March 22, 1999

PAW version 2.10/09

The PAW version 2.10/09 is part of the 99 release of CERNLIB. The source files, binaries and libraries are available, via anonymous ftp, on the asis system:

                       ftp asisftp.cern.ch

The PAW support is done by:

Olivier Couet.

The PAW related packages versions are:

Package	Version	Date
PAW	2.10/09	01/03/1999
KUIP	2.07/17	05/01/1998
HBOOK	4.25/03	05/03/1999
COMIS	1.21/00	13/05/1998
HPLOT	5.21/04	07/12/1998
HIGZ	1.25/05	07/01/1999
SIGMA	1.10/03	10/02/1994

The following subjects can be found in this page:

Year 2000 bug
PAW

Improvements
New commands
Bugs fixed
HPUX specific bugs fixes
Linux specific bugs fixes
Solaris specific bugs fixes

HBOOK

Improvements
Bugs fixed

COMIS

Improvements
Bugs fixed

HIGZ

Improvements
Bugs fixed

HPLOT

Improvements
Bugs fixed
Alpha OSF specific bugs fixes

KUIP

Bugs fixed
NT specific bugs fixes

Year 2000 bug

COMIS. HBOOK, HIGZ, HPLOT, KUIP and PAW have been modified to be ``Y2K compliant''. In these packages there is no computations on dates. All the modifications were to change printing with two digits into printing with four digits. These packages use now the new version of DATIME. The modifications have been tested by changing the date on one of our machines and running PAW on it. Every things looked fine ... so now wait 2000 ! ...

PAW

Improvements

Help improvements in the commands: BOX, UWFUNC, HISTOGRAM/FIT, VECTOR/FIT, METAFILE and CALL
The HELP of several commands has been improved to have a better look in the html version of the reference manual
New routines callable via COMIS: ITX3, MNSTAT.

SUBROUTINE ITX3(X,Y,Z,TEXT)
Draw a text in 3D space

SUBROUTINE MNSTAT(FMIN,FEDM,ERRDEF,NPARI,NPARX,ISTAT)
Provides the user with information concerning the current status of the current minimization. Namely, it returns:
- FMIN: the best function value found so far
- FEDM: the estimated vertical distance remaining to minimum
- ERRDEF: the value of UP defining parameter uncertainties
- NPARI: the number of currently variable parameters
- NPARX: the highest (external) parameter number defined by user
- ISTAT: a status integer indicating how good is the covariance matrix:
  - 0= not calculated at all
  - 1= approximation only, not accurate
  - 2= full matrix, but forced positive-definite
  - 3= full accurate covariance matrix
In the command METAFILE the unused parameter CHMETA removed
Commands in OBSOLETE/GRAPHICS/ATTRIBUTES/ print now an error message instead of a warning and the command is NOT executed.
PICTURE/PRINT can now scale output GIFs
New option in VLOCATE: "Q" Do not print the number of points entered

KUIP/UNITS has been replaced by FORTRAN/UNITS and is now able to list the HBOOK files again.

PAW > units
 +----+---------------------------------------+------------+-----------+------+
 | Lu | File name                             | Format     | Access    | Recl |
 +----+---------------------------------------+------------+-----------+------+
 | 10 | paw.metafile                          | FORMATTED  | SEQUENTIAL|      |
 +----+---------------------------------------+------------+-----------+------+
 |  1 | bla.hbook                             |            | DIRECT    |      |
 |  2 | Staff.hbook                           |            | DIRECT    |      |
 |  3 | test.hbook                            |            | DIRECT    |      |
 |  4 | /home/couet/cernlib/paw/demo/pawdemo  |            | DIRECT    |      |
 |    | .hbook                                |            |           |      |
 +----+---------------------------------------+------------+-----------+------+

If the file name is too long it is printed on several lines.

New commands

A set of graphics commands to produce 3D plots has been added: 3DNULL, 3DPLINE and 3DPMARKER

Bugs fixed

The ELLIPSE command was always drawing ellipses with the first angle equal to zero even if a non zero value is given.
```
 
  null 0 10 0 10
  ellipse 5 5 4 2 90 270
```
The command H/GET/ABS gives a wrong values for histograms with one bin only. The following macro shows the problem:
```
 h/cre/1d 100 'essai' 1 0.9 1.1
 v/cre x(1)
 h/get/abs 100 x
 v/pri x
```
In the command UWFUNC the option "P" allowed to generate code to print the ntuple variables (for RWN only). Since the last version of the ntuple queries processor (PAW version 2.07) the code generated was not working properly.
The MATCH parameter in NT/READ was case insensitive.
The H/GET/ABS command produced wrong value for histograms with one bin only. The following macro shows the problem:
```
 h/cre/1d 100 'essai' 1 0.9 1.1
 v/cre x(1)
 h/get/abs 100 x
 v/pri x
```
UWFUNC : Bug fixed in case of option P. The previous version was not working properly.

HPUX specific bugs fixes

Compiled fortran code using math functions produced a crash:

      subroutine aaa()
      print*,tan(1.279)
      end

PAW > call aaa.f77

 *** Break *** Bus error
( 0)  0x00254348   signal_handler + 0x1a0  [/cern/97a/bin/pawX11]
( 1)  0xc08562e0   _sigreturn  [/usr/lib/libc.1]
( 2)  0xc1174d88   aaa_ + 0x70  [/tmp/aaa_4517.sl]
( 3)  0x00161e54   cscali_ + 0x64  [/cern/97a/bin/pawX11]
( 4)  0x00159a34   cskcal_ + 0x2f4  [/cern/97a/bin/pawX11]

Workaround: Change the compiler options by executing the following macro:

appl comis quit
!setopt 'f77 -c +Z +ppu' f77
quit

Linux specific bugs fixes

A RZ file, which was created with:
```
          ZEBRA/RZ/MAKE 20 [fnam] 4096 ! ! H CAL-ID
```
on a linux system, cannot be red on an AIX system after binary ftp. It crashed there with the error message telling that this file was not an rz file. ON linux:
```
 ZEBRA/RZ/MAKE 20 bla.rz 4096 ! ! H CAL-ID
 cdir //lun20
 mdir FOPI ! H Cal-id
 fortran/close 20
```
on AIX:
```
 RZ/FILE 24 bla.rz 4096
```
Mods to avoid a black picture when the geometry panel is displayed in PAW++.

Solaris specific bugs fixes

On SunOS 5.6 (Solaris 2.6) PAW++ crashes each time a panel containing toggle buttons is displayed (Open HBOOK file, Ntuple viewer etc ...). This is due to a bug in the SUN Xm library. The patch T105284-23 fixes this problem. With a more recent version of libXm.so.3 (revision 23) the problem is also fixed.

HBOOK

Improvements

New option P for profile histograms(HBPROF). Implemented by: Nello Nappi
This option allows to overcome the severe precision problems which occur in the error calculation when the standard deviation is much smaller than the mean. In such a case, the use of the standard formula:
```
             
                      < ( y -  )^2 > = < y^2> - ^2
        
```
involves the calculation of a small difference between two large numbers. The problem is avoided by accumulating the sum of the squares of the deviations of y with respect to the mean, instead of the sum of the squares of y. The prior knowledge of the mean is avoided by the use of an iterative formula making use of the running average.
Solaris machines are now flaged as SYSTEM V architecture. Previously they were consider as BSD (like SUNOS). This (being BSD) disabled the shared memory facility which was emulated with files.

Bugs fixed

Bug found and fixed by Kees van der Poel (Kees.van.der.Poel@nikhef.nl):
I use integer profile histograms (chopt='I') for plotting efficiencies. But I found that the errors were really big, so I checked in the 97a source code (the CERNLIB version I am using) how the errors are calculated. In the routine hcx.F the errors and values are calculated. But there are inconsistencies between manual (4.23 and 4.24 version) and code (also for the 98 version):
This last error caused my big errors for the integer histograms.

In HOPER1 (a service routine used by HOPERA), the option L in MZDROP was missing. Without this option the following PAW macro didn't work:

v/cr p3(3) r 0 1 0
v/cr a(10) r 1 3 2 5 4 8 6 10 7 9
1d 1 '' 10 0 10
put/con 1 a
h/fit 1 p2.f q 3 p3
add 1 1 2
h/pl 2

with p2.f:

      REAL FUNCTION P2(X)
      REAL X, PAR(3)
      COMMON /PAWPAR/PAR
      P2 = PAR(1) + PAR(2)*X + PAR(3)*X*X
      END

When filling a CWNtuple and using a 24bit packing of a real, for each entry equal to the maximum value XMAX the value actually stored into the CWN was XMIN. This only seems to happen with 24 to 30 bit packing. This error occurred on HPUX and IRIX (at least).
FORTRAN I/O is the default for VMS
CWNs containing REAL*8 were not portable from small endian machines to big endian one.
The following test program shows the problem:
```
      Program PR8
*
      Parameter (Nhmax = 100000)
      Common /Pawc/ Hmemor(Nhmax)
      REAL*8 R8
      Common /P/ R8,R4
*
      Call Hlimit(Nhmax)
*
      Call Hbnt(100,'R8 Test',' ')
      Call Hbname(100,'P',R8,'R8:R*8,R4')
*
      Do N=1,9
         R8 = N*111111.
         R4 = N*111.
         Print*,' R8 = ',R8 ,' R4 = ',R4
         Call Hfnt(100)
      EndDo
*
      Call Hrput(0,'r8.hbook','N')
      Close(10)
*
      End
```
This program generate the file r8.hbook which you can look at with the following PAW macro:
```
hi/file 0 r8.hbook
nt/scan 100
```
If the file r8.hbook is generated on on a HPUX, IBMRT or SUN and if you look at it on DEC OSF or LINUX The macro gives:
```
+-------+-----------------------+--------------+
| Event |   R8                  |   R4         |
+-------+-----------------------+--------------+
|     1 |  5.3863020109007d-315 |  111.        |
|     2 |  5.3914826646872d-315 |  222.        |
|     3 |  5.3944674733548d-315 |  333.        |
|     4 |  5.3966633184738d-315 |  444.        |
|     5 |  5.3985502045818d-315 |  555.        |
|     6 |  5.3996481271413d-315 |  666.        |
|     7 |  5.4007460497008d-315 |  777.        |
|     8 |  5.4018439722603d-315 |  888.        |
|     9 |  5.4029418948198d-315 |  999.        |
+-------+-----------------------+--------------+
```
Which is obviously wrong. If you do the contrary: generate the file on DEC OSF or LINUX and look at it on HPUX, IBMRT or SUN you get also something wrong. All the other combinaisons give a correct result.
It is a "word swapping" problem. In HBOOK code REAL*8 variables are stored in two REAL*4 without taking care of possible swapping.
This problem is now fixed with the following limitations:
- CWNs containing REAL*8 variables generated with the new HBOOK (4.25/03) on any kind of machine will be readable by the new PAW on little endian machines and all the PAW versions on the big endian machines.
- CWNs containing REAL*8 variables generated with any HBOOK version on big endian machines will be readable by all PAW versions on big endian machines and only by the new PAW version on little endian machines.
- CWNs containing REAL*8 variables generated with an old HBOOK version (before 4.25/03) on little endian machines will be readable ONLY by the old PAW versions on little endian machines.

COMIS

Improvements

Shared libraries mechanism implemented for LINUX.
Default compilation options for HPUX have been changed. With HPUX 10.20 the option -O produced a crash when COMIS routines containing math function(s) were called with the f77 extension.
Operators .EQV. and .NEQV. have been added.

Bugs fixed

notation like 1.e-2 were not understood by COMIS.

HIGZ

Improvements

New version of the routine imgpickpalettecoming from a C++ version from Fons Rademakers. This version allows to work on TrueColor devices. ixdogif in x11int.c has been modified in order to use this new version of imgpickpalette. Protections have been added in ixdogif to prevent some crashes when one try to create a gif after a window resizing or when the program (for example PAW) started with a workstation type equal to 0. All the static arrays in ixdogif have been converted into dynamic one.
imgscaling.c: new procedure to scale bitmap images.
In x11int.c:
- ixdogif can scale pictures
- uses image size instead of window size in ixdogif
- loop on MapNotify event in ixopnwi to be sure the window is mapped
- suppress some warning at compilation time (LINUX)
ixdogif can scale pictures
APOLLO version removed from the source files.
GDDM driver removed from the source files.
IGTABL: It is now possible to set the size of the characters in case of option 'T' and 'C' (in percent of the bin heigh)

Bugs fixed

For filling patterns 1 to 25 the PostScript output was wrong in some cases. For those the pattern we now close the polygon (move to X1 Y1).
The following macro generated a wrong PostScript file: the histogram fill pattern was incomplete.
```
size 26 20
set hcol 1002
set htyp 4012
for/file 66 test.ps
meta 66 -112
h/pl 110(1:40)
close 66
```

IGELLI always drew the ellipse with the first angle equal to zero even if a non zero value is given.

        null 0 10 0 10
        ellipse 5 5 4 2 90 270

Some improvements have also be done in the ellipse drawing:

      null 0 10 0 10 ; ellipse 5 5 2 4 1 360 30
      set bord 1
      null 0 10 0 10 ; ellipse 5 5 2 4 1 360 30
      set fais 1
      set faci 2

HPLOT

Improvements

APOLLO version removed from the source files.
2SIZ option is available again in the SET command. It defines the size of the characters when 2D histograms are plotted with options TEXT or CHAR. The size is defined in percent of the bin heigh.

Bugs fixed

According to the online help, a histogram can be plotted without axis labels and tick marks using hist/plot [id] a. This works fine until one issued the command set fcol 1100 , which has the effect that despite the option A the tick marks are drawn.

The axis title position was changed with SET TSIZ. The following macro shows the problem:

set tsize
null 0 20 0 20
atit xtitle ytitle
set tsiz 0.2
atit xtitle ytitle
set tsiz 0.5
atit xtitle ytitle

if FCOL was greater than 1000 the axes were redrawn even if option "A" was used.

Alpha OSF specific bugs fixes

With 2D histograms if one tries to, set its minimum value, then plot its x projection, PAW stopped with a 'floating invalid' error.

KUIP

NT specific bugs fixes

The following macro produced a ``core dumped''.
```
a = $sigma(1.)
b = $sigma(1.)
c = $sigma(1.)
```