Interlisp-D Implementor's Reference Filed as: {Eris}Internal>Doc>ImplManual.Tedit Contents * Updating the Lisp Manual * Making a SYSOUT, the Loadup process * Low-level Storage Reference and Manipulation Functions * ABC, EXPORTS.ALL, etc. * SYSRECORDS * Installing new DLion microcode * WHEREIS & JONL's WHEREIS database * Archiving files onto MAXC * Installing New Op-Codes * Creating new DEVICES * VMEM.PURE.STATE * Partial list of system atoms Chapters elsewhere: * The DLion Low Level Disk Drivers (see ImplManual-DlionDiskDriver.tedit) * XDE Tutorials (see ImplManual-XDE.tedit) Updating the InterLisp Manual [Sannella]: Documentation files are kept on {eris}*.im. LispCore members are encouraged to modify these documentation files to reflect the changes made as the system is modified. Please be careful. One important note it is necessary to keep track of how the manual is changed, in order to provide a list of changes with the next revision of the manual. Therefore, I would strongly suggest that whenever anyone makes a significant change to the manual (adds/deletes a function definition, adds new arguments to a function, non-trivial rewording, etc.) that they send a short message to me (Sannella, not LispCore^). The manual is stored in a large number of separate files, and it is difficult to know which file contains a particular function definition. Therefore, I have created a small package that will take an "IM Name" (a function, variable, property name, etc), create a TEdit window on the appropriate IM file, and position the TEdit cursor at the right place. To use this, do _(FILESLOAD (FROM LISPUSERS) IMNAME) _(INSPECT.IM 'FOO) INSPECT.IM uses the hash file package to search a hash file containing index information for the name FOO. If it is found, it will put up a pop-up menu listing references in different files. Selecting one of the references will move move the cursor in the appropriate TEdit window (if there is an active TEdit window to the appropriate file), or create a new TEdit window to the appropriate file. Sometimes, a particular name is defined as more than one "type" (function, variable, etc.). In this case, a pop-up menu will prompt you to declare which type you are interested in. A somewhat more convenient way of using this facility, if you want to use it repeatedly, is to do _(MAKE.IM.INSPECTOR) This sets up an ``IM Inspector Window'', which contains a menu. Initially, this contains the single selection "Type an IM name", which prompts the user to type a name which will be looked up in the database. Below this window will appear type-selection and reference-selection menus, which do not disappear until another selection is made above them. This is hard to describe.... try it out. [It works great! -- LMM] MAKING A LOADUP [Masinter]: There are various command files on {Eris}Next>: LoadFull.CM makes a LISP.SYSOUT and a FULL.SYSOUT from scratch. LoadFullFromLisp.CM makes a FULL.SYSOUT directly from the LISP.SYSOUT. LoadDemoFromFull.CM makes a DEMO.SYSOUT LoadSmallFromDLINIT.cm makes a SMALL.SYSOUT from INIT.DLINIT. The first part of the command file runs MAKEINIT to create a file INIT.SYSOUT and then DLFIXINIT to make it dandelion bootable and merge in the dandelion microcode to create a INIT.DLINIT. The next part starts up the INIT.DLINIT (which will run on all machines), and calls LOADUP(HUGE). The function LOADUP (on the file APUTDQ which is merged in at MAKEINIT time) has directions on how to do various kinds of loadups. If you have a special kind of loadup that you think should be supported for some applications, it is quite reasonable merely to add a separate clause to LOADUP and include that in the standard source. How MAKEINIT works: Basically, all the storage-modifying functions are redefined so they make their changes to a file. Then the normal LOAD code runs, but the effects take place in the new SYSOUT file instead of in memory. Things to watch out for: The data type STREAM must be the first datatype declared after MAKEINIT time. This means that no file loaded before FILEIO can declare a datatype. Internal Storage Reference Functions [Masinter, van Melle]: There are a number of low-level functions for directly accessing memory as if it were an enormous array of 16-bit words, bytes, cells, etc. In general, don't call any of these directly if you can help it. They are generally "unsafe" and can confuse your system in subtle ways if misused. Also, there are often alternatives that, if not completely safe, are at least less prone to error: (A) Write functions or macros to do the accesses, and have them perform suitable type checking. See the GETPUPWORD and PUTPUPWORD macros, for example. (B) Define a BLOCKRECORD to overlay a given data structure. This is much better than using \GETBASE et al if you are using fixed offsetsit is safer (less error-prone) and generally produces better code. With creative use of LOCF and ACCESSFNS, you can often avoid using explicit \GETBASEs altogether, and your code is much more readable. Also check out the MESATYPES package, written by Tayloe Stansbury, for producing such expressions from Mesa type declarations. The argument BASE in the following functions refers to an address, an Interlisp pointer. For example, if the value of X is an instance of a datatype, then X is actually a pointer to the first cell of that instance. There are essentially only two operations that perform "pointer arithmetic": \ADDBASE and \VAG2; these compile directly into the ADDBASE and VAG2 opcodes. (\ADDBASE BASE OFFSET) Produces a new address that is OFFSET 16-bit words beyond BASE. (\VAG2 HI LO) Produces an address whose left 8 bits is HI and whose right 16 bits is LO. There are, however, many other ways to produce addresses that ultimately perform \VAG2 or \ADDBASE, and these are usually preferable. The record POINTER is useful for decomposing pointers into page# and word-in-page or cell-in-page quantities. LOCF is useful in conjunction with BLOCKRECORDs and DATATYPEs. (LOCF (fetch FIELDNAME of datastructure)) [Macro] Produces a pointer to the first word containing FIELDNAME. E.g., if BAR is declared as a WORD field in a record, then (fetch BAR of X) is equivalent to (\GETBASE (LOCF (fetch BAR of X)) 0). (INDEXF (fetch FIELDNAME of T)) [Macro] Returns the word offset to the first word containing FIELDNAME. Since this is independent of the actual datum being operated on, the datum is often given as "T". E.g., if BAR is declared as a WORD field in a record, then (fetch BAR of X) is equivalent to (\GETBASE X (INDEXF (fetch BAR of T))). There is rarely any need for INDEXF. Note that \ADDBASE, LOCF and other pointer-producing operations are not in general safe in Interlisp-D. The garbage collector can get very confused if you save away arbitrary pointers anywhere other than in local variables. This is because the reference count of an object is associated only with the pointer to its beginning, i.e., only with the address that the public traffics in (the pointer returned from create, for example). If you must store away internal pointers, be very careful that you continue to hold on to the pointer to the start of the object for as long as you maintain the internal pointer. This assures that the object will not get garbage-collected out from under you, the most common source of such confusion. (\ADDBASE2 BASE N) Equivalent to (\ADDBASE BASE 2*N). (\GETBASE BASE OFFSET) (\PUTBASE BASE OFFSET VALUE) These fetch and store, respectively, the 16-bit word (as a Lisp small positive integer) located at OFFSET words beyond BASE. \PUTBASE is really dangerous. E.g., (\PUTBASE NIL n) for many small values of n will smash your system beyond repair. Not good for a residential environment where a smashed system can lose a lot of work. (\GETBASEBYTE BASE OFFSET) (\PUTBASEBYTE BASE OFFSET BYTE) Fetch and store 8-bit quanta. BASE is a word address, and OFFSET is a byte offsetcounting the high byte of the base word as offset zero. (\GETBASEPTR BASE OFFSET) Fetches a pointer at OFFSET from BASE. A pointer is a 24-bit quantity, which is stored right-justified in a 32-bit cell. Note, however, that BASE and OFFSET are both still in terms of 16-bit words. (\PUTBASEPTR BASE OFFSET PTR) Stores pointer PTR at OFFSET from BASE. This is not a direct inverse of \GETBASEPTR, because it stores a full 32 bits, never mind what used to be the high 8 bits originally stored there. \PUTBASEPTR does not do reference counting, so this can be especially dangerous if not used carefully. BASE is a word address, and OFFSET is in words, not cells! \RPLPTR is similar to \PUTBASEPTR, except that it does do reference counting. (\RPLPTR BASE OFFSET PTR) Stores a 24-bit pointer, similar to \PUTBASEPTR, except that (a) it stores only 24 bits, preserving whatever used to be in the high 8 bits; and (b) it does reference-counting operations. This is the proper way to smash a pointer field if you must. However, there is almost never any need for you to call this directly; the usual way to smash a pointer field is to use records. Implementation notes: \GETBASEBYTE and \PUTBASEBYTE compile directly into the corresponding opcodes, and execute entirely in microcode when the OFFSET and VALUE arguments are small positive integers. \GETBASE, \PUTBASE, \GETBASEPTR, \PUTBASEPTR, and \RPLPTR compile directly into the corresponding opcodes when the OFFSET argument is a constant less than 256; for other OFFSET arguments (variable quantities, larger integers), they require an ADDBASE in addition. (\GETBASEFIXP BASE OFFSET) (\PUTBASEFIXP BASE OFFSET VALUE) These fetch and store 32-bit integers. (\GETBASEFLOATP BASE OFFSET) (\PUTBASEFLOATP BASE OFFSET VALUE) These fetch and store 32-bit floatps. (\GETBASESTRING BASE OFFSET NCHARS) Creates a string NCHARS characters long whose characters consist of the bytes located starting at OFFSET (a byte offset) from BASE. Thus, the first character of the result is (\GETBASEBYTE BASE OFFSET). (\PUTBASESTRING BASE OFFSET STRING) Stores the characters of STRING as consecutive bytes starting at OFFSET (a byte offset) from BASE. (\BLT DBASE SBASE NWORDS) Copies a sequence of NWORDS words starting at SBASE to corresponding words starting at DBASE. This compiles directly into the BLT opcode. In the case where the source and destination ranges overlap, the behavior is well-defined: words are copied from the end of the range backwards to the beginning. Thus, this is equivalent to (for I from NWORDS-1 to 0 by -1 do (\PUTBASE DBASE (\GETBASE SBASE I))). Very fine point: this operation is defined to be completely uninterruptable if NWORDS is less than 10; thus, you can use opcode to make small indivisible transfers. (\MOVEWORDS SBASE SOFFSET DBASE DOFFSET NWORDS) Obsolete predecessor of \BLT. (\MOVEBYTES SBASE SBYTEOFFSET DBASE DBYTEOFFSET NBYTES) Copies a sequence of NBYTES bytes starting at SBYTEOFFSET bytes beyond SBASE to DBYTEOFFSET bytes beyond DBASE. If the ranges overlap, the result is formally undefined. (\ZEROBYTES BASE FIRST LAST) Stores zeroes into the bytes at offsets FIRST thru LAST, inclusive, from BASE. Thus, a total of LAST-FIRST+1 bytes are cleared. (\ZEROWORDS BASE ENDBASE) Stores zeroes into the words from BASE thru ENDBASE, inclusive. There are obscure reasons for the lack of symmetry among \ZEROWORDS, \ZEROBYTES, and \MOVEBYTES. ABC, EXPORTS.ALL, &c [JonL]: I changed ABC and FILESETS to not reference ; also, placed all references in the DIRECTORIES list before any ones. Currently, looks in ...>SOURCES> and ...>LIBRARY> and finally in LISPUSERS> and . Note that for Interlisp-D, these files mutually expect each other to be on {PHYLUM}SOURCES>. We can make copies of EXPORTS.ALL on, say, Library>, but don't delete the copy from Sources>. Just a reminder: (LOAD 'MAKE-EXPORTS.ALL) will connect to {PHYLUM}SOURCES> and gather all exports into the EXPORTS.ALL file. ``ABC'' stands for ``A Byte Compiler''meaning the augmented environment required to compile Interlisp system code. The augmentation includes any of the definitions found under the EXPORT FILEPKG command. The variable EXPORTFILES, set up by loading the file FILESETS, contains the rootname of all system files which have any EXPORT commands. A file wil generally export those items that other files need (e.g., records or macros) which are DONTCOPY, and thus not part of the user's system. SYSRECORDS [van Melle]: In order for the inspector to be able to inspect an object of some user-declared datatype, it needs a declaration for it. The declarations for system datatypes are omitted from the loadup (by being marked DONTCOPY and being initialized with INITRECORDS). In order for their instances to be inspectable, they should be added to SYSTEMRECLST by the filepkg command SYSRECORDS, which is syntactically identical to RECORDS. The datatype declaration is actually stripped of comments, subrecords and initialization info before being put out. Putting new DLion Microcode into a Sysout: Load the file SPLICE.DCOM from Sources>. (NCLIP MICROCODEFILE SYSOUTFILE) Copies the entire contents of MICROCODEFILE, a DLion .db file, into SYSOUTFILE, which must be a Lisp sysout. Both files must be random access. WHEREIS, and the WHEREIS database [JonL] [Note: This is out of date. There is now a hashfile whereis database, that works as described in the manual.] If you load in Library>WHEREIS.DCOM, then you can ask _(WHEREIS 'FNS T) Of course, this definition is upwards-compatible, so you can ask for MACROS, RECORDS, PROPS, and VARS in addition to FNS; an arg of NIL there means "any of the above". No other definitions are catalogued on SYSTEM.WHEREISDATA. Actually, this extended WHEREIS is extremely fast; it will typically do a couple dozen GETPROPs, ASSOC's, and FMEMB's. What isn't fast is loading the data (it has to come in by READ right now). As you might expect, my "tailorized" sysout, {Phylum}ABC.SYSOUT, has this data pre-loaded (along with the new WHEREIS too). I haven't looked at the amount of LISTP space taken up by the various FILEDEFS properties, but the file itself is comparable in size to the largest of the system source files. How do we make a new WHEREIS.HASH? Archiving files onto MAXC: pass 1 at how to archive sources to maxc chat to MAXC CONN BLISP MY At this point, you will be inside an Interlisp-10. _ LOAD(FTPFILES) (PULL) Pull will bring over copies of any files that are out of date. The destination directory is determined by looking at the source directory and the value of the variable CORRES. After it brings over the files, do @ARCHIVE *.*;*, @@Don'T delete @@ And when you get done with that, DELVER YY If you get a DIRECTORY FULL error during the FTP phase, you can do the DELVER and try again (if the DELVER got rid of any files.) The problem is that the maxc directory system has a fixed limit on the number of files on a directory and the limit is smaller than 2*(number of files on LispCore>Sources). Installing New Opcodes [Masinter]: [abstracted from messages to Jim desRivieres] The various functions (CALLSCCODE, PRINTCODE, CHANGENAME) know about opcodes via the list \OPCODES. If you want to add some new opcodes (in this case, for 2-LISP), you can just edit the list. The format of \OPCODES is documented, I think, in the function PRINTOPCODES, which is on the file ACODE. Note that if you install new opcodes on the fly you should then reset the variable \OPCODEARRAY to NIL. You can tell if you have installed the opcodes by calling \FINDOP directly. (\FINDOP 'CAR) should return the opcode-description-record for CAR, while (\FINDOP 231Q) should look up opcode 231. If you add opcodes, you should send a message to Charnley, JonL, and vanMelle (and LispCore^) outlining what opcodes you want to reserve. The file OPCODES.DOC (on DOC>) I think has a listing of opcodes too, and if you are reserving a range, that reservation should be documented there too. I think there are enough unused opcodes to accomodate 2/3-LISP. Subject: adding UFNs The UFN mechanism hasn't really been extended for simple experimentation but is workable with a little effort. Normally, UFN entries get set up at MAKEINIT time by a renamed version of a function, I think it is called SETUFNENTRY or some such. (LLCODE, LLBASIC, LLNEW or one of those). The entries in the OPCODES record is used to set up the ufns. Now, it is currently the case that UFN's can't do anything like push N things on the stack -- all they can do is pop N arguments (N>=0) and push 1 result. Writing ufn's that do something other than that, e.g., that don't follow the normal function call paradigm, are a lot more work. Basically I think you have to get into the level of stack-hacking that is found inside LLSTK. For example, a UFN that wanted to push a bunch of NIL's would have to do something awful, like steal space out of its own basic frame to give it back to the caller. This kind of code is tricky to write and debug, especially because you can't do things like insert BREAKs. Popping N off of course is easy since that is what function calls do. In order to do a jump operation, doing something like (add (fetch PC of (\MYALINK)) 10) would do a relative jump to byte +10. It may actually be necessary to extend the UFN mechanism to allow some of the extensions that you want. Why don't you figure out what you can do with the current mechanism, and come back with the ones that you can't figure out how to implement. Adding new opcodes to the Interlisp-D system Written by: Herb Jellinek Revised: 14 June 1984 The process of adding new opcodes to the Interlisp-D system has long been a mysterious one. This document is an attempt to shed some light on these mysteries. The document covers: Creating new opcodes, Writing UFNs, and The OpcodeTool. Enjoy. Creating new opcodes There are a number of global objects and properties that one must know about in order to install new opcodes/UFNs. Here's a list of them: \OPCODES [List] A list of the current opcodes, each of which is a record of type OPCODE. \OPCODEARRAY [Array] An array-ified version of \OPCODES. If set to NIL it will be reinitialized from the contents of \OPCODES. \OPCODEARRAY is recreated when needed by the function \FINDOP. DOPVAL [Property] Information on how to emit code for a given function. There are two formats: 1. (nargs . opcode-sequence) If the number of arguments supplied matches nargs, compile into the sequence opcode-sequence. 2. ( (nargs1 . opcode-sequence1) (nargs2 . opcode-sequence2) ... (nargsN . opcode-sequenceN) . other-cases) If the number of arguments supplied matches nargs1, compile into opcode-sequence1, otherwise see if the number of arguments supplied matches nargs2, etc. One may also supply a function name as the tail of the list; the code generator will call that function if none of the other cases apply. The function OPT.COMPILERERROR is typically used for this purpose; it is equivalent to HELP. DOPCODE [Property] The OPCODE record for a given atom. OPCODE [Record] A record describing the structure of the DOPCODE property and the elements of the list \OPCODES; it has the following fields: UFNFN name of the ufn. Actually read by MAKEINIT OPLAST specifies the last in the range of instructions of which this one is the first. only for opcodes like IVAR LEVADJ stack effect (+/-n) or token. used by PRINTCODE. See code in PRINTCODE for details OPPRINT used only by PRINTCODE. OPNARGS number of extra bytes OPCODENAME name of opcode OP# number of opcode Herb will document OPPRINT. Writing UFNs UFNs are Lisp functions that either run in the place of unimplemented instructions or when the microcode detects a situation that is too complex for it to handle. (This is termed punting.) There are two cases involved in writing UFNs: those for single-byte opcodes, and those for multi-byte opcodes. UFNs for single-byte opcodes For example, assume we have a single-byte opcode called SQRT, which takes a FLOATP as operand and returns its square root. The instruction is designed to punt out to its UFN (named \SQRT) when its operand is of the wrong type, at which time the UFN can either attempt to coerce the operand to a FLOATP or signal an error. \SQRT need be no more than a function of a single argument. UFNs for multi-byte opcodes These UFNs are slightly more complicated, but not much. The difference between single-byte UFNs and multi-byte ones is in the handling of the "extra" (alpha, beta, gamma) byte or bytes. To wit: all multi-byte opcodes that begin with the same byte have the same UFN, and the extra bytes get passed to this UFN in the form of extra arguments. We might have a group of three bit-vector operators (we'll call them BITOP), that all begin with a bytecode of 72Q and vary from 0 to 2 in the second byte. The bytecodes each expect one argument on the stack. The UFN (\BITOP), would probably have the following form: (DEFINEQ (\BITOP (LAMBDA (BITVECTOR OP) (SELECTQ OP (0 (\BITOP.MASK BITVECTOR)) (1 (\BITOP.SHIFT BITVECTOR)) (2 (\BITOP.ROTATE BITVECTOR)) (HELP "\BITOP - illegal operation" OP))))) The OpcodeTool The OpcodeTool is a package that makes it easy to set up and test new opcodes in a running Lisp system. Do (LOAD '{Phylum}Misc>OpcodeTool.dcom) (MAKEOPCODE) to run it. It prompts for the following information: Name of opcode, Number, UFN name, Number of args, Stack effect, OpPrint, DOPVAL. After you've run MAKEOPCODE, you can compile functions that use the new opcode and test them out. Creating New Devices: Date: 15 MAR 84 09:47 PST From: MASINTER.PA Subject: AR for Implementors Manual To: LispSupport cc: Kaplan, LispCore^ Section on making new file devices, How To. VMEM.PURE.STATE [van Melle] When preparing a demo, it is often nice to set things up in such a way that you can push the boot button at any time to instantly restart the sysout, rather than having to go back to some installation utility to reinstall the sysout. VMEM.PURE.STATE is a hack that lets you do this. Basically, while it is on, the page fault handler is altered to write dirty pages beyond the original end of the vmem, thus keeping the original vmem "pure". (VMEM.PURE.STATE FLG) [Function] When FLG is true, enables "pure vmem" as of the next operation that writes out a consistent vmem, e.g., LOGOUT, SAVEVM, or SYSOUT. While in this state, as long as you do not perform another vmem-writing operation (LOGOUT, etc.), you can boot the machine (or slightly more cleanly, call (LOGOUT T)) and be back in the same state as the LOGOUT (or whatever) that initiated the pure image. When FLG is NIL, returns to normal page fault operation. This is usually not too interesting, unless you really do want to LOGOUT, etc and forgo the "checkpoint" you set up. Note, however, that in either case, your virtual memory file is bloated by whatever pages had been written to the end of the vmem file instead of where they belonged. There is a mode in which LOGOUT compresses the pages back to where they belong, but I never got it fully debugged. There is a new MP error in this state: 9316. It means you wrote out so many dirty pages, you ran into the absolute end of the vmem file (8MB) even though you still have plenty of "virtual memory" left. There are two typical modes of operation: (1) Call (VMEM.PURE.STATE T) before calling SYSOUT, thus making a sysout that has the pure feature turned on for anyone running it. (2) Start up a sysout not so made, and then call (PROGN (CLEARW (TTYDISPLAYSTREAM)) (VMEM.PURE.STATE T) (LOGOUT)) to turn the vmem into which this sysout was loaded one with the pure property. System Atoms: These were collected automatically. Feel free to prune this list, or to related atoms under a new heading! The full list is in the file ImplManual.Atoms. Please examine that listing and start documenting. !! used in <> by clisp !!IT !0 TTY edit command !AT !CHECK !DW TTY edit command !E !ED !F !IT !N !NX !SCRATCHLIST !SCRATCHTAIL !UNDO !UNDOFLG !Undo %" # ## Function to access current edit expression #SPELLINGS1 Sizes of spelling lists 'NIL an opcode name 'T an opcode name + in CLISP +- in CLISP , in printout ,! ,, in printout ,,, in printout ,. in BQUOTE ,@ in BQUOTE - in CLISP -- in edit pattern, printout -> in MATCH expressions -IN- as in BREAK A in B. -X- end-of-function in printcode .CALLAFTERPUSHINGNILS. Interpreter opcode hack .PUSHNILS. Interpreter opcode hack /FNCLOSER garbage from Interlisp-10 undoable smash swapped array /FNCLOSERA garbage from Interlisp-10 undoable smash swapped array /FNCLOSERD garbage from Interlisp-10 undoable smash swapped array /FNS undoable functions /push undoable clisp version of PUSH /pushnew undoable clisp version of PUSHNEW 10:3 in NS packet trace printout �NILNIL#TIMESROMAND$ TIMESROMAN�NILNIL TIMESROMAN�NILNIL7 TIMESROMAN TIMESROMAN�dNILNIL HELVETICA TIMESROMANHH�$NILNIL HELVETICA TIMESROMANHH�NILNIL% HELVETICA TIMESROMANHH�NILNIL8 HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL# HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMANHH�NILNIL HELVETICA TIMESROMAN�NILNIL HELVETICA TIMESROMANHH�NILNIL# HELVETICA& HELVETICA HELVETICAHH�NILNIL+ HELVETICAHH�NILNIL HELVETICA TIMESROMANHH�NILNIL TIMESROMAN�NILNIL TIMESROMANNILNIL) HELVETICA TIMESROMANNILNIL� TIMESROMAN NILNIL� TIMESROMAN NILNILc TIMESROMAN NILNIL TIMESROMAN NILNIL$GACHA TIMESROMAN NILNILGACHA TIMESROMAN NILNIL GACHA \ TIMESROMAN GACHA & TIMESROMAN NILNIL� TIMESROMAN NILNILb TIMESROMAN NILNILGACHA TIMESROMAN NILNIL� TIMESROMAN NILNIL TIMESROMANNILNIL HELVETICA TIMESROMAN�NILNIL TIMESROMANNILNIL: TIMESROMAN �NILNIL TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN HELVETICA TIMESROMAN �NILNIL TIMESROMAN GACHA TIMESROMAN GACHA HELVETICA TIMESROMAN �NILNIL TIMESROMAN GACHA TIMESROMAN �NILNIL TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN NILNIL( TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA P TIMESROMAN GACHA TIMESROMAN NILNIL TIMESROMAN GACHA - TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA = TIMESROMAN NILNIL� TIMESROMAN GACHA ( TIMESROMAN TIMESROMAN�NILNIL HELVETICA TIMESROMANNILNILt TIMESROMAN GACHA 2 TIMESROMAN GACHA TIMESROMAN NILNIL( TIMESROMAN GACHA + TIMESROMAN GACHA . TIMESROMAN GACHA TIMESROMAN NILNIL TIMESROMAN �NILNIL TIMESROMAN�NILNIL$ HELVETICA TIMESROMAN�NILNIL HELVETICA TIMESROMANNILNIL� TIMESROMAN $$NILNILi TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN $$NILNIL TIMESROMAN GACHA D TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA % TIMESROMAN GACHA G TIMESROMAN GACHA c TIMESROMAN NILNIL TIMESROMAN TIMESROMAN TIMESROMAN GACHA TIMESROMAN GACHA " TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL) TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN NILNILQ TIMESROMAN GACHA TIMESROMAN GACHA 0 TIMESROMAN GACHA ] TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN �NILNILGACHA HELVETICA GACHA HELVETICA GACHA TIMESROMAN $$NILNIL0 TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN $GACHA TIMESROMAN �NILNIL GACHA HELVETICA GACHA HELVETICA GACHA TIMESROMAN $$NILNIL5 TIMESROMAN GACHA a TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN &GACHA TIMESROMAN GACHA TIMESROMAN NILNIL TIMESROMAN GACHA TIMESROMAN GACHA , TIMESROMAN TIMESROMAN U TIMESROMAN HELVETICA ? TIMESROMAN x�xNILNIL GACHA TIMESROMAN TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNILc TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN GACHA TIMESROMAN TIMESROMAN TIMESROMAN GACHA TIMESROMAN GACHA ~ TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMANJ TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMANj TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN* TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN# TIMESROMAN GACHA i TIMESROMAN GACHA ] TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN TIMESROMAN TIMESROMAN NILNIL GACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL$ TIMESROMAN GACHA L TIMESROMAN NILNIL TIMESROMAN GACHA TIMESROMAN GACHA ] TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN) TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA : TIMESROMAN TIMESROMAN1 TIMESROMAN TIMESROMANC TIMESROMAN GACHA TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL' TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL& TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMANK TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN. TIMESROMAN GACHA TIMESROMANGACHA TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN" TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN NILNILGACHA TIMESROMAN TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN$ TIMESROMAN TIMESROMAN# TIMESROMAN GACHA � TIMESROMAN TIMESROMAN TIMESROMAN NILNILGACHA TIMESROMAN TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN GACHA TIMESROMAN NILNILGACHA TIMESROMAN TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN< TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL( TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN NILNILGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL" TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMANG TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN x�xNILNILGACHA TIMESROMAN�NILNIL TIMESROMAN�NILNIL HELVETICA TIMESROMANNILNIL� TIMESROMAN NILNIL� TIMESROMAN NILNIL� TIMESROMAN NILNIL� TIMESROMAN �NILNIL TIMESROMAN�NILNIL HELVETICA TIMESROMANNILNIL TIMESROMANNILNIL� TIMESROMAN GACHA TIMESROMAN GACHA L TIMESROMAN GACHA TIMESROMAN GACHA & TIMESROMAN GACHA x TIMESROMAN NILNIL, HELVETICA TIMESROMAN NILNIL TIMESROMAN NILNIL TIMESROMAN GACHA TIMESROMAN NILNILGACHA TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMANB TIMESROMAN NILNIL TIMESROMANNILNIL( HELVETICA TIMESROMANNILNIL TIMESROMANNILNILo TIMESROMANNILNIL@ TIMESROMAN�NILNILGACHA NILNIL� TIMESROMANNILNIL� TIMESROMANNILNIL" TIMESROMAN TIMESROMAN�NILNIL TIMESROMAN�NILNIL HELVETICA TIMESROMAN�NILNIL TIMESROMAN�NILNIL) TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL3 TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL� TIMESROMAN�NILNIL# TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL0 TIMESROMAN�NILNIL/ TIMESROMAN�NILNIL TIMESROMAN�NILNIL" HELVETICA TIMESROMAN�NILNIL TIMESROMANNILNIL. TIMESROMAN NILNIL TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA " TIMESROMAN GACHA n TIMESROMAN GACHA ) TIMESROMAN GACHA TIMESROMAN GACHA U TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN NILNIL: TIMESROMAN GACHA TIMESROMAN GACHA 1 TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN NILNIL, TIMESROMAN NILNIL@ TIMESROMAN NILNIL TIMESROMAN NILNIL� TIMESROMAN GACHA ? TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA ' TIMESROMAN GACHA � TIMESROMAN NILNIL TIMESROMAN GACHA � TIMESROMAN GACHA TIMESROMAN NILNILF TIMESROMAN NILNIL~ TIMESROMAN NILNIL� TIMESROMAN NILNIL TIMESROMAN NILNIL, HELVETICA HELVETICANILNIL TIMESROMANNILNIL TIMESROMANNILNIL TIMESROMAN TIMESROMANNILNIL TIMESROMAN TIMESROMANNILNIL� TIMESROMAN �NILNIL TIMESROMAN NILNIL TIMESROMAN TIMESROMANNILNIL� TIMESROMAN ��NILNILGACHA TIMESROMAN $$NILNILA TIMESROMAN GACHA TIMESROMAN ��NILNILGACHA TIMESROMAN $$NILNIL TIMESROMAN GACHA TIMESROMAN GACHA / TIMESROMAN GACHA TIMESROMAN GACHA * TIMESROMAN GACHA TIMESROMAN ��NILNILGACHA TIMESROMAN $$NILNILN TIMESROMAN �0TNILNIL TIMESROMAN <<�N`fNILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN $$NILNIL, TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN �06NILNIL TIMESROMAN <<�N`fNILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN <<�N`fNILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN <<�N`fNILNIL TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN $$NILNIL, TIMESROMAN TIMESROMAN TIMESROMAN TIMESROMAN < TIMESROMAN TIMESROMAN � TIMESROMAN GACHA 9 TIMESROMAN GACHA TIMESROMAN ��NILNILGACHA TIMESROMAN $$NILNIL TIMESROMAN GACHA TIMESROMAN ��NILNILGACHA TIMESROMAN $$NILNIL) TIMESROMAN GACHA ' TIMESROMAN GACHA TIMESROMAN H��NILNILGACHA $ TIMESROMAN GACHA TIMESROMAN H��NILNILGACHA g TIMESROMAN GACHA TIMESROMAN H��NILNILGACHA ' TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN H��NILNILGACHA TIMESROMAN GACHA TIMESROMAN H��NILNILGACHA TIMESROMAN H��NILNIL GACHA TIMESROMAN H��NILNILGACHA TIMESROMAN $$NILNIL TIMESROMANGACHA TIMESROMAN TIMESROMAN NILNIL TIMESROMAN TIMESROMANNILNIL� TIMESROMAN TIMESROMAN s TIMESROMAN NILNIL TIMESROMAN TIMESROMANNILNIL8 TIMESROMAN GACHA TIMESROMAN GACHA d TIMESROMAN GACHA m TIMESROMAN GACHA TIMESROMAN GACHA 7 TIMESROMAN NILNIL TIMESROMAN TIMESROMANNILNIL4 TIMESROMAN GACHA + TIMESROMAN �0H`x�NILNIL TIMESROMAN�0H`x�NILNIL GACHA TIMESROMAN�0H`x�NILNIL GACHA TIMESROMAN�0H`x�NILNILGACHA TIMESROMAN�0H`x�NILNILGACHA TIMESROMAN�0H`x�NILNIL GACHA TIMESROMAN�0H`x�NILNIL!GACHA TIMESROMAN�0H`x�NILNIL"GACHA TIMESROMAN�0H`x�NILNIL/GACHA TIMESROMANNILNIL TIMESROMAN TIMESROMANNILNIL TIMESROMANNILNILl TIMESROMAN �0H`x�NILNIL TIMESROMAN �0H`x�NILNIL/GACHA TIMESROMAN �0H`x�NILNIL GACHA TIMESROMAN NILNIL6 TIMESROMAN NILNILI TIMESROMAN GACHA TIMESROMAN NILNIL TIMESROMAN GACHA G TIMESROMAN �NILNIL TIMESROMAN�NILNIL HELVETICA TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL$ TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL TIMESROMAN�NILNIL, TIMESROMAN�NILNIL TIMESROMAN�NILNIL HELVETICA TIMESROMANNILNIL TIMESROMANNILNIL� TIMESROMAN GACHA � TIMESROMAN �NILNILGACHA TIMESROMANGACHA TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN` TIMESROMAN GACHA TIMESROMAN GACHA TIMESROMAN GACHA V TIMESROMAN GACHA B TIMESROMAN GACHA ' TIMESROMAN GACHA . TIMESROMAN $$NILNIL TIMESROMAN TIMESROMAN TIMESROMAN GACHA m TIMESROMAN GACHA � TIMESROMAN $$NILNILr TIMESROMAN TIMESROMAN NILNIL� TIMESROMAN NILNIL* TIMESROMAN NILNIL TIMESROMAN GACHA TIMESROMAN GACHA R TIMESROMAN NILNIL1 TIMESROMAN $$NILNIL"GACHA TIMESROMAN $$NILNILGACHA TIMESROMAN $$NILNILGACHA TIMESROMAN NILNILO TIMESROMAN �NILNIL TIMESROMAN �NILNIL HELVETICA TIMESROMANNILNIL TIMESROMANNILNIL� TIMESROMANNILNIL TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL.GACHA TIMESROMAN�NILNIL TIMESROMAN�NILNIL#GACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL GACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL GACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL GACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL GACHA TIMESROMAN�NILNIL.GACHA TIMESROMAN�NILNIL"GACHA TIMESROMAN�NILNIL TIMESROMAN�NILNIL@GACHA TIMESROMAN�NILNILAGACHA TIMESROMAN�NILNILAGACHA TIMESROMAN�NILNILGACHA TIMESROMAN�NILNIL TIMESROMAN�NILNIL%GACHA TIMESROMAN�NILNIL+GACHA TIMESROMAN�NILNIL TIMESROMAN�NILNIL!GACHA TIMESROMANbz�