Fri Jan  7 13:50:53 1994  John Salmon  (johns@haggis)

	*  Terminate this ChangeLog.  Future changes to be logged
	in the parent directory.

Wed Dec 29 08:57:39 1993  John Salmon  (johns@sampson)

	* fseekrd.c (fseekrd): #include protos.h and error.h

Mon Dec 27 18:05:36 1993  John Salmon  (johns@lux)

	* fseekrd.c: Bail out with Error in the event of an error.

Sun Oct 10 21:02:27 1993  John Salmon  (johns@sockeye)

	* SDFget.c: Added protos.h.  Silenced a warning.

Tue Jul 13 11:19:20 1993  John Salmon  (johns@lux)

	* libSDF/: changed some formats to silence complaints about
	incorrect argument passing to printf.  These SHOULD all be
	changed to singlWarning.

Thu Jun 10 15:34:01 1993  John Salmon  (johns@haggis)

	* SDF.h, SDFget.c: Added the functions SDFgetint, SDFgetdouble
	SDFgetfloat and SDFgetstring to the library.  In addition, added
	the macros SDFget*OrDie and SDFget*OrDefault to SDF.h.  The macros
	use Error and singlWarning, which can be found in the swutils
	library, or which the user may redefine to taste.

Thu Jun  3 16:32:18 1993  John Salmon  (johns at haggis)

	* Revert back to the i860_utils style of managing the PGI,
	IPSC_XDEV, etc. environment variables.  "The management apologizes
	for the inconvenience."  The reason is that i860_target doesn't
	interact well with Make running inside emacs.

Tue Jun  1 20:51:16 1993  John Salmon  (johns@sockeye)

	*  (../SDF): Changed Make.common so it creates libraries in, e.g.,
	sun4/libSDF.a.  Make the old-style SDF_sun4.a a symbolic link to
	the new file.  Use of <ARCH>/libSDF.a is encouraged.

Tue Jun  1 20:47:04 1993  John Salmon  (johns at haggis)

	* Added (this) ChangeLog to the list of docs copied into .tar files!

Wed May 26 20:02:14 1993  John Salmon  (johns at sampson)

	* Removed the -delta flags from Make.delta.  Use the i860_target
	technology instead.

	* Fixed two bugs related to ascii data.  On line 399 of
	SDF-parse.y and line 646 of SDFfuncs.c.  Recompiled sun4 and delta
	versions.  It should now correctly handle ascii data.

Tue Mar  2 20:26:09 1993  John Salmon  (johns at sampson)

	* Conditionalized the #define YYDEBUG 1 so it won't get linked on
	the delta.  It surely can't do us any good!

	* Added a comment (but no code) to SDF-parse.y that worries about
	what to do when a string constant is too long to fit in a char array.

	* Changed SDFtst.c so it prints the first element of every vector
	in the file it reads.  Just another diagnostic that might show up
	an error...

Wed Feb 10 20:12:15 1993  John Salmon  (johns at sampson)

	* Don't close stdin in SDFissdf!

Wed Dec  9 11:36:53 1992  John Salmon  (johns at haggis)

	* Changed the #ifdef SEEK_SET conditionals in SDF.h.  Now
	the 'preferred' usage is SDF_SEEK_SET, etc., but they are
	almost guaranteed to be the same as SEEK_SET from stdio.h.
	If they aren't, complete and untraceable confusion will surely
	result.  Changed SEEK_SET, etc. in SDFfuncs.c to SDF_SEEK_SET,
	etc. 

Wed Dec  2 17:55:08 1992  John Salmon  (johns at pollux)

	* Added ncube2 support.  This means setting up a Make.ncube2,
	and getting alloca.c (unmodified) from the emacs distribution,
	and saying  'extrasrc:=alloca.c' in Make.ncube2.  Also added
	a call to alloca(0) to SDFopen just after yyparse, which should
	clean up any mess left behind by the C alloca.

Fri Nov 13 18:42:33 1992  John Salmon  (johns at sampson)

	* Reset yylineno in SDFyyprepare().  Otherwise error message
	line numbers accumulate!

Tue Nov 10 20:21:11 1992  John Salmon  (johns at haggis)

	* Changed the typedef for SDF in SDF.h.  It still doesn't tell
	you what's really going on in an SDF hdr, but at least prototypes
	get checked for correctness, and printing an SDF* under gdb might
	even tell you most of what you care to know.

Tue Oct 27 12:24:50 1992  John Salmon  (johns at haggis)

	* Fixed a bug in SDF-lex.l that didn't recognize signed
	integer constants.  How did this last for so long?!  The
	{Sign} specifier was simply missing from the regex.

Fri Oct 16 11:10:13 1992  John Salmon  (johns at haggis)

	* Added halocenterx, halocentery and halocenterz to the
	generic tree.sdfh and tree_ap.sdfh files.  These are used
	by the isothermal halo integrator (i.e., Carl).

Thu Oct 15 21:34:20 1992  John Salmon  (johns at haggis)

	* Try to be even more defensive in SDFclose, in order to not
	get SEGV when bailing out of a parse error.  This is not
	a finished project!  Careful cleanup should be attempted in
	yyerror().

Fri Oct  9 08:34:50 1992  John Salmon  (johns at delilah)

	* SDF now only uses SEEK_CUR to reposition itself in the file.
	fseekrd() will work with SEEK_CUR and any positive offset by
	freading into a junk buffer, regardless of whether the file is
	seekable.  Thus, operations on unseekable files need not be
	sequential.  They must still be "monotonic increasing".

Thu Oct  8 10:06:49 1992  John Salmon  (johns at delilah)

	* Added support for reading from unseekable files.  This means
	keeping track of the "seek pointer" ourselves, and checking
	whether seeking is really necessary to accomodate requests.  If
	the caller doesn't require us to seek, then we don't.  If he does,
	then we fail.

	* Added SDFsetbufsz/SDFgetbufsz, which induce SDF to malloc a
	buffer and call setvbuf in all subsequent SDFopen's.  Use for
	tuning access to tape devices and pipes.

Wed Sep 16 11:12:40 1992  John Salmon  (johns at haggis)

	* Added the keyword "long" to the set of possible types.
	It behaves EXACTLY the same as "int".  There is no corresponding
	SDF_LONG.  Is this really a good idea (the same was done for unsigned).

	* Fixed a bug in SDF-parse.y which caused miscounting the
	lengths of implicit arrays, e.g., char greeting[] = "hello";

	* Added Const to many of the arguments in SDF.h.

Tue Sep 15 17:10:48 1992  John Salmon  (johns at haggis)

	* Eliminated the huge redundancy in SDFrdvecsarr, and made it
	call SDFseekrdvecsarr.  This not only shortens the code, but it
	probably means more reasonable behavior on return from errors.
	Ever wonder what would happen if you do an SDFrdvecsarr and the
	last "name" you give it doesn't exist?  Well it's much better now.

Wed Sep  2 15:12:06 1992  John Salmon  (johns at haggis)

	* Changed the strategy for grabbing temporary space in SDFrdvecsarr
	and SDFseekrdvecsarr.  Now if malloc fails, we try successively
	smaller requests until we get something.  We only give up with
	an error if we can't get enough for a single "record" (i.e.,
	struct).  The memory is freed at the end of the execution of
	SDF*rdvecsarr.

	* In the interests of reducing namespace pollution, changed the enum 
	names in SDF_type_enum to be SDF_INT, SDF_FLOAT, etc.  This will
	break some old codes.  Old codes can be made to compile by
	adding a #define OLD_ENUM_NAMES before #include "SDF.h".

	* Changed the external name of swapn to SDFswapn.  This should
	not effect anything.

Tue Sep  1 10:13:05 1992  John Salmon  (johns at haggis)

	* Added an if( hdr==NULL ) test to each of the exported SDF
	functions.  This should cut down on memory faults from users
	who don't check the return from SDFopen.  It might even be
	worthwhile to put a magic number in the SDF hdr.

Mon Aug 31 17:43:25 1992  John Salmon  (johns at haggis)

	* Fixed a typo in SDF-lex.l and recompiled delta and sun4.

	* Checked the return of realloc against NULL in the middle of
	SDFrdvecsarr and SDFseekrdvecsarr.  The program now fails with
	a message rather than getting a Segv.  It is probably not recoverable.
	We really should reduce the size of some of these buffers by
	staging them.

Tue Jun 30 02:39:30 1992  John Salmon  (johns at haggis)

        * Removed some spurious newlines from the ends of the .sdfh files.
	These make trouble if you try to do "cat tree.sdfh foo.tree > foo.sdf"
     
	* INCOMPATIBLE CHANGE:  changed the calling convention for
	SDFopen.  Now we pass in two char strings, both filenames, one
	for the header and one for the data.  This has a number of
	beneficial effects:
	1) It means we don't have to 'agree' on what kind of FILE is
	in use with the 'user'.  (e.g., CUBIX, NX, SRV, etc.)  All the nasty
	business gets handled inside SDF in the Makefile and such.
	2) It's easier for a 'user' to specify a header that comes from a
	file.  The user doesn't have to stat the file, malloc space, read
	it in and then call SDFopen with the contents as an arg.
	3) SDF is reasonably careful about checking out whether fopen
	returns NULL.  This means less obscure failure when a bad name
	is provided.

	If the hdrname is NULL, or if it is empty or if it strcmp's
	equal to the datafname, then the data file is treated as an
	SDF file with a header attached.

	* Made a similar change to SDFissdf, so that one calls it with
	a name rather than a FILE *.  "-" is stdin.

Tue Jun  9 17:30:13 1992  John Salmon  (johns at delilah)

	* Created SDF_sun4.a SDF_risc6000.a and SDF_delta.a.

	* Added some new functions to SDF.h and SDFfuncs.c:
	SDFissdf, SDFbyteorder, SDFswap, SDFnoswap, SDFisswapping.

	These make it simpler to use generic header files for things
	like pjq, tree, etc. files.  The nbio_ routines are no longer
	the preferred way to read old-style files.  It might be worth
	writing a function that reads an alternative header directly
	from disk, rather than from a string.  It really is a shame
	that there is no portable way to treat a string as a FILE *.

Sat Apr  4 09:34:16 1992  John Salmon  (johns at haggis)

	* Created SDF_risc6000.a.

	* Changed the declaration of do_value_param so it DOES NOT say
	Static int do_value_param(...., const_t const);
	This caused the AIX compiler to fail in a very obscure way!

Thu Mar 26 05:20:04 1992  John Salmon  (johns at kastor)

	* Some minor surgery on SDF-parse.y to deal with zero-length
	blocks at the end of the data file.  We now try to fstat the
	file to figure out how many records will fit in the file.  This
	should make writing generic tree-file readers easier, because
	"npart" is implicit.

Tue Mar 24 09:32:01 1992  John Salmon  (johns at haggis)

	* "Released" v1.2.0.

	* Changed the Makefile, SDFfuncs.c and SDF-parse.y so the
	bison-generated	externals now start with SDFyy, i.e., 
	SDFyyparse, SDFyyerror, etc. This makes it easier to merge 
	SDF with programs like sm.

Thu Mar 19 05:05:12 1992  John Salmon  (johns at haggis)

	* Removed #pragma once from the rest of the .h files

	* Changed malloc to calloc in SDFopen.	

Fri Mar 13 08:24:32 1992  John Salmon  (johns at haggis)

	* Rebuilt SDF_delta.a and SDF_sun4.a

	* Removed #pragma once from all the .h files.

	* Added dependency for SDF-lex.c and SDF-parse.o to Makefile
	so changes to SDF-lex.l are understood by make.

	* Added the 'unsigned' keyword to SDF-lex.l.  It's a total
	no-op.  It may make it somewhat easier (and dangerous?) to
	convert C structs into SDF headers.

	* Changed the definition of the end-of-header symbol.
	Now it's any comment line containing the string SDF-EOH.
	The newline is the last character in the header.  It is
	highly recommended that users add their own form-feeds
	to this line, but it is now at the user's discretion.

Thu Mar 12 22:23:37 1992  Mike Warren  (msw at sampson)

	* (nbio_open) Added a break to the case NBIO_SDF_TYPE:

Sun Feb  9 13:26:20 1992  John Salmon  (johns at haggis)

	* Added some more error reporting to nbio.  On most errors,
	something is now written into nbio_errstring.  If the error
	is detected by SDF, then we (usually) copy SDFerrstring too. 

	* Added the SDFhasname() and nbio_hasname() functions to test 
	for the existence of a name in a file.

Fri Feb  7 18:32:04 1992  John Salmon  (johns at haggis)

        * Added missing ';' in the "parameter byteorder=blaa; "
	strings in nbio.c

        * Added the char nbio_errstring[] array for messages.

	* Added the char *nbio_type_names[] array.

	* Added the "seekrd" functions to SDF and nbio, which use the
	fseekrd(...) "primitive".  Hopefully, this will be supported
	by servix on all future parallel machines.

	* Fiddled around with the header files.  There's now only
	one copy of the prototypes in the "public" files, nbio.h and
	SDF.h.