========================
                  Revision history of Tela
                  ========================

-----------------------------------
Version 2.02 (Released 20 May 2005)
-----------------------------------

USER-VISIBLE CHANGES

- incorporated 2005 Tsyganenko geopack.f model and added T96_NOFAC field model option
  (the T96 model without field-aligned currents, opt/geopack.f, opt/T96.f, opt/geopackintf.C;
  affects functions geopack_model, geopack_coordtrans, geopack_trace)

INTERNAL CHANGES

- two formal changes to allow compilation under g++-3.4.0 (I think it's g++ bug or strange feature at least,
  but the fix is not harmful)

BUG FIXES

- functions fprintf/printf were not working at all (fileio.ct). Now fixed by rewriting fileio.ct
  to use system snprintf. Gcc extensions are not needed using this method. Hope snprintf is
  found on all systems...
- format,sformat and fformat with array argument inserted erroneously null character after each
  number written; especially seroius this effect was in sformat which broke the string after the first
  outputted number. Fix was to remove line oss << ends; in object.C:OutputComplexNumber when
  USE_STRINGSTREAM is defined as 1. The ends inserts the null character, in this way <string>
  and the old <strstream.h> behaviour differs rather lot.
- "make clean" deleted flex++-3.0.2/parse.c etc., which causes trouble if bison++ does not exist
  in user's system and user types "make clean" in main tela directory; now "make clean" only
  removes flex++-3.0.2/*.o and "make veryclean" removes also the other regeneratable files.

--------------------------------------
Version 2.01 (Released 23 August 2004)
--------------------------------------

USER-VISIBLE CHANGES

- added function accum (numerics.ct). (Actually the function has been there for some time,
  but the behavior was changed slightly, and it is now "official").
- added function Lomb (numerics.ct)
- renewed Tsyganenko model support and GEOPACK coordinate transform routines
  (everything under opt/, new file geopack.t, autosources now defined in telainit.t)
  More general and powerful functions geopack_coordtrans, geopack_model and geopack_trace
  are now available, using the latest GEOPACK Fortran version (2003). The old functions
  (e.g., geotogsmXYZ) are now defined in geopack.t and call the general functions.
  Supported field models are T89C and T96.

------------------------------------
Version 2.0 (Released 9 August 2002)
------------------------------------

USER-VISIBLE CHANGES

- it is now demanded that if a t-function is defined with obligatory output arguments
  (you do this by placing the semicolon inside the output arg list and enclosing the list
  in brackets, as in function [x;] = f() {x=1}), the function must be called with the 
  bracketed notation, like [y] = f(), not simply f(). If you do, version 2.0 generates
  a runtime warning message "Calling function with obligatory output arg(s) with non-
  bracketed call form". Motivation: see the KNOWN BUGS section of Version 1.35 below.
  This fixes the bug, if the programmer also obeys the rule that if an output arg is defined
  as obligatory by using the semicolon as above, he avoids reading the variable in
  the function body before assigning a value to it. To check this automatically would
  be hard in the general case (would require data flow analysis). The current version (2.0)
  does not attempt to do even simple checking of this.
  Summary: Existing code should work as before, but may give the above warning messages.
  To remove the warnings, change the call forms to use the bracketed output arg notation
  for functions having obligatory (i.e., input/output) output args. After you have done this
  it is in principle possible that the results of your program change, but towards being
  more correct than in the past:-)
- telainit.t now checks that envvar HOME is defined before attempting to source $HOME/.telainit.t
  (thanks to Mikko Syrjäsuo for noticing this, mikko at phys.ucalgary.ca).

KNOWN BUGS

- at least with HDF4.1r3 library version, load()ing of new-style HDF file crashes
  A fix will be provided soon...

-----------------------------------
Version 1.35 (Released 5 July 2002)
-----------------------------------

USER-VISIBLE CHANGES

- added many special functions (specfun.C,specfun.H,ode.C,ode.H,specfun_ct.ct,specfun.t,telainit.t).
  Functions implemented now include Euler gamma, incomplete gamma, incomplete beta, erf, erfc,
  Gauss' and Kummer's hypergeometric functions, Chebyshev, Legendre, Laguerre, Bessel.
  The functions work with arbitrary complex arguments and are also automatically threaded
  over array arguments. Bugs probably remain, although systematic verification
  with Mathematica was done.
- aniopen/aniput/aniclose can now write also MNG animation in addition to MPEG
  (ani.ct,png.C,png.H,image.C,image.H)
- added function poissonrand (numerics.ct)
- added function fwrite (files.ct)
- added Emacs Tela mode to distribution (tela.el, contributed by Mikael Hedin, First.Last@irf.se)
  (Also available under http://www.geo.fmi.fi/prog/tela.html/)
- somewhat improved Matlab file importing (reads 32-bit reals also, contributed by Micce Hedin)
  (Still, doesn't read Version5 matlab files, but this is in the plans...)
- functions import_PBM and export_PBM renamed the more logical import_PNM and export_PNM.
  Old names are still aliases to the new functions, but it is recommended that you
  use the new names in new code from now on (files.ct,telainit.t).

KNOWN BUGS

- using the output argument also for input (i.e. an input/output argument) and binding
  it to indexed expression in caller is legal, but may produce incorrect results. Example:
    function [a;] = f() {a+=1}; x=ones(2); x[1] = f(); disp(x);
  produces runtime error "Unary '++' applied on <Undefined>" rather than giving the
  expected output #(2, 1).
  To avoid problems, use the following rule: if function argument is both read and written,
  always bind it to simple variable in caller. The above example rewritten using this rule
  would be:
    function [a;] = f() {a+=1}; x=ones(2); tmp=x[1]; [tmp] = f(); x[1]=tmp; disp(x);
  I hope to somehow fix this in the future (for version 2.0).
  Thanks to Neilen Marais for reporting the bug (nmarais at hertz.ee.sun.ac.za).

BUG FIXES

- 2^(-1:1) produced core dump (i.e. case when integer was raised to integer
  vector power, when the integer vector contains negative elements) (objartihm.C)
- min() and max() of an empty array produced a very large number,
  nowadays they produce an empty vector (objarithm.C, Bugfix by Mikael Hedin)
- function gsetogsmXYZ didn't work correctly (opt/geopack_ct.ct)

INTERNAL CHANGES

- now uses FFTW library (www.fftw.org), which increased FFT speed by factor of two.
  Thus far the sine/cosine transforms still use the old FFTPACK routines, however (fftw.ct,fft.ct).
- some changes, hopefully improvements, in configure script (HDF libraries and includes etc) (configure.in)
- now uses most recent version of autoconf (2.53)
- updated to using newest version of opt/geopack.f (dated 7 Aug 2001)

-----------------------------------
Version 1.34 (Released 17 Sep 2001)
-----------------------------------

BUG FIXES

- The function whos() didn't work in the previous release,
  for which reason there is this early bug fix release now.

-----------------------------------
Version 1.33 (Released 4 Sep 2001)
-----------------------------------

BUG FIXES

- Since introduction of KShallowObjectArray in 1.32, I forgot to increase the constant
  12 to 13 in #define pair(ka,kb) ... in objarithm.C. Result: E.g., 2/x where x is undefined
  did not produce error message but silently evaluated to 0. There could be other much
  more obscure errors also. Now the constant 12 is no longer hard-coded but is defined to be
  int(Kundef)+1. This works as long as Kundef remains the last in enum Tkind (object.H).
  There is comment telling it...
- in tela.C:FindMachineNan(), using switch() statement can cause compile-time error due to
  duplicate case labels, fixed now by rewriting switch() as nested if's (tela.C)
  (This is in code segment that was put in tela only in the previous release.)
- some changes to allow compilation with gcc-3.0 (contributed by Ch. Spiel)

-----------------------------------
Version 1.32 (Released 24 Apr 2001)
-----------------------------------

BUG FIXES

- t-functions using ellipsis argument (...) crashed Tela (this was due to the 
  "deep" copy object array feature introduced in previous version 1.31, cure
  was to introduce another internal type KShallowObjectArray which is now used
  in Tprg::execute() in prg.C)
- memory leak in object arrays (object.C:Tobject::setsize, deep clear was missing)
  (also this bug existed only in version 1.31, not before, because object array
  semantics was changed from shallow copy to deep copy in 1.31).

USER-VISIBLE CHANGES

- new diag() function contributed by Ch.Spiel which allows a second input argument
  specifying the super- or sub-diagonal (the default zero meaning the main diagonal),
  this function is compatible with Matlab's and also compatible with the old diag()
  function (std.ct)
- added global symbol NaN if compiler supports NaN's (ieee754.h header is present and working) (tela.C)
- some enhancements to telapath() (std.ct; contributed by Ch.Spiel)

----------------------------------
Version 1.31 (Released 8 Mar 2001)
----------------------------------

USER-VISIBLE CHANGES

- copying an object array (see help voids) is now "deep" copy rather than "shallow" copy.
  This means that when an object array is assigned to another variable, a completely new
  copy is made (before, only the first array level was copied). Mostly this won't affect
  you at all because (1) you are using object arrays only if you are using voids() in some
  phase in your program, (2) if you are, then you are probably using only single-level
  object array. The old behaviour was simply stupid. Object arrays are a crude but working
  mechanism for using heterogeneous combined types in Tela; the only relevant function
  (in the present Tela version) is voids(), which acts as a constructor for such things.
  Thanks to Ch. Spiel for suggesting this change (object.C).
- quit() and exit() now accept input argument (the exit value) (std.ct)
- sort() function now accepts second input argument (user-defined comparison function) (std.ct)
  (old function still available as sort_old in case of problems).
- slight change in semantics of --shared option for telakka (telakka.in)
The above were contributed by Christoph Spiel <cspiel at hammersmith-consulting.com>

- added incomplete gamma and beta functions (specfun.t)
  (specfun.t not autoloaded, but after source("specfun.t") functions are available).
- added function polymul and added polyadd,polymul to autoload (poly.t,telainit.t)
- added function applyfilter (numerics.ct)
- added functionwise profiling option (tela -p) (tela.C,prg.C)

BUG FIXES

- doing dot product (**) of two zero-length integer vectors produced a LAPACK error message
  and Tela quit (la.ct)
- tree.C:DeleteObjects() accessed memory after it being freed, which however did not cause
  any problems in practice (unless using malloc debug libraries), but is anyway now fixed

INTERNAL CHANGES

- Some modernizations in configure script (e.g., now tries to link with
  libatlas by default instead of libblas if available) (configure.in)

-----------------------------------
Version 1.30 (Released 5 Dec 2000)
-----------------------------------

USER-VISIBLE CHANGES

- wrote ani.ct for writing MPEG-1/MPEG-2 animations
  in a memory-efficient way (aniopen,aniput,aniclose)
- added function bsearch for binary search (std.ct)
- added function interp (std.t)
- Introduced some flatcode optimization, to use, invoke with tela -O1 ...
  The optimization is loop tiling and it may speed up some codes that
  uses long vectors with lots of intermediate results that are reused.
  At least it speeds up third-order interp() by factor 1.5 .. 2.
  (new source file: optimize.C)
- added function read_CDF for reading variables from a CDF-file as
  values, not into global variables as import_CDF does (CDF.ct)
- export_CDF now creates 1- or 2-byte unsigned integer files if the int array
  elements are in the range 0..255 or 0..65535, respectively (previously, only
  signed types were specially handled)
- export_PBM is now able to write 16-bit grayscale images (in ASCII mode) (files.ct)
- publicly released SPPC-1.0 plot package (ftp://ftp.funet.fi/pub/sci/math/tela/needed/sppc-1.0.tar.gz).
  The interface is in sppc.ct.

BUG FIXES

- fitlinear didn't work correctly when 4th arg (sigma) was given (fit.t)
- import_CDF now correctly transposes arrays in Fortran-ordered CDF-files (CDF.ct)
- import_CDF now correctly treats also strings and string matrices (CDF.ct)

-----------------------------------
Version 1.29 (Released Nov 2 1999)
-----------------------------------

USER-VISIBLE CHANGES

- added C-tela file sppc.ct as an interface to Simple Panel Plot Composer
  (a program just written and hopefully distributed soon separately)
- added command line option --noreadline (-n), use it if reading
  from a pipe or file, like tela -n <file, this bypasses the GNU-Readline
  library (tela.C)
- added function LinearProgramming (new t-file lpmin.t)
- added function gaussrand (std.ct)
- added function putenv (std.ct)
- added function export_CDF (CDF.ct)
- minor bug fix in import_CDF (CDF.ct)
- made atan2 work for (real) array arguments (std.ct)
- blasbench.t, bench.t modernized
- HDFNewMode(1) is now the default (files.ct)
  I.e., it no longer writes the HDF3.1r5-compatible files by default.

BUG FIXES

- import_PBM can now read also 1-bit images, binary of ASCII (files.ct)
- pcoloring or contouring large matrices using xgrid,ygrid options
  confused plotmtv, fixed this by writing them in binary (plotmtv.ct)
- failed fopen calls consumed one file descriptor, yielding to running out of
  them after 100 failed opens (fileio.ct).
- import1 hung if called with empty filename (files.ct)
- save("filename.hdf") produced incorrect HDF file if VOID variables were
  in workspace (These files can now be nevertheless load()'ed, warnings
  are given though).
- new-style HDF: string matrices are now saved properly with the string flag set
- made it to work with HDF4.1r3 (files.ct)

KNOWN BUGS

- increasing MAXRANK from 4 (def.H) leads to core dump in telainit.t

-----------------------------------
Version 1.28 (Released 11 Sep 1998)
-----------------------------------

USER-VISIBLE CHANGES

- added function fread for fast binary reading of files (fileio.ct)
  (Anyone needing fwrite out there..?)
- added functions set_alt_display and reset_alt_display to temporarily direct graphics to another
  X11 display (plotmtv.ct). Previously this was possible only for an entire Tela process by
  setting DISPLAY environment variable, but now it can be done from within Tela also.
- added -display option to plotmtv (works together with previous one)
- import_PBM and export_PBM can now read/write pipes by using the notation "! ..."
- exporting and importing of v4-Matlab MAT-files now works in little-endian 386-Linux also
  (Internals: It should work on DEC-Alpha also but not tested. WE_ARE_LITTLE_ENDIAN should be
   defined as 1 on little-endian machines in files.ct.)

BUG FIXES

- import_PBM could not read P2,3,5,6 files where maxcol was on the same
  line as w,h (files.ct); we hear that these are produced by IDL for instance.
- If Tela was exited using quit() or exit() (instead of typing Control-D),
  open figure windows were not closed (std.ct)
- inserted call to exitfunction() of std.ct at end of main() (tela.C) to prevent core dumps on pressing control-D to exit
  (for now, this is done only for GNUC, since the problem has thus far arised only on Linux).

INTERNAL CHANGES/"FEATURE" FIXES

- Include -ldl now by default for Linux (dynamic linking library, configure.in)
- For Linux, define MAIN__ in tela.C (required by some versions of libf2c.so)
- Some syntactic changes to better compile withi newer Linux systems etc.
- Incorporated flex++-3.0.2 under source tree to avoid name conflicts with other programs
  also named flex++ but which are incompatible

-----------------------------------
Version 1.27 (Released 4 July 1997)
-----------------------------------

BUG FIXES

- export_PBM made core dump if file was unwritable, now gives nonfatal error message (files.ct)

USER-VISIBLE CHANGES

- parser now complains if there is some garbage between last legal token and EOF (d.l,d.y,tela.C)
  (This is done by effectively wrapping the input in {...}.) Thanks to Gary Holt <holt at monet.klab.caltech.edu>
  and Olaf Amm <amm at geophys.nat.tu-bs.de>. For example, if there is a semicolon missing between
  function definitions in a t-file, old Tela silently ignored the rest of the file; now it complains.
- "calling" a numerical object previously printed its value, now it is forbidden. For example,
  c = 3; c(1) before printed "3" and then "<Undefined>". Now it complains: "3 (integer) is not a function".
  (The reason why it outputted <Undefined> before is a bit obscure, but who cares any more...)

-----------------------------------
Version 1.26 (Released 6 June 1997)
-----------------------------------

USER-VISIBLE CHANGES

- whos() does not use $PAGER when in batch mode (std.ct)
- whos() tells about string-flag also on rank>1 int arrays (std.ct)
- whos() now explores environment variable TELA_SORTER (default "sort -b -f -k 1") to
  find the Unix program to sort the variables. On some Unix systems these options might
  not work in which case it is necessary to set up TELA_SORTER to fix the problem (std.ct)
- added function import_CDF (new file CDF.ct)
  This should work if you have CDF library and cdf.h.
  Notice that CDF (Common Data Format) is different from netCDF and HDF,
  which Tela has been able to do for rather long time. The CDF format is mainly used by NASA.
- added function strstr (std.t)

BUG FIXES

- *** SERIOUS *** a^b where a,b were ints and b was negative produced incorrect zero result (objarithm.C)
- Plotting data which has large offset was inaccurate because only 6 digits
  were passed internally to PlotMTV [for example, plot(1e4+sin(0:0.1:4)) produced
  "quantized" result] (plotmtv.ct)
- Installation/configure: Configure script failed to detect -lreadline library
  even if one was present. Now uses GNU autoconf-2.10
- The help command did not work on all systems for t-files (For example, typing
  "help stddev" resulting in message "No comment string"). Fix: Replaced "more" with "tail" (telahelp.sh).
- The autoglobal function returned a <Undefined> value, although it should
  return no value (std.ct)

INTERNAL CHANGES

- C++ ANSI standard says that in for (int i=0;i<10; i++) stmt; the scope of i is limited to stmt.
  Almost all Tela source files contained at least one instance of this problem. These have been
  fixed by taking the declarations out of for statements, thus it can be compiled on both
  old and new c++ compilers.
- function where() is now C-tela function in std.ct, not t-function in std.t
  as it was previously (std.ct,std.t,telainit.t). This should not be a user-visible change,
  except that the function is faster.
- some formal changes for the new Cray C++ compiler (2.0). The new compiler is still buggy,
  for it cannot compile objarithm.C without crashing (!).
- global::nops is now double (previously was int), unless SLOW_FLOATING_POINT_MATH is defined as 1.
  This makes it possible for Ct-modules to record FLOP counts exceeding 2 billion for one call (common.H, tela.C)
- #define exit(x) _exit(x) (def.H). On some machines Tela sometimes makes a core dump when exiting.
  This hack prevents global destructors from being called; _exit() is the low-level C-library function
  whereas exit() knows tries to call C++ global destructors first.

-----------------------------------
Version 1.25 (Released 4 July 1996)
-----------------------------------

USER-VISIBLE CHANGES

- added function meminuse (std.ct)
- whos now gives sorted output with pager ($PAGER envvar, or more) (std.ct)
- fopen now recognizes also "a" flag for appending (previously only read ("r") and write ("w")) (fileio.ct)
- import1 can now read ASCII matrices which have '#' comment lines in between data lines
  (previously only comment lines at the beginning of the file were accepted). However this feature
  does not work for files containing only one number per line (vector files). (files.ct)

BUG FIXES

- Sometimes 'Runtime error at line ...' reported line number zero. No more. (prg.C)
- import("file.hdf","app") did not append "app" to the names if "file.hdf" was
  new-style HDF file (files.ct)
- izeros, zeros, rzeros, czeros now complain if the dimension is
  negative (previously crashed) (prg.C)
- wrote a workaround for an operating system memory managament bug on Silicon Graphics machines.
  This involved writing a completely new memory management package (newdel.C), which is used
  instead of the standard new/delete operators on these machines.
- fixed some bugs in importing new-style HDF SDS and netCDF files written by program other than Tela (files.ct)
  But the netCDF stuff may still not work completely...
- function minimization t-file: some bugs fixed (fmin.t)
- further optimization (traverse only nonequal-sized blocks in Freelist) of the memory management patch (newdel.C)
  (It is now fast enough so it may be worthwile to use it on other platforms too.
   The algorithm is now the 'bestfit' method with sorted Freelist.)

-------------------------------------
Version 1.24 (Released 19 March 1996)
-------------------------------------

USER-VISIBLE CHANGES

- import can now read Matlab integer matrices (previously only floating point)
- closefig("all") now closes all figures
- added function clear (std.ct)

BUG FIXES

- sum,prod,mapmin,mapmax produced garbage result if called
  with vector first argument and present second argument
  (equal to 1). For example:
  >size(sum(0:5,1))
  #()		// second argument present ==> wrong
  >size(sum(0:5))
  1			// sum called with only one argument ==> ok

  However the printed value was correct:
  >sum(0:5,1)
  15

  The object produced was damaged but indexing etc. nevertheless
  produced the correct numerical value.
  
  If the second argument was not present the result was always
  correct.

  Now these all work correctly. The bug was in std.ct.

- (PlotMTV) Fixed bug when saving GIF file from 24-bit visual
  (previously worked only on some hardwares)

KNOWN BUGS

- "save" may not work if the saved variable is a degenerate array
- there is disk space overhead with HDF files if large one-dimensional
  arrays are saved, if HDF version > 3.1r5 is used. This is actually a
  bug in HDF and there is not much Tela can do about it. The overhead
  is four bytes per data element for one-dimensional arrays. For multidimensional
  arrays it is four bytes times the sum of the sizes of the dimensions,
  which is negligible if the arrays are roughly square.

------------------------------------
Version 1.23 (Released 20 Oct. 1995)
------------------------------------

USER-VISIBLE CHANGES

- added function fparse (fileio.ct)
- Made link function to work on Iris ELF dynamic shared objects.
  Probably works also on other machines with elf. See FAQ list for more.
- added --shared option to telakka to ease DSO creation (see above)
- added function stencil2d_4 for 2D five-point molecule (numerics.ct)
- added function axpy (BLAS-1 style function) (la.ct)
- added functions NewHDFSupported, HDFNewMode (files.ct)
- changed behavior of save: do not save variables hidden in packages
  (previously these were saved with names starting with '$') (files.ct)
- added function export_netCDF, which is operational only if Tela was
  compiled using HDF4.0b1 or later (see also below; files.ct)
- added function colormap (plotmtv.ct). You need the latest (1.4.2t) PlotMTV
  version to use this feature.
- added possibility to use nonuniform grid in contour/pcolor/mesh plots
  by using the "xgrid", "ygrid" options with real vector.

INTERNAL CHANGES

- made SetJmp into a macro (error.H, error.C, tela.C)
  On some non-GCC compilers the old version might have caused
  some crashes after error.
- made limit a builtin function (previously was in std.t) (std.ct)
- Now uses HDF4.0b1 by default. It is possible to link with HDF3.1r5
  by defining OLD_HDF_VERSION when compiling files.ct, however.
  Versions 3.2 and 3.3 are known NOT to work.
  Pros and cons of old and new versions:
  3.1r5: + Tela executable size will be smaller
         - Real number data lose precision since they are saved as 32-bit
           numbers
  4.0b1: - Executable will be larger (approx. 400 kB on Linux)
         + Unidata netCDF files become possible to load, import and export
         + No precision loss occurs if HDFNewMode(1) is called
           before saving (the HDF files will also be about twice
           as big).
- obj.SetToVoid() now also deallocates previous object by calling clear().
  This should be always safe and may remove some memory leaks, but is
  probably insignificant (object.H)

BUG FIXES

- Assignment b[ind] = a (a,b,ind all vectors) where length(ind)!=length(a)
  was silently accepted. No more!
- Matrix product (inner product) of two integer vectors produced
  garbage result (la.ct; added  && A.kind()!=KIntArray above "Faster branch")
- More than one goto to the same label did not work (codegen.C)
- It sometimes crashed when it tried to give a warning of undefined labels (codegen.C)
- MAXFILES was 5 before, now increased to 100 (fileio.ct) (Was this a bug or not?)
- Cray C++ compiler produced incorrect code for range operation (e.g.,
  0:1.5 returned #(0,1) not #(0,1,2) as it should). This is now fixed
  by small source code modification in gatscat.C. It is now also fixed in the
  Cray C++ compiler (!)
- Likewise, the same bug in round() function on Cray!!! (Also fixed in the compiler I guess).
- A small bug in perf() self-overhead removal (std.ct)
- mapmin,mapmax and sometimes map produced unpredictable results when the array
  was degenerate (0-length), now they produce a zero-length int vector in such cases
- In some rare situation save function crashed because DFSDsetdims was not
  called. This should not be strictly necessary as the corresponding parameters
  are passed in DFSDputdata but nevertheless it is now fixed (files.ct:HDFAdd)
- The syntax error messages reported a wrong source line number if the file contained
  lines that ended with '/' within multi-line /* ... */ comments (d.l)

KNOWN BUGS

- Xmosaic may not be able to follow all the cross-references in docview() command,
  but Lynx and Netscape can.
  
---------------------------------
Version 1.22 (Released 25.5.1995)
---------------------------------

This version was delayed by a self-caused bug in the FFT
code of the development version. New features include simple
functions for numerical integration, function minimization,
root finding, linear and nonlinear fitting, goto statement,
and support for image processing with XV.

USER-VISIBLE CHANGES

- added label and goto statements (d.l, d.y, tree.H, tree.C, codegen.C)
- added function DebugQueryMode. When debug query mode is on, pressing
  Control-C causes a question to be posed, enabling one to interrupt
  in the usual way or enter debug level (std.ct, tela.C)
- added functions export_PBM, import_PBM (files.ct)
- import() can now read also a raster-8 dataset from HDF file,
  if no SDSs are found (files.ct)
- added function fmin to do function minimization (fmin.t)
- added function fsolve for root finding (fsolve.t)
- added function integrate to do one-dimensional numerical integrals
  over finite intervals with no singularities (integrate.t)
- added functions fitline and fitlinear for fitting data to straight
  line and to linear combination of arbitrary basis functions (fit.t)
- added function fitnl for Levenberg-Marquardt nonlinear fitting (fit.t)
- added function diff (diff.t)
- added function xv to display matrix in XV window (xv.t)
- export_matlab() and save() do not save hidden variables
  unless explicitly named (files.ct)
- added command line options --execute and -e for executing
  strings as Tela expressions on startup
- added optional second argument for import (files.ct)
- generalized map so that the mapped function may return also vectors
  instead of scalars, this allows e.g. sort function to be used with
  map (std.ct)
- added function flip to reverse arrays (std.ct)
- added optional argument (prompt) for input_string() (std.ct)
- cd("") and cd("~/my/dir") now work (std.ct)
- hold(-1) can be used to reset the internal counter (plotmtv.ct)
- eval("!...") no longer works, use system("...") instead.
  The !... notation still works on top level command line.
  Special macro (?, help, !) processing is now done only top level
  command lines.

INTERNAL CHANGES

- Cray C90 support, the macros VECTORIZED and NOVECTOR as well as 
  #ifdef VECTOR_MACHINE ... #endif appear in several places
- now uses MAXFILENAME which is always at least 1024 instead of FILENAME_MAX,
  which was on some system as small as 14, which surely would have caused problems
- some reorganization on def.H
- Makefile no longer assumes the INSTALL works for more than one source argument
- The BROKEN_UCHAR_OUTPUTTER recognition in configure was buggy before
- The object copy operation is now inlined differently (object.H, object.C).
  The new way should be faster (but is it?).

BUG FIXES

- import() from Matlab binary file now honors the string flag
  for string matrices (files.ct)
- save() marks string matrices as such (files.ct)
- the Tsymbol::ishidden member was not initialized in symbol.H,
  therefore the 'hidden' attribute was sometimes on initially (symbol.H)
- the parser failed to count return characters inside strings, resulting
  in slightly wrong line numbers being displayed in error messages 
  if strings contained return characters (d.l)
- Using #() in an expression sometimes resulted in internal error in
  code generator (d.y)
- fixed a small bug in Ct-file line number tracking (ctpp.C)

EXPERIMENTAL FEATURE WORTH MENTIONING

- If Tela receives USR1 signals it breaks execution and goes to debug level.
  Anyone knows how to generate USR1 signals from e.g. Control-G keypresses?

----------------------------------
Version 1.21 (Released 24.2.1995)
----------------------------------

*** IMPORTANT: For the first time, two FATAL bugs were discovered.
    See below.
*** There is now User's Guide in doc/ and in WWW page.

USER-VISIBLE CHANGES

- A new PlotMTV version (1.4.1) is now shipped with Tela
- Added function figure for realtime applications (plotmtv.ct, PlotMTV).
- PlotMTV can now save the figure as a GIF file.
- Added function pixmap (plotmtv.ct). This controls the -nopixmap option to
  PlotMTV. The default is pixmap(on) as before. If your X server is configured
  to use backing store, it may be advantageous to turn pixmap(off).
- pause(n) sleeps n seconds. pause() waits for a key as before.
- Indexing scalars is now permitted, when all indices are either 1,
  VOID, or #(1) For example, if a = 3.14, a[1,#(1),:] is 3.14 but
  a[#(1,1)] is not allowed (gatscat.C).
- C-tela functions now output their names, for example if you type 'sin',
  the system replies 'sin', not '<C-tela function>' as before (object.C).
  If you do the assignment fn=sin, 'fn' will still output 'sin'.

Minor points:
- Some functions in std.ct expecting a string argument no longer work for a
  character argument. Very unlikely that this causes problems to anyone.
- Removed some obsolete functions in std.ct: echo, dump
- Added rough MFLOPS counting in FFT routines.

INTERNAL CHANGES

- Upgraded to autoconf2.1. Numerous changes in configure script. Maybe some bugs?
- Cleaned up std.ct somewhat.
- SIGINT (Control-C) is now handled in a way which should not leave too much
  garbage in memory (prg.C, tela.C). TODO: Rethink macro CATCH_INTERRUPTS.
- Translated FFTPACK in C. Wrote a vectorized version also (vecfftpack.C).
  These source files can also be used as standalone FFT packages for
  C programmers. The functionality is the same as for FFTPACK and the performance
  may even be slightly better.

FATAL BUGS (incorrect results from correct programs)

- Fatal and very nasty bugs in codegen.C were fixed. The variable unique_index
  was not always zeroed where it should. Statements unique_index = 0; were added
  in three places, these are marked with /* Added 30.1.1995 */.
  In addition to fixing the bugs, this might also make the code more efficient
  in some cases. This bug is the worst one I have ever fixed in Tela. Precise
  conditions are difficult to give. To see the simplest example I found,
  see the file "bug1.t" in the distribution; that file can also be used to
  test the Tela version you are using against this bug.
- #(x;y) worked incorrectly if any of the x,y,... was complex (gatscat.C).
  Might also have affected #(x,y,...) notations.
  (The reason was that variable cptr was not incremented in case KComplexArray
  in function MakeArray in gatscat.C.)

ORDINARY BUGS

* "Ordinary" bugs produce core dumps on incorrect programs, incorrect
  behavior in incorrect programs, or correct behavior on incorrect programs.

- Inputting "a[]" for any a produced core dump, now produces syntax error (d.y).
- Last instruction in a function contributed megaflops rates twice (prg.C).
- Doing fn=sin; fn(3) resulted in core dump because the Cfunc.infoptr member
  was not copied by Tobject::init() (object.C)
- sum() and prod() on degenerate input arrays resulted in core dump (std.ct)

----------------------------------
Version 1.2 (Released 18.1.1995)
----------------------------------

USER-VISIBLE CHANGES

- Improved helpgiving for t-functions. Added telahelp.sh shell script
  to do the pattern matching. Now help for t-functions works almost
  similarly as help for C-tela functions. (std.ct, telahelp.sh)
- Modified function sort: return index vector if 2 output args,
  sort complex vectors by real parts, disallow inplace sorting (std.ct).
- Added small utility functions limit and nrange (std.t)
- Added functions mapmin, mapmax (std.ct)
- Added functions cumsum, cumprod (std.ct)
- prod(<integer array>) now returns real result if overflow would
  result in integer multiplication (std.ct)
- Added optional second input arg to sum, prod, mapmin, mapmax;
  like sum(A,1): sum only over first dimension (std.ct)
- Added optional second output argument to return min/max position
  to min, max, mapmin, mapmax (intrinsic.C, std.ct)
- mean and stddev also support second argument (std.t)
- zeros, izeros, czeros now support integer vector argument,
  so you can use zeros(size(a)) etc. Also improved error checking
  in these opcodes (prg.C)
- Added function export_RIS8 (files.ct)
- Added function remove (fileio.ct)
- Function length now returns 0 for VOID object
- Added a full set of FFT functions. There are now 12 functions:
  FFT, realFFT, sinFFT, cosFFT, sinqFFT, cosqFFT, and the inverses.
  A test module testfft.t was also added. These functions can take
  the transform along any dimension in a multidimensional array.
  (fft.ct, fftpack.H).

BUG FIXES (-Read carefully. Why so many this time?)

- Fixed bugs with arithmetic operations where output operand is the
  same variable as one of the input operands and where the input operands
  would be an illegal combination of noncompatible arrays. Previously Tela
  failed to give an error message and the result was garbage (objarithm.C,
  templ/*.C).
- Fixed bugs with negative real or integer x in x^y, when y is real.
  Previously e.g. (-1)^2.3 produced random garbage, now it produces
  a complex number as it should. This required taking Pow() away from
  template file tbinop.C and incorporating it explicitly in objarithm.C.
- Fixed a major memory leak in the lexical analyzer -- the input buffer
  was never deallocated! This is now fixed by defining an explicit
  destructor in d.l, but this should really be handled by flex++ automatically.
- Assignment x=x produced garbage because object assignment failed to
  check for identical objects being assigned (prg.C)
- As a general rule, operations are only permitted on equal-sized arrays.
  An exeption is that one of the operands may be a vector, e.g.
  rand(3) + rand(1,3) is allowed. Garbage resulted if both operands were
  complex arrays, for example 1i*rand(3)+1i*rand(1,3) (try it!).
  This was fixed at the end of templ/tbinop.C. But I'm not quite sure whether
  a similar bug might still hang out somewhere. A final solution to this
  problem would be to disallow the abovementioned extra liberty, but then
  again I'm not quite sure if something else breaks. Be prepared that someday
  rand(3) + rand(1,3) may be forbidden.
- Made where() to work for scalar arguments also (std.t)
- Mflops rates for complex operations are now somewhat closer to
  reality (objarithm.C, templ/tbinop.C)
- (Minor point) Removed self-overhead in perf() (std.ct)
- prod(<real-or-complex-array>) returned always 0. Fixed (std.ct)

INTERNAL CHANGES

- Added new machine instruction MMPOS (return last MIN/MAX position)
  (machine.H, machine.C, prg.C, Ctgen.C)

OTHER

- Started a special function package (not yet in public)

----------------------------------
Version 1.13 (Released 20.12.1994)
----------------------------------

USER-VISIBLE CHANGES

- Added function chol for Cholesky decomposition (la.ct, lapack.H)
- Added some color and line style definitions (telainit.t)
- Added functions autoglobal and hidesym (std.ct)
- Functions size, length and arg now cause error on undefined argument (std.ct)
- Added function telapath (std.ct)
- Enhanced function hist to handle two output arguments (plotmtv.ct)
- Added function source_silent (std.ct)
- Tries to source (silently) ~/.telainit.t on startup (telainit.t)

BUG FIXES

- export_matlab("f",[variable-list]) actually saved all variables.
  Now it saves only specified ones, as it should (files.ct: missed
  if (DoWrite)... in export_matlab). Thanks to Irix6.0 CC compiler
  for detecting this!!
- export_matlab now saves strings as row vectors, which is usually
  what is wanted.
- A'**B did not work. This was because matprod() did not accept
  int valued third argument but only char-valued (la.ct).
- function run left one file descriptor open if called without inputs (std.ct)
- several m2t fixes

INTERNAL CHANGES

- HPUX support

----------------------------------
Version 1.12 (Released 8.11.1994)
----------------------------------

INTERNAL CHANGES

- AIX support (configure.in; see also grep _AIX *.C *.H; see below)
- changed int->Tint in TDimPack members (object.H, object.C).
  Also in some places in gatscat.C. Should allow more flexibility with Tint.
- configure now tries to detect if pointer size (sizeof(char*) etc.) is longer
  than sizeof(int) and to define TPtrInt accordingly. This is needed in some
  Alpha/OSF systems.

KNOWN PROBLEMS

- To compile under AIX, you have to manually edit file d.y.c.
  Locate the lines

#ifndef __GNUC__
#include <alloca.h>
#endif

   Change these to

#ifndef __GNUC__
#ifdef _AIX
#  include <malloc.h>
#else
#  include <alloca.h>
#endif

   and you are done.

----------------------------------
Version 1.11 (Released 25.10.1994)
----------------------------------

USER-VISIBLE CHANGES

- Parser now recognizes the statements a+=b, a-=b, a*=b, a/=b (d.l, d.y)
- MTV options can be given to any graphics command (plotmtv.ct)
  (for example, plot(x,y,"-3d -landscape") works)
- Started new t-file matmenu.t
- Added functions matmenu, smatmenu (matmenu.t)
- Fixed some HTML link problems in documentation and wrote
  example file (telagraphs.html)

BUG FIXES

- Bugs associated with break and continue statements fixed. Previously worked
  incorrectly if used outside loops (now are ignored with a warning).
- Dividing integer array by integer scalar resulted in crash, for example:
  v = 1:5; v/=5; (objarithm.C:Div(), branch KIntArray,Kint, moved copydimsRealArray
  inside if statement).

INTERNAL CHANGES

- Tidy-up of codegen.C (classes TBreakPoint, TBreakPointList, Tcompilation),
  also fixed the abovementioned break/continue bugs. Thought about syntax
  break n; continue n (n positive integer), this should now be rather easy
  to implement if wanted (just replace LoopLevel-1 by LoopLevel-n in 
  FCONTINUE, FBREAK in codegen.C).

----------------------------------
PLANS FOR FUTURE VERSIONS (>=1.2):
----------------------------------
- Object arrays (lists), plus some new syntax for constructing them.
- Record/struct notation built on object arrays and associated syntax extension.
- Methods to write functions with arbitrary number of args may change.
  If you use the intrinsics argin, argout, SetArgOut a lot, contact the author!
- Current Tela does function definitions in code generation phase.
  Therefore you can use functions in t-files even if they are defined later
  in the file. But this also means that it is impossible to attach function
  definitions to local variables. This practice may change, so it is best to
  use functions only after they are defined in t-files. No severe problems should
  ever result, however, you could always fix them by doing some reordering.
  Ultimately functions should become Group A citizens, and there should
  also be a mechanism for writing pure functions (a'la lambda in lisp).

----------------------------------
Version 1.1 (Released 4.10.1994)
----------------------------------

USER-VISIBLE CHANGES
- Modifying function input arguments now causes a warning (prg.C, codegen.C)
- Added function holdmode (plotmtv.ct)
- Added function bar (plotmtv.ct) and a new demo that uses it
- Added function hist (plotmtv.ct)
- Added functions strmat, strmat2 (std.ct)
- Added function where (std.t)
- Modified tostring to stop on zero and to work for rank>1 also (std.ct)
- Added functions hide, unhide (std.ct)
- Added optional arg to whos to include hidden symbols (std.ct)
- Started new C-tela module fileio.ct
- Added functions fopen, fclose, fformat, printf, fprintf, sprintf (fileio.ct)
  (thus far sprintf only works for two args and may overflow).
  Fformat, sformat were already there.
- Added function export_matlab2 for variable name control (files.ct)
- Started new C-tela module numerics.ct
- Added function intpol for linear interpolation of N-dimensional arrays
  (numerics.ct).
- Added function grid3 (plotmtv.ct)
- Enhanced import1 to read "plain" ASCII files (files.ct)
- Added functions perf, GetInstructionData for telametric measurements (std.ct)
- Started new T-file perf.t
- Added telametric functions Mflops and Mips (perf.t)
- Started new T-file poly.t
- Added functions polyval, roots (poly.t)

BUG FIXES & alterations of weirdish featurings
- f(,x) no longer core dumps but produces syntax error (d.y: arglist_nonempty)
- A = A[i:j,k:l] now works correctly (gatscat.C:introduced Tobject a1
  in Gather) Previously worked incorrectly for rank>1 arrays.
- Made string matrices output as such (object.C). Tobedone: generalize to
  rank>2. Useful?
- Fixed bugs with SysVish CPUSeconds() (Solaris-2) (tela.C, common.H) that was
  introduced in 1.04. Thanks to Antti.J.Tanskanen at helsinki.fi!!
- Extracting part of a integer array carrying string flag now yields
  string flagged array.
- [a[1],b[3]] = f() has never worked correctly, but is now forbidden also (d.y)
  (only identifiers are allowed, as in [a,b] = f()).

INTERNAL CHANGES
- Now uses PlotMTV1.4.0
- 'Hidden' attribute for symbols (symbol.H); make attributes bitfields
- Allows chaining of indexing: a[1][2] etc (d.y)
- New flatcode instruction OZEROS
- New intrinsic function voids() for constructing object arrays.
  It is for now undocumented because object array support is still partial.
- Various documentation and configuration enhancements

----------------------------------
Version 1.04 (Released 8.9.1994)
----------------------------------

USER-VISIBLE CHANGES

- Added LUbacksubst function (la.ct)
- [lu,p] = LU(A) now returns pivot vector in p (previously returned
  the permutation matrix).
- Added function version() (std.ct)
- Added read pipe support (FindAndOpenFile() in tela.C).
  Example: source("!rsh uranus 'cat fyle.t'");
  (Tobedone: now fclose is used also for these -- any harm?)
- Added t-file madala.t (general elliptic solver).
- Added function run (std.ct)

BUG FIXES

- Fixed FindMachineEpsilon() to give correct result also in Linux (tela.C)
- v = rand(10); v[2:5] = -2.3; is now legal. I.e. implicit promotion
  of a scalar (-2.3) to a vector (#(-2.3,-2.3,-2.3,-2.3)) before
  indexed assignment. Before this was only permissible if rank(v)
  was greater than one, which was so bizarre feature that I consider
  it a bug.
- Removed redefinition of YY_READ_BUF_SIZE (d.l). 512bytes is too small.
- Removed exiting after error in argument file (tela.C)
- Removed TObjectStack destructor since it only causes core dumps (prg.C)

INTERNAL CHANGES

- Added SysVish /proc based CPUSeconds() support (for Solaris2)
- Possibility to increase MAXRANK from 4 to 8 by just changing
  a #define in def.H. To go beyond 8-dimensional arrays (!) you
  would have to modify gatscat.C. The default MAXRANK will stay
  at 4. Increasing it from four will increase the memory consumption
  somewhat and may also slightly degrade performance.

----------------------------------
Version 1.03 (Released 23.8.1994):
----------------------------------

USER-VISIBLE CHANGES

- Previously all arithmetic ops for two arrays were allowed, iff the
  LENGTHs of the arrays were equal. This means eg. that A+A' was legal.
  No more. Now the dimensions must agree, unless one of the operands
  is a vector, in which case it is sufficient that the lengths agree.
  For example if A=rand(2,3), B=rand(3,2), V=rand(6), then
  A+V and B+V are legal but A+B is not. The type of the result is the
  type of the first operand, ie. A+V is a matrix but V+A is a vector.
  Internally, this change amounts to introducing TDimPack::iscompatible
  and calling it in objarithm.C and templ/*.C with LengthError.
- Comparison a==b or a!=b where a or b is undefined now leads to error
  (templ/teqop.C, at end).
- Assignment a=b where b is undefined now leads to error (prg.C).
- Added function(al) map (std.ct)
- Added support for TELAPATH_SYSTEM environment variable, which is
  read before TELAPATH

BUG FIXES

- SGI: FFT did not work on Irix5.2 because wrong routines from
  libcomplib_sgimath.a were linked in. The complib_sgimath preceded
  libfftpack.a on link line, when the order was reversed, FFT
  started working again

INTERNAL CHANGES

- Moved some doc files to doc, reorganized local/ directory and
  made corresponding changes in Makefile, makeinc.in and local/makeinc.
  Also needed to change mkdocfile.
- Several ./configure enhancements

-------------
Version 1.02:
-------------
- numerous formal 'const' related changes in source code, to enable
  compilation under g++-2.6.0
- reintroduced ifdef BROKEN_UCHAR_OUTPUTTER in def.H.
  You must define it if your compiler output a hex address in case of
  cout << (const unsigned char*)ptr;

-------------
Version 1.01:
-------------
- enhanced Tobject assignment ops (object.H, object.C)
- telakka.in: can mix options with cc opts; some internal beautification
- other very minor changes

-----------------------------
3.8.94:		Tela-1.0 released
-----------------------------