR.L		-(SP)							; and a Nil clipping region.
	_CopyBits									; copy the screen bits.

; We have written that band out to the file using the bottleneck.  If QD failed by not having
; enough RAM we need to shrink the band size down.  We know whether it failed or not
; by checking the gWroteBits variable to be sure the bottleneck got called, and we'll also 
; check QDError just to be sure.  QDError is only valid for color machines.

	TST.B		sysInfo.hasColorQD(A5)			; do we have color QD?
	BEQ.S		@1								; if not, skip QDError check.
	CLR.W		-(SP)							; return result,
	_QDError									; get last error,
	MOVE.W	(SP)+,D0						; get value QD last set,
	BNE.S		ShrinkCopyHeight				; if nonZero, try shrinking the band Height.
@1
	TST.B		gWroteBits(A5)					; and check the write flag,
	BNE.S		UpCopyTop						; if it is true, we wrote bits, and can go on.

; We have a case where the CopyBits to save the image didn't work, so the bottleneck never
; got called to write any data.  Working under the assumption that it is a memory full error,
; we want to shrink our band by half, and try again.

ShrinkCopyHeight
	MOVE.W	copyHeight(A5),D0				; get the current copyHeight
	LSR.W		#1,D0							; divide the height by two.
	MOVE.W	D0,copyHeight(A5)				; save off the new number.
	
	CMP.W		#1,D0							; are we down to 1 pixel high yet?
	BLE.S		DeathCity						; if we are, just skip it, not enough ram.
	BRA.S		BandLoop						; try this smaller band instead.

; We didn't get any error while doing that band.  Up the copyTop to the next band to be done,
; and check if we are done with the entire max rect yet.  If we are done, we can fall into the
; outer loop of the gDevices on the system.  If not, we'll go back to do another band for this
; current gDevice.

UpCopyTop
	MOVE.W	copyTop(A5),D0					; get the band top we just used,
	ADD.W		copyHeight(A5),D0				; add in the current bandsize,
	MOVE.W	D0,copyTop(A5)					; save that new band top.
	CMP.W		maxRect+bottom(A5),D0			; are we off the end of the device yet?
 	BLT.S		BandLoop						; not off bottom of device, so do another band.
	
; If we have colorQD we need to get the next device and copy it.  If we don't have color QD
; skip this stuff, since it will cause a system error if we run it.  Can you say 'special case?'
; Sure, I knew you could.

EndDeviceLoop
	TST.B		sysInfo.hasColorQD(A5)			; do we have color QD?
	BEQ.S		PictureDone						; if not, skip the GDevice stuff.
	
; We have done that monitor.  Now go to the next one in the list.  When we get to the
; end we will get NIL back.

	CLR.L		-(SP)							; make room for the result.
	MOVE.L	nextGD(A5),-(SP)				; where we are now,
	_GetNextDevice								; get us the next device in the list.
	MOVE.L	(SP)+,nextGD(A5)				; move the gDevice handle.
	BNE		DoNextDevice					; jump back to CopyBits more if there is more.
	

;------------------------------------------------------------------
; That was a ton of work, but we didn't see it, PictWriter did.
; Now we are done saving the bits to disk, we can shut down the picture saving operation.

PictureDone
	_ClosePicture								; close the picture.

; We are done with the picture operation, see if we had an error while it was going on.
; We have to check afterwards, since QD has no error checking, by design, for some
; unknown reason.

	MOVE		gPictError(A5),D0				; was there an error during spooling?
	BNE.S		DeathCity						; if so, we are dead meat, and must leave.

; Now we know the file was written OK, so we have to drive back to the front of the
; file and update the pict size so it is correct.  Do the Positioning and Write operation
; in the same call.  
; Now write the 10 byte header off the pict handle to the disk, so that info is correct
; in the Pict file.  The size and rect need to be written, since they won't be right 
; until the picture is complete.

	LEA		filePB(A5),A0					; the PB to use.
	MOVE.W	gPictRefNum(A5),ioRefNum(A0)	; use the refnum we opened with,
	MOVE.L	gPictHandle(A5),A1				; the handle to the pict,
	MOVE.L	(A1),ioBuffer(A0)				; the buffer to write.
	MOVE.L	#picData,ioReqCount(A0)		; the number of bytes to write.
	MOVE.W	#fsFromStart,ioPosMode(A0)		; do offset from the start of file,
	MOVE.L	#headerSize,ioPosOffset(A0)		; and offset past the 512 byte header.
	_Write										; write the final data.
	BNE.S		DeathCity						; if an error, split.

; From here on out, we are cleaning up the world to what it was before we came in.
; That means we won<6F>t have any failures (recoverable anyway) during this exit.

	BSR.S		CloseNClean					; close the file, clean up blocks.

	BRA.S		CommonExit					; and leave cleanly.
	

;------------------------------------------------------------------
; This is the error handler.  This is where we have died from some error during the
; whole operation.  When we get here, we have to dispose of all the junk we created,
; clean up, let the user know we chooked, and exit.  *** does there need to be something
; more sophisticated to let the user know we died?  Notification request perhaps?

DeathCity
	MOVE.W	#8,-(SP)							; length for SysBeep,
	_SysBeep									; a beep to signify errors.
	
	BSR.S		CloseNClean					; close the file and cleanup.
	
; Since we had some skanky error, attempt to delete the file we created at the start,
; to clean up bogus junk.  If we get an error on the delete that<61>s OK.

	LEA		filePB(A5),A0					; the parameter block,
;	LEA		fileName(A5),A1				; the pointer to the file name string,
;	MOVE.L	A1,ioFileName(A0)				; the file name to delete,
;	MOVE.W	sysInfo.sysVRefNum(A5),ioVRefNum(A0) ; use startup volume, as above.
;	CLR.W		ioFileType(A0)					; and no file type.
	MOVE.L	#fsRtDirId,ioDirID(A0)			; file was in the root directory.
	_HDelete									; try to kill it.
	
	
;------------------------------------------------------------------
CommonExit
	MOVEM.L	(SP)+,A0-A2/A5/D0-D2/D7		; restore registers we trashed.  (9/25/90)
	UNLK		A6								; kill local vars from stack.
	RTS										; Exit from FKEY, successful or not.
	

;------------------------------------------------------------------
; Close the File and clean up our tracks.
; 	Everything is done, put the CrsrVis/CrsrState back before someone sees that
; 	we were playing with it.  If we get caught, we face life in prison.  This is done
;	whether we have an error or not.

CloseNClean
	MOVE.B	oldCrsrVis(A5),CrsrVis			; restore the cursor visibility.
	ADD.B		#1,CrsrState						; Turn cursor back on.
	
; Restore the state of the itl4 resource handle if we got one
	MOVE.L	itlHandle(A5), D0				; did we have this handle
	BEQ.S		@3								; if not, don<6F>t bother
	MOVE.L	D0, A0							; get the handle into A0
	MOVEQ	#0, D0							; a clean register is probably not needed
	MOVE.B	itlState(A5), D0					; get the saved handle state
	_HSetState									; and make it like it used to be

@3
; Close it in case it is open.  If it is not open, gPictRefNum will be zero.

	LEA		filePB(A5),A0					; the pb block to use,
	MOVE.W	gPictRefNum(A5),ioRefNum(A0)	; the file refnum we just used.
	BEQ.S		@2								; if no file was open, skip it.
	_Close										; close the file.
@2
; Kill the picture handle that was created at OpenPicture time, if it exists.
; If it never made it, the handle will be nil, allowing us to skip it.

	MOVE.L	gPictHandle(A5),D0				; get the handle to the picture,
	BEQ.S		@1								; if it is NIL, skip the Kill.
	MOVE.L	D0,-(SP)							; else, the handle to the stub on stack,
	_KillPicture									; kill the swine.
@1
	MOVE.L	currPort(A5),-(SP)				; and the old port,
	_SetPort									; so we are back.

; And dispose the port that we created before, since we are done.

	TST.B		portIsValid(A5)					; is the port valid, set up?
	BEQ.S		@0								; if not, skip the close,
	PEA		picPort(A5)						; address of port to blast.
	_ClosePort									; close the dog, releasing memory.
@0
	RTS
	
	
;------------------------------------------------------------------
; PictWriter:
; 	This routine is the bottleneck procedure that actually gets called when QuickDraw
; is slamming through the bits while copying the screen.  For each time it gets called
; it will save off the bytes passed and check the error stuff. 
; As found in a Pascal routine:
;
;	{ This routine will save the current image as it is created.  As the data requests
;		go by that data will be written to the file.  The data is being created by the
;		OpenPicture/CopyBits in DoWrite, this is the bottleneck for that operation.
;		Any errors found while doing this will make us skip any further requests
;		to write data to the disk.  No memory is allocated.   Communication with 
;		DoWrite is done through globals, since bottlenecks must be at the main
;		level.  The bottleneck must also keep track of how many bytes are written,
;		so that the header on the picture can be fixed up to be correct.  This must
;		be done to avoid creating bogus pictures.  The picSize field of the handle
;		must be updated continuously so that when the picture is done, the ClosePicture
;		can create a valid picture.  The check for the NIL handle is to handle the
;		problem of when the OpenPicture is called.  The proc gets called before 
;		the handle is valid.  Be very careful of these bottleneck things, it is 
;		easy to run into problems that are very hard to figure out.  QuickDraw
;		has no facilities to give you info when things go wrong so it makes it
;		a bit tougher. }
;PROCEDURE  PictWriter (dPointer: Ptr; nextHunk: Integer);
;
;VAR	longHunk:	LongInt;
;
;BEGIN
;	IF gPictError = noErr THEN  BEGIN
;		longHunk := nextHunk;
;		gPictError := FSWrite(gPictRefNum, longHunk, dPointer);
;		gPictSize := gPictSize + longHunk;
;		IF gPictHandle <> NIL THEN  gPictHandle^^.picSize := LoWord (gPictSize);
;	END;
;END;

; For the PictWriter parameters, we will just use an A7 relative stack frame.  The
; A5 stack frame for our globals will still be in effect from the WITH at the top, and
; we will explicitly use the PictWriterStack parameters when we need them here,
; doing the relative (SP) as expected, instead of (A5).

PWStack		RECORD	{EndRegs},DECR
PBegin			EQU		*					; start parameters after this point
dPointer		DS.L		1					; offset to the pointer,
nextHunk		DS.W		1					; offset onto stack to parameter,
PWSize			EQU		PBegin-*			; to remove these dudes.
RetAddr		DS.L		1					; return address on stack.
SaveRegs		DS.L		6					; save/restore 6 registers in here.
EndRegs		EQU		*					; to set the location.
				ENDR

PictWriter
	MOVEM.L	A0-A2/D0-D2,-(SP)				; save registers we trash. 

	MOVE.W	gPictError(A5),D0				; get current error result.
	BNE.S		NoCanDo						; if we have an error pending, exit now.

	ST			gWroteBits(A5)					; mark us as having had enough ram.
	
	CLR.L		D0								; start with empty register
	MOVE.W	PWStack.nextHunk(SP),D0		; get size of next piece to write.
	
	LEA		filePB(A5),A0					; address of i/o parameter block,
	MOVE.W	gPictRefNum(A5),ioRefNum(A0) ; the refnum to use.
	MOVE.L	PWStack.dPointer(SP),ioBuffer(A0) ; address of data to write.
	MOVE.L	D0,ioReqCount(A0)				; number of bytes to write, as longInt.
	MOVE.W	#fsAtMark,ioPosMode(A0)		; start at mark in file,
	CLR.L		ioPosOffset(A0)					; no offset from current mark.
	_Write										; write the data to the open file.
	MOVE.W	D0,gPictError(A5)				; set the error code to yea/nay.
	
; Now the data is out there in the file, update the picture size variable.

	MOVE.W	PWStack.nextHunk(SP),D0		; retrieve the count of bytes to do,
	ADD.W		D0,gPictSize(A5)					; add, ignore overflows, QD does.
	
; And update the word in the header of the Picture Handle to match.

	MOVE.L	gPictHandle(A5),D0				; handle to the data block,
	BEQ.S		NoCanDo						; if nil, skip it for now.
	
	MOVE.L	D0,A0							; put back in Address land.
	MOVE.L	(A0),A0							; deref to pointer.
	MOVE.W	gPictSize(A5),picSize(A0)			; set the size parameter.

NoCanDo
	MOVEM.L	(SP)+,A0-A2/D0-D2				; restore registers we trashed. 
	MOVE.L	(SP)+,A0						; retrieve return address.
	ADD.L		#PWStack.PWSize,SP			; kill the parameters,
	JMP		(A0)							; and return.

	ENDWITH									; done with the global Frame.
	

;------------------------------------------------------------------
; Here we have a string constant that we want to use in order to do a Munger thing
; on the creation of the filename.  We will search a given file string for the ^0 and replace
; it with the current file number we are creating.  This ^0 allows us to be more 
; internationally compatible, since the number can be at the front of the string which is
; nicer and more grammatically correct for some languages.  Not to mention it is easy
; for us to do it here.  This ^0 is defined here in the code as a constant, and isn't modified,
; but is used as the source for the string search in Munger.

	
	String		Pascal							; put length byte on this one.
	
upSnick0Str		DC.W	'^0'						; the string piece to find and replace.
	
	String		Pascal							; put length byte on this one.
	
hardName		DC.W	'Pic ^0'					; default name, in case STR is lost.
												; set to funny name to show something is wrong.
												
extendedOne									; the constant 1 in extended format
				DC.W	$3FFF, $8000, $0000, $0000, $0000, $0000
				
numFormat
				DC.B	$30,$00,$00,$05,$00,$00	; internal representation of the format
				DC.B	$00,$00,$00,$05,$00,$05	; for our number <20>####<23>
				DC.B	$00,$05,$00,$04,$00,$00
				DC.B	$00,$00,$00,$00,$00,$00
				DC.B	$00,$00,$01,$00,$00,$00
				DC.B	$00,$00,$00,$00,$00,$00
				DC.B	$00,$00,$00,$00,$00,$00
				DC.B	$00,$00,$04,$05,$05,$05
				DC.B	$05
				
	ENDP										; end of proc.
	
	END