June 18, 2001

PAW version 2.12/22

The PAW version 2.12/22 is part of the 2001 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	2.12/22	13/06/2001
KUIP	2.08/03	23/05/2001
HBOOK	4.26/01	19/12/2000
COMIS	1.22/03	07/02/2001
HPLOT	5.23/04	06/03/2001
HIGZ	1.27/03	31/05/2001
SIGMA	1.10/03	10/02/1994

The following subjects can be found in this page:

PAW

Improvements
Help improvements
Bugs fixed

HBOOK

Bugs fixed

HIGZ

Improvements
Bugs fixed

HPLOT

Improvements
Bugs fixed

KUIP

Improvements
Bugs fixed

COMIS

Bugs fixed

PAW

Improvements

New routines callable via COMIS: HF1N, DIVDIF, REXPIN, RLUXGO, DRTEQ3.
New accessible COMMON block via COMIS: HCFLAG.
The DOUBLE COMPLEX functions DCMPLX, CDSQRT, CDLOG, and CDEXP are now loaded in the PAW executable module to make them available in COMIS functions via dynamic linking.
' ' is allowed as separator in NT/PLOT options. This prevent to have the message "Unknown option ' '" when ntuple plot is used with BOX option from the ntuple panel in Paw++.
D is recognised as a valid option in NT/PLOT (meanning "Default") because it is used by Paw++. The query processor printed an error message saying it was not a valid option.
New oprion B in /MLP/LEARN to keep the weights from the smallest error on the test sample.

Help improvements

New option 'A' in 3DNULL documented.
contour plot help updated
update FUN/PLOT help
update HELP of the command CHAIN
Nt/Dup help had a mistake
^ is replaced by ** on the MLP/LPAT/SET help. It was producing bad LaTex reference manual
update the help of MANY_PLOTS
run ispell on the cdf files
revise the keywords in pawcdf.cdf

Bugs fixed

Because of the long file names, the following macro produced wrong results. When an HBOOK file was used, its name was stored in a character strings array of 80 characters. The chain mechanism used this array to check if a file was already attached. When the file names was too long (like here) the comparison was done with a truncated name. If the file names were very similar and vary only at the end (like here) the truncated character string stored in the HBOOK common block were the same for two different files and the chain mechanism thought that the file was already attached whereas it was not. This ended up analysing twice the same file !
```
filecase keep
chain bpipimbJ Y01319_1-4.his Y01319_5-8.his
chain bpipimbJ -P /afs/rl.ac.uk/atlas/data/level2/ctrighis/011303p314/Bpipimb/
cd //bpipimbJ/t2gl
nt/plot 1202.nKin nKin<100
```
Extending the lenght of the character strings would just push the limit but not solve the problem. To highlight the fact that a problem may occur, a warning is now printed when the file names are too long. A workaround is to create short symbolic links (UNIX) to the long directory path common to all files.
On LINUX platforms only . Due to several bugs (some linked to Motif) the Paw++ fitting panel didn't work on LINUX.
The parameters used to describe colour and filling of the FUN/PLOT appearance change if it was a first plot or a superposition. The following macro shows the problem:
```
   set HCOL 2
   f/plot x 0 10
   set FCOL 5
   f/plot 10-x 0 10 S
```
A workaround is to use HCOL for the first plot and FCOL for the superposition.
The command MASK/FILE didn't work with mask names containing a full path name like:
```
MASK/FILE /a/b/c/m Ni
```
The error message in case of old MASK format was wrong.

If one used a 8-digit histogram identifier [hid1] and used

    h/op/div [hid1] [hid2] [hid3]

the result [hid3] was created but [hid1] was deleted afterwards. The following macro demonstrates the problem. If one uses a 7-digit number for [hid1] instead of the 8 digits it works properly.

 h/del *
 v/del *
 hid1=16221810
 hid2=4211888
 hid3=16221820
 zone 1 3
 v/cre data(36) r 9*0.0 37.6516 22.0652 30.6946 26.6565 18.05   22*0.0
 v/cre err(36)  r 9*0.0 7.02594 5.42831 6.11107 6.00046 5.18477 22*0.0
 v/cre one(36)  r 36*1.0
 v/cre eone(36) r 36*0.01
 1d [hid1] momentum 36 0 1.8
 1d [hid2] one      36 0 1.8
 h/plo [hid1]
 put_vec/cont [hid1] data
 put_vec/err  [hid1] err
 put_vec/cont [hid2] one
 put_vec/err  [hid2] eone
 h/plo [hid1]
 h/li
 mess **** divide histos ********
 h/op/div [hid1] [hid2] [hid3] 1.0 1.0 e
 h/li
 mess **** plot histo *******
 h/plo [hid1]

This bug fix had a side effect (now fixed) which prevented the following macro to work as in the past:

 h/file 45 behist_test.rz
 h/copy 11 991
 h/copy 12 992
 opera/add 991 992 993
 h/plot 991

The histogram 991 was not seen.

Negative fonts identifiers were not properlly treated in the Paw++ fonts panel.
The Paw++ fitting panel works on LINUX.
Call GDELS via cfortran.h to make it work on NT.

HBOOK

Bugs fixed

To be "Y2K" compliant routine HUWFUN has been changed to use the new version of DATIME. The calling sequence used was CALL DATIME(ID,IT). This is the one used is the manual. ID and IT are not used because after the call to DATIME, the time and date are taken in a common block. Unfortunately, the variable name ID was used in the common block HCFLAG. So this variable was corrupted by the call to DATIME and the ntuple identifier in the routine generated by HUWFUN was wrong.

In some case the RMS calculation of 1d histograms was wrong, on SGI, HPUX, DEC/OSF and Solaris.

1dh  1 'Peak HFILL' 50 0. 5.
idopt 1 STAT
appl comis quit
      subroutine fillpeak
         do i=1,3000
            call hfill(1,4.35022,0.,1.)
         enddo
      end
quit
call fillpeak
h/pl 1

With this macro the RMS should be around 10E-6. On SGI, HPUX, DEC/OSF and Solaris it gaves 10E-3.

HIGZ

Improvements

newpath has been added at the beginning of the PS and EPS file to close some graphics path which may have been left opened by the application in which the HIGZ PS or EPS file is included.

Bugs fixed

With an external window as parent, the position of the window created by ixopnwi was not correct if the external window was itself attached to a parent window at a relative position different from (0,0).
In Pie Chart drawing the labels were truncated after 12 characters.

HPLOT

Improvements

New option 'A' in HPLFR3 to draw the axis only.

Bugs fixed

Sometimes on SGI machines, empty bins become big spikes when opt LOGY is set. The optimisation level of the HPLOT routine HPLH was too high. The following macro demonstrate the problem:

macro bummer
*
* Vectors trash1 and trash2 differ only in that -
*          trash1 has 0.0 in bin 1 and something non-zero in bin 2
*      but trash2 has something non-zero in bin 1 and 0.0 in bin 2
*
vec/del *
h/del *
vec/create trash1(30) r  0.0  2.0  3.0  0.0  5.0  0.0  7.0  0.0  9.0  0.0 _
                        11.0  0.0 13.0  0.0 15.0  0.0 17.0  0.0 19.0  0.0 _
                        21.0  0.0 23.0  0.0 25.0  0.0  0.0  0.0  0.0  0.0
*
vec/create trash2(30) r  1.0  0.0  3.0  0.0  5.0  0.0  7.0  0.0  9.0  0.0 _
                        11.0  0.0 13.0  0.0 15.0  0.0 17.0  0.0 19.0  0.0 _
                        21.0  0.0 23.0  0.0 25.0  0.0  0.0  0.0  0.0  0.0
zone 1 2
*
h/create/1d 98 'trash1' 30 0.0 30.0
h/put/cont 98 trash1
opt liny
h/plot 98
opt logy
h/plot 98
wait ! 0.0
*
h/create/1d 99 'trash2' 30 0.0 30.0
h/put/cont 99 trash2
opt liny
h/plot 99
opt logy
h/plot 99
*
return

The axis titles after a 3DNULL where missplaced.
```
PAW > 3DNULL
PAW > ATITLE AAA BBB CCC
```
There was some rounding problems on IEEE machines.The following commands are enough to demonstrate the problem:
```
PAW > LABEL 2 5 a b c d e
PAW > SET NDVX 10.20
PAW > NULL 0 1
```
On RS6000, Linux and Windows the labels are not visible. On these machine 10.20 becomes 10.1999999 and the label list used is 1 instead of 2. To fix that 0.001 is now added to the number of divisions.

KUIP

Improvements

The commands VECTOR/READ and VECTOR/WRITE could not handle file names longer than 80 characters. This limit has been extended to 160 characters.
The system function $FEXIST did not work on file names containing ~. Even on an existing file, it returned 0:
```
PAW > message fexist('~/aaa')
 0
```
In some KUIP include files, C++ keywords "template" and "delete" were used for variables names in C prototypes. This prevent to use this include files with the GNU C++ compiler g++.
Modifications to suppress a warning at compile time on Solaris 2.7.

Bugs fixed

On LINUX: Mods to allow the closing of KUIP Motif panels. Closing KUIP panels crashed previously on LINUX. The following macro demonstrate the problem:
```
    panel 0
    panel 1.01 'hello'
    panel 0 d 'first' 120x100+0+0
    panel 0 c 'first'
```

COMIS

Bugs fixed

On NT the following macro did not work:

MACRO rl.kumac

APPLICATION COMIS QUIT
      SUBROUTINE readhead
      CHARACTER*80 TEXT
      IFILE=15
      OPEN(IFILE,FILE='aes0212a.lsb',STATUS='OLD',ERR=820)
      READ(IFILE,789,ERR=822,END=821)TEXT
      CLOSE(IFILE)
      print*,TEXT
      STOP
789   FORMAT(80A)
820   WRITE(*,*)' # ERROR OPENING FILE'
      STOP
821   WRITE(*,*)' # END OF FILE DETECTED'
      CLOSE(IFILE)
      STOP
822   WRITE(*,*)' # ERROR READING FILE'
      CLOSE(IFILE)
      STOP
      END
QUIT

CALL readhead

RETURN

The A format was not handled correctly.

The SGI systems has a minor inconsistency in dynamic loading. Both f77 and cc default options had -n32 but ld didn't. As a result, attempts to build a shared library file from a COMIS function failed.
On SunOS platforms older than SunOS 5.7 dynamic linking inside PAW didn't work properly. When one try to change a COMIS program and relink, PAW didn't pick up the new version of the code. If one changes the main COMIS routine, the new version is used. However, if one changed a function called by the main COMIS routine, the new version was not used.
In a DO loop, we expect a type conversion of the index limits in integer if it is real. Eg. consider the following program compiled with a GNU-Fortran:
```
       r=3.5
       do i = 1, r
          print *, i
       enddo
       end
```
The output of this is:
```
 1
 2
 3
```
But when this is done in a function handled by the COMIS interpreter there was much more than three lines of output.