Google
 

Trailing-Edge - PDP-10 Archives - bb-bt99e-bb - cbl12b.d06
There is 1 other file named cbl12b.d06 in the archive. Click here to see a list.
                 EDIT DESCRIPTIONS FOR COBOL-20-V12B                            
  
  
                             EDIT 44     FOR COBDDT
  
[SYMPTOM]
  
  
  
  
A  histogram  table  produced  by  some  programs  will   be
incorrect  (some paragraph names are omitted and some values
are wrong) whereas similar but smaller  programs  produce  a
correct table.
  
  
[DIAGNOSIS]
  
The histogram table is being overwritten.  A  TECOing  error
was   made   during   v12B  development  which,  when  other
conditions are right, will cause this to happen.
  
The full story is:
  
When the History Initialize or Begin command is given COBDDT
calculates  how  much  space  it needs to hold the histogram
values.  It calls GWORDS to allocate this much.
  
However, GWORDS allocates space on a page boundary.  We  are
not  sure  if  this is required here but the routine is also
used by overlay code.  In any case this  causes  no  problem
except  to  waste  some  space.  However, it does leave some
space free below the histogram  tables.   At  the  end  when
COBDDT  wants  to print the histogram it calculates how much
space is required for two buffers and asks for  that  space.
If the previous hole is big enough it gets space from it and
all is well.  However, if the hole is not big enough  it  is
given  that  address anyway and eventually the output buffer
wipes out the histogram data giving  the  incorrect  results
seen.
  
The reason for this is  a  TECOing  error  made  during  12B
development.   The  arg  block  for the FUNCT.  call (HSTFFN
through HSTADR) was originally all EXP 0s.   At  some  point
the  first  four EXP 0s were changed to BLOCK 1s except that
somehow they ended up as BLOCK 0s.  Thus  giving  the  first
five words the same address.  The only effect of this was to
replace the size required  (406  words)  with  the  function
requested  (6).   Thus,  FUNCT.  thought it was allocating 6
words rather that 406 and almost always found space  in  the
previous hole.
  
  
[CURE]
  
Turn the BLOCK 0s into BLOCK 1s.
  
********************************************************************************
  
  
                             EDIT 45     FOR COBDDT
  
  
  
  
[SYMPTOM]
  
COBDDT does not ACCEPT COMP-1 numbers if they contain a decimal point.
  
  
  
[DIAGNOSIS]
  
There is no code to input floating point numbers.
  
  
  
[CURE]
  
Implement ACCEPT of floating point numbers.
********************************************************************************
  
  
                             EDIT 1446   FOR COBOL
  
  
  
  
[SYMPTOM]
  
  
A user-defined currency symbol that is used in  place  of  a
floating  dollar  sign  in a PICTURE definition along with a
minus sign will be overwritten by  the  minus  sign  if  the
number being displayed is negative.
  
  
[DIAGNOSIS]
  
  
To use the extended  instruction  EDIT,  a  pattern  command
string  is  built  by  the  compiler.   This  pattern string
dictates how to process the source characters to  build  the
desired  output  from the given PICTURE definition and data.
While building this string, a test  is  made  on  the  float
character,  and if it is not a dollar sign, it is assumed to
be a plus or minus sign.  This causes an extra routine to be
built  on  to  the end of the pattern string to determine if
the source number is negative, and if so, to insert a  minus
sign  before  the  first significant digit.  This is causing
the user-defined currency symbol to be overwritten.
  
  
[CURE]
  
  
Put in an additional test after the dollar sign test to  see
if  the  float  character  matches  the  field DOLLR., which
contains any user-defined currency symbol, and on  a  match,
skip over the routine which generates the leading sign code.
********************************************************************************
  
  
                             EDIT 1447   FOR COBOL
  
[SYMPTOM]
  
Program names containing any character other than "A" - "Z",
"a" - "z", or "0" - "9" as the first character will generate
an illegal instruction during compilation.
  
  
[DIAGNOSIS]
  
During code generation,  the program name is interpreted  as
a UUO call rather than a literal.
  
  
[CURE]
  
When the program name is encountered, skip over the test  to
see if it is in the form of a UUO call.
********************************************************************************
  
  
                             EDIT 1450   FOR COBOL
  
  
  
  
[SYMPTOM]
  
FREE does not work on key = 0 for dynamic access.
  
  
  
[DIAGNOSIS]
  
FREE has to be followed by key name or  EVERY  for  COBOL-74,  but  no
error message is generated by the compiler to show the syntax error.
  
  
  
[CURE]
  
Edit 1450 to IOGEN and DIAGS to catch this syntax error.
********************************************************************************
  
  
                             EDIT 1451   FOR COBOL
  
[SYMPTOM]
  
  
  
  
COBOL generates fatal error message 'MUST BE A HIGHER  LEVEL
CONTROL THAN THAT OF THIS GROUP' at random.
  
  
[DIAGNOSIS]
  
To test to see if the RESET  control  identifier  is  higher
than  that  of the group level control identifier, a test is
made to see which of  the  two  control  identifier's  table
addresses is smaller.  If the RPWTAB is expanded between the
time the first address is saved and when the test  is  made,
then the address saved is invalid to use.
  
  
[CURE]
  
Rather  than  test  the  absolute  address  of  the  control
identifiers, use the relative address into the RPWTAB.
********************************************************************************
  
  
                             EDIT 1452   FOR COBOL
  
  
  
  
[SYMPTOM]
  
     COBOL generates fatal error  message  'MUST  BE  A  HIGHER  LEVEL
CONTROL THAN THAT OF THIS GROUP' at random.
  
  
[DIAGNOSIS]
  
     To test to see if the RESET control  identifier  is  higher  than
that  of  the  group  level  control identifier, a test is made to see
which of the two control identifiers table addresses is  smaller.   If
the RPWTAB is expanded between the time the first address is saved and
when the test is made, then the address saved is invalid to use.
  
  
[CURE]
  
     Rather than test the absolute address of the control identifiers,
use the relative address into the RPWTAB.
********************************************************************************
  
  
                             EDIT 1453   FOR COBOL
  
  
  
  
[SYMPTOM]
  
  
Edit 1411 gives a warning messgae for left  truncation,  but
the source line number is wrong.
  
  
[DIAGNOSIS]
  
  
The code to get hte source line number is not correct.
  
  
[CURE]
  
  
Use the same routine that is used in  other  places  in  the
compiler.  Note edit 1411 must be applied before this edit.
********************************************************************************
  
  
                             EDIT 1454   FOR COBOL
  
  
  
  
[SYMPTOM]
  
     The 68274 converter when converting the SYMBOLIC KEY clause to  a
comment, places the key on the line following the SYMBOLIC KEY clause.
  
  
[DIAGNOSIS]
  
     When converting a line to a comment, the converter inserts a CRLF
before  the  next  statement.   The module that updates where the next
statement is does not update the pointer if the next  statement  is  a
period.   So  when  the  converter  inserts  a  CRLF into the line, it
inserts it before the last word in the statement.
  
  
[CURE]
  
     Update the pointer to the start of the next clause if a period is
seen.
********************************************************************************
  
  
                             EDIT 1455   FOR COBOL
  
[SYMPTOM]
  
  
  
  
The 68274 converter does not correctly convert the statement
following  the  NOTE  clause if the statement is of the same
line as part of the NOTE clause and if the first word of the
statement is replaced by another word.
  
  
[DIAGNOSIS]
  
The 68274 converter uses two internal  buffers  to  generate
the translated code.  When the converter makes a statement a
comment, it must make sure that it is the only statement  on
that  line.  To do this, the converter writes out to the CVT
file the contents of the first buffer.  The second buffer is
then  copied  to  the  first  buffer.  The statement that is
being commented out is removed from the  second  buffer  and
replaced  by  spaces.   The  statement  that  is  not  being
commented out is removed from the first buffer and  replaced
by  spaces.   When this is completed, the converter does not
update the position and state of the current line.
  
  
[CURE]
  
Set the pointer to the current word on  the  second  buffer,
and  reset  the  flag  that  tells if the statement is to be
commented out.
********************************************************************************
  
  
                             EDIT 1456   FOR COBOL
  
  
  
  
[SYMPTOM]
  
  
If the first term in a COMPUTE statement uses a 'ZERO' as  a
factor the next term will be ignored.
  
  
[DIAGNOSIS]
  
  
The module MATGEN turns on a flag when it determines that  a
factor  in  a multiplication is zero, and does not bother to
generate  code  to  perform  the  multiplication.   If   the
processing  of  the  next  part of the expression requires a
subexpression be calculated before the zero  term  is  used,
the  zero  field  is  moved to a temporary location, but the
flag signifying 'zero in an  accumulator'  is  never  turned
off.  The next term is also considered to be zero, and if it
also requires a multiplication, it is not performed.
  
  
[CURE]
  
  
In  the  module  EXPGEN,  when  it  is  determined  that  an
accumulator  must be moved to a temporary location, shut off
the flag that signifies 'zero in an accumulator'.
********************************************************************************
  
  
                             EDIT 1457   FOR COBOL
  
[SYMPTOM]
  
Very occasionally the usage of the syntax IF SET-NAME SET for DBMS-10/20
Data Manipulation Language will cause garbage to be placed in the Literal
generated for the set-name by the compiler.
  
[DIAGNOSIS]
  
The code in IFGEN which creates the Literal calls STASHL to put it into
the Literal Table. With the particular example provided the Literal
Table had to be expanded in the middle of putting the Literal in place.
In the process the VALTAB area which was the source of the text of the Literal
got moved down in core because it is below the Literal Table. However,
the pointer to the address of this area was never updated. So, when the
IFGEN routine went to get the rest of the name it was now pointing
at garbage.
  
[CURE]
  
Make sure that the VALTAB pointer gets updated by passing it to the
STASHL routine in the field CURVAL. If the Literal Table has to be
expanded, the XPAND routine will be called and it will update the
pointer. This is done in edit 1457 to the COBOL-10/20 compiler.
********************************************************************************
  
  
                             EDIT 1467   FOR COBOL
  
[SYMPTOM]

When relinking the COBOL compiler, the undefined  global
ALCPA.  may be detected.


[DIAGNOSIS] 

The routine ALCPA.  is called from SCHIO2, as  a  result
of  DBMS-20  V.6 edit 473.  If a custoem installs that edit,
and then subsequently attempts to relink the COBOL  compiler
using the DBMS-patched SCHIO2, the undefined ALCPA.  will be
detected.


[CURE] 

Resolve ALCPA.  within the module  DBGETF,  where  other
DBMS-type  memory  management  calls  are  resolved  for the
compiler.
********************************************************************************
  
  
                             EDIT 1047   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
During simultaneous access, if the process that  has  the  lock  on  a
record encountered a fatal error and was aborted, another process that
has been in ENQ I/O wait for this locked record would hang  until  the
aborted process is reset.
  
  
  
[DIAGNOSIS]
  
When the first process encountered the fatal error, it took the  abort
route  without dequeuing the locked record.  Therefore other processes
cannot access this record even though the first process is killed.
  
  
  
[CURE]
  
Edit 1047 to LSU to do the dequeue before the job is aborted.
********************************************************************************
  
  
                             EDIT 1050   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
Accessing a  relative  file  randomly  or  dynamically  will
result  in  the  end-of-file  path  being taken on the first
empty record encountered.
  
  
[DIAGNOSIS]
  
  
The last block number field, D.LBN, is  not  being  set  for
random  files opened for input only.  When reading the file,
if an empty record is encountered, a routine is  entered  to
add  one  to the count of records processed and to check the
current block number against the last block  number  in  the
file.   Since this field has not been set up, end of file is
always set.
  
  
[CURE]
  
  
The OPEN routine should set the  last  block  number  field.
Edit  1050 to CBLIO.MAC will add a call to OPNEL2 which will
use the file size in the extended lookup argument  block  to
determine the last block number and store it in D.LBN.
********************************************************************************
  
  
                             EDIT 1051   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
REWRITE doesn't work after READ NEXT on  relative  file  with  dynamic
access.
  
  
  
[DIAGNOSIS]
  
When LSU tries to  get  the  actual  key  for  comparsion,  the  wrong
instruction  is  used  and  therefore the data from a wrong address is
assumed as the actual key and the comparsion fails.  This  failure  of
comparsion leads to the error message of "record not retained".
  
  
  
[CURE]
  
Edit 1051 to LSU to get the correct address.
********************************************************************************
  
  
                             EDIT 1052   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
  
When extending a blocked file LIBOL tries  to  determine  if
the  last  block is completely filled or not.  This is being
done incorrectly for SIXBIT files,  resulting  in  one  less
record than is possible.
  
  
[DIAGNOSIS]
  
  
When determining how many additional records can  be  placed
in  the last block of an extended file, LIBOL fills a buffer
with that block.  It calculates the number of words used  in
the  buffer,  multiplies it by the number of bytes per word,
yielding the number of bytes used, then divides this  result
by  the  number  of  bytes  per record, expecting to get the
number of records in the last block.  The record  size  used
ignores  the  header  word  contained in each record and the
fact that all sixbit records will start at the beginning  of
the  next word.  This means the computation may be off by up
to two words per record.
  
  
[CURE]
  
  
Since the used portion of the buffer is stated in words  and
the  file table contains the record size in words, use those
fields to determine the number of records contained  in  the
last buffer.
********************************************************************************
  
  
                             EDIT 1053   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
Trying to open RMS files for I-O while dumper is backing them up  gets
error message "?LBL520 File is being modified".
  
  
  
[DIAGNOSIS]
  
DUMPER has got the files open frozen and RMS could not open the  files
for I-O.  But DUMPER is not modifying any files.
  
  
  
[CURE]
  
Edit 1053 to LBLERR to issue a correct error message.
********************************************************************************
  
  
                             EDIT 1054   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
When opening an ISAM file for Simultaneous  access  and  the
IDX  file  is  present  but  not  the  IDA  file  and  using
Declaratives to trap fatal file I-O errors and  setting  the
action  code  to force continued execution of the program, a
fatal program failure is  encountered  with  an  "unexpected
return from ENQ/DEQ" and the program execution is terminated
immediately.
  
[DIAGNOSIS]
  
When Simultaneous Update is  in  effect,  C.OPEN  elects  to
ignore  the error on the Open of the IDA file.  However upon
return to LFENQ., this routine does  not  even  test  for  a
failed Open and merely passes on in its main line of code to
call its ENQ  handler  where  the  program  failure  finally
occurs.   In the case of a failed Open, the Unavailable path
should not be taken.  However, the execution should  proceed
at  the  "Next  Sentence" in the user's program in the event
that the action code is set to 1.
  
[CURE]
  
Make LFENQ.  check for an  Open  failure  upon  return  from
C.OPEN.   Install  edit  1054  to  LSU.MAC and CBLIO.MAC and
rebuild Libol.
********************************************************************************
  
  
                             EDIT 1055   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
  
Reading a SIXBIT file results in the fatal error:
  
        The maximum record size may not be exceeded
        File FILE-ID [FILE ID] on device DSK: cannot do INPUT.
  
  
[DIAGNOSIS]
  
  
When a file is extended, the last physical block is  brought
into  memory  and examined so the buffer pointers and record
counts can be set.  In  the  sample  program  provided,  the
logical  block  of five records required two physical blocks
to contain it.  The physical block count was not  being  set
correctly  if  the  file  ended on a logical block boundary.
When LIBOL determined that the file last ended on a  logical
block   boundary,  it  reset  the  buffer  pointers  to  the
beginning of the buffer and the  number  of  records  to  be
added  to  the  new buffer, but the physical block count was
not reset to two.   This  caused  an  incorrectly  formatted
block  to be written out to the file.  When reading the file
back the fatal error occurred.
  
  
[CURE]
  
  
At the routine OPNNXX in CBLIO which requests a dummy  write
to  reset  the buffer pointers and also resets the number of
records per logical block, insert edit  1055  to  reset  the
physical block count.
********************************************************************************
  
  
                             EDIT 1056   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
     If DBMS has been added to  LIBOL/C74LIB,  DBMS  modules
are loaded even for programs that do not use DBMS.
  
[DIAGNOSIS]
  
     Each cobol  program  unit  requires  a  file  from  the
link-time  libraries  that  produces a link-time request for
symbol DBSTP$.  In a library that has DBMS  installed,  this
request is satisfied by loading the DBMS module that defines
all the entry vectors,  and  if  the  OTS  is  being  loaded
non-share, this also causes ALL of DBMS to be loaded.
  
[CURE]
  
     Replace the request for symbol DBSTP$  with  a  request
for  symbol  %DBSTP.   Add a new module %DBSTP which defines
%DBSTP==DBSTP$.  and make sure that this module gets  placed
before  the dummy (non-DBMS) DBSTP$ routine in the link-time
library.  If DBMS is loaded, make sure that this module, and
the  dummy  DBSTP$  routine are placed after the actual DBMS
entry module.  In this way, if  no  DBMS  entry  points  are
requested, the %DBSTP request will be satisfied by the dummy
DBSTP$.  And if the DBMS entry module  has  been  loaded  to
satisfy DBMS DML calls, the %DBSTP request will be satisfied
by the actual DBMS DBSTP$.  This edit requires edit  517  to
DBMS,  which  merely  changes  the  DBMS20.CTL  file to take
advantage of this new capability.
********************************************************************************
  
  
                             EDIT 1057   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
The patched version of LIBOL sets file status to  zero  when
error  occurs  if  file  is opened for input;  and the error
number for lookup error is set to 2 instead of 3 if the file
is not found.
  
  
  
[DIAGNOSIS]
  
The patched version has edit 1007, which  did  not  set  the
file status to 30 to indicate a permanent error occurred, it
also has a wrong value for error number.  Edit 1015 tried to
set the error number to the correct value but failed.
  
  
  
[CURE]
  
Edit 1057 to CBLIO to make the necessary corrections.
********************************************************************************
  
  
                             EDIT 1060   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
  
UNSTRING returns a value of zero into a numeric  destination
if the input data field contains leading spaces.
  
  
[DIAGNOSIS]
  
  
To UNSTRING a data field, the delimiter is first looked for.
After  it's  found,  the  source  characters  preceding  the
delimiter are transferred into a temporary hold  area  equal
in length to the destination field.  Padding characters will
be added and then a conversion routine called to convert the
data  and put it into the destination field.  In the example
provided the source field was alphanumeric and  contained  a
leading space, the destination field was numeric and greater
than needed.  After the delimiter was found,  leading  zeros
were  put  into  the  temporary field followed by the actual
data.  A conversion was then performed  on  the  data  which
then  consisted  of leading zeros, a space, and then numeric
characters.  The conversion halted when it found the  space,
and  left the destination field containing the leading zeros
only.
  
  
[CURE]
  
  
Since the conversion instruction will ignore leading spaces,
but  not  imbeddened spaces, edit 1060 to NSTRNG will change
the padding character to spaces.
********************************************************************************
  
  
                             EDIT 1061   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
     LIBOL generates a RECORD SEQUENCE error when  reading  a  magtape
with  the RECORDING MODE IS SIXBIT and at monitor level, a TAPE FORMAT
INDUSTRY COMPATABLE.
  
  
[DIAGNOSIS]
  
     LIBOL opens a sixbit magtape with the .IOBIN set in the OPEN  uuo
call.  The OPEN uuo sets the byte size for the magtape to 36 bit bytes
if the hardware characteristics for the magtape  drive  is  CORE-DUMP.
If  the hardware mode for the drive is SIXBIT, the file is opened with
a byte size of 6.  If the hardware mode is ASCII, the file  is  opened
with  a byte size of 7.  In all other cases, the file is opened with a
byte size of 8.
  
  
[CURE]
  
     If the device being opened has RECORDING MODE IS SIXBIT,  and  it
is  a  magtape,  set  the hardware mode to be industry compatible core
dump.
********************************************************************************
  
  
                             EDIT 1062   FOR LIBOL
  
  
  
  
[SYMPTOM]
  
  
A program using indexed files halts with a memory protection
violation.
  
  
[DIAGNOSIS]
  
  
LIBOL compares the blocking factors  of  an  index  file  as
stated  in  the  file's  statistics  block and in the user's
program.  If the factor in the statistics block is less than
or equal to the factor in the program, it is assumed correct
and used to request core to build a table  for  storing  the
addresses  of the records within a data block.  If the value
is not correct,  not  enough  core  may  be  reserved  which
results in the run-time failure.
  
  
[CURE]
  
  
Edit 1062 to CBLIO will change the test on blocking  factors
to  require that the program's factor match the index file's
factor.
********************************************************************************
  
  
                             EDIT 1063   FOR LIBOL
  
[SYMPTOM]
  
When doing Simultaneous Update with Relative files, data records are
lost without warning when being added to a file.
  
[DIAGNOSIS]
  
The user has failed to use the routine SETEOF to set the file size to
the maximum byte count before subjecting the file to Simultaneous
Updaters. As a result, when either of two updaters extends the file
the other is unable to discover this. So, when the other wishes to
extend the file, he may bury the first user's new records when he does
so.
  
[CURE]
  
Install edit 1063 to LIBOL to give a warning message if a Relative file
is being opened for Simultaneous Update and its file size has not been
set to the maximum byte count. A fatal error message is not being given
because it would force other similar executions which do not extend the
file to terminate unnecessarily.
********************************************************************************
  
  
                             EDIT 1064   FOR LIBOL
  
[SYMPTOM]
  
When doing Simultaneous Update with Checkpoint, data records may be
lost without warning when being added to a file.
  
[DIAGNOSIS]
  
The routine CHKPNT, which does the checkpointing, is executed in LIBOL
after the routine LRDEQX, which releases ENQ locks on file resources.
As a result, a race condition can exist in which two or more users are
doing overlapping accesses to the same shared file.
  
[CURE]
  
Install edit 1064 to LIBOL to eliminate this race condition in the
several places where it can exist by moving the calls to LRDEQX.
********************************************************************************
  
  
  
END OF  COBOL-20-V12B