ProfitPress Mega CDROM2 …eeware (MSDOS)(1992)(Eng)/ProfitPress-MegaCDROM2.B6I/PROG/PASCAL/TP6BUGS6.ZIP/TP6BUGS6.LST

File List | 1992-05-03 | 11.1 KB | 309 lines

TP6BUGS6.LST - Version 6 of TP 6.0 bug list

This is a completely *un*official list of bugs in TP 6.0.  There have
been (at least) two releases both numbered 6.00:  the original one in
October 1990, and another in June 1991.  You can identify the later
release by the date or by the file size:  TURBO.EXE is 325982 bytes in
the later release.  Several bugs were fixed in the second release; some
of those are marked below.  I've also heard of a 6.01 release dated July
1991, but haven't tested any of these bugs against it.

The list is very abbreviated;  if any of the items are too obscure, ask
and I'll expand on them.  One abbreviation I've been asked about:
"BASM" is the built-in assembler used in asm blocks.

Juergen Schlegelmilch sent me a very detailed list of bugs & fixes in TP
6.0 & Turbovision; the file version (TP6BUGSn.ZIP) of this list contains
all of his comments.  It also contains a few patches and fixes that I've
collected.  The file version will be available on Fidonet on the PDN
Pascal file echo area, and on Internet from garbo.uwasa.fi.  It is also
available on the PC Techniques section of CLMFORUM on Compuserve in DL
16, as TPBUGS.ZIP.

Many other people have sent me their bug lists; thanks to all of them.
In particular, some recent additions are from Arno Haket, Hans
Schleichert, Norbert Juffa, and Jud McCranie.  I know I've missed a few
recent contributors; sorry!

I'd appreciate it if anyone sent me their own bug list; I'll publicize
it on Usenet, Fidonet and Compuserve. I'd also appreciate hearing about
patches.

Duncan Murdoch

 [e-mail]   (Internet address)
 71631,122 (Compuserve)
 DJ Murdoch at 1:221/177.40.


April 17, 1992:

If the Objects unit is overlaid, calls to stream or collection Error
methods can cause system crashes.

The Drivers unit misses keystrokes if the keyboard buffer has been
reallocated.

Extensions to the Drivers unit INT09 (keyboard interrupt) handler won't
work when run within the IDE, because the interrupt handler won't be
installed.

A window that has neither the wfGrow now wfMove flags set does not
disable the cmResize command when it is active.  (A fix is available:
see TWindow.fix in TP6BUGSn.ZIP).

April 13, 1992:

The compiler doesn't detect overlapping case constants.  Only the first
matching case is executed.

The Delay() routine in CRT gives inaccurate delays on some very fast
machines. A patch (DELAY.FIX) from Borland is included in TP6BUGSn.ZIP.

The SetViewPort procedure accepts X1=X2 or Y1=Y2 (contrary to the
documentation), but clipping doesn't work properly with these settings.

TopText, CentreText, and BottomText are defined differently for
DefaultFont and stroked fonts.

Single and Double types typed constants are not correctly initialized
for absolute values smaller than 1.17xxxxE-38 respectively 2.22xxxxE-308
(their value wil be zero instead).  The lower limit of the dynamic range
should be 1.5E-45 and 5.0E-324 respectively.

App.ISqr enters an endless loop for arguments greater than 32760.

Graph.SetTextJustify doesn't properly account for descenders; they can
be clipped at the bottom of the screen.

BASM has the ranges of legal values incorrect for IN, OUT, INT, ENTER,
AAM, AAD instructions.

BASM allows references to nonexistent registers on the coprocessor, e.g.
FADD ST, ST(123456)

The IDE doesn't properly single step across AAM xx and AAD xx
instructions.

BASM doesn't do proper checking on CALLs and JMPs to near/far procedures
or absolute variables, and sometimes doesn't generate proper code, e.g.
 JMP FAR PTR AbsoluteVariable    { AbsoluteVariable is declared absolutely }
 CALL FAR PTR AbsoluteVariable

BASM doesn't flag the statements
  MOV AX, [WORD]
  MOV BYTE PTR AL,5
as nonsense.

The coprocessor emulator doesn't handle special arguments like INF and
NaN properly.  Quiet NaNs are treated as signalling NaNs, and operations
on INF return NaN.

The Random function will return 1.0 once every 2^32 calls in $N+ mode.
This value should be folded to 0.0, so that Trunc(N*Random) can't return
N.

TRect.Union and TRect.Intersect don't perform set operations.  If two
rectangles don't intersect, TRect.Intersect always returns
((0,0),(0,0)); the union of that empty rectangle with any other
rectangle always includes the point (0,0).

The Str procedure rounds nnn.5 differently than either the Round or
Trunc functions when printing reals with no decimal places.  In $N-
mode, it is inconsistent; in $N+ mode, it consistently rounds up.

Feb 2, 1992:

The Drivers.MoveBuf procedure puts junk into the buffer when Attr is 0.
(A fix is available; see MOVEBUF.FIX in TP6BUGSn.ZIP.)

The Random(N) function doesn't produce a uniform distribution unless N
is a power of 2, especially for large values of N. Not fixed.

The Val procedure can't convert the smallest longint value, -2147483648,
nor can is it handled properly by Read/Readln.  All routines work
properly on the hex equivalent, $80000000.  Not fixed.

The base type for a pointer type definition can't be a qualified
identifier. Not fixed.

The following TV bugs are more fully described (and fixes are given!) in
the file TVBUGS in TP6BUGSn.ZIP:

TFileEditor.NewBuffer allocates 0 bytes, but does not set BufSize
to 0.

If the search string given to Editors.IScan nearly matches the last
chars in the text, the routine runs over the end of the text, eventually
replacing text.

TMemo cannot coexist in a TDialog with other objects having scrollbars,
because TEditor.HandleEvent consumes all cmScrollBarChanged events.

TEditor.InsertBuffer does not reset the block marker.  If a call to it
is followed by a backspace, the block is deleted.

If StdDlg.NoWildchars is called with an empty argument, it overwrites
the stack.

THelpTopic.AddCrossRef allocates memory and then forgets to use it.

THelpIndex.Position returns garbage values if called with a negative
argument.

THistory.Draw uses characters 221 and 222 which are not standard outside
codepage 437.

December 3, 1991:

TP doesn't handle the denormal exception properly on an 8087 or 80287,
so underflow sometimes creates non-zero results.  Not fixed.

Compile time evaluation of the ROUND() function is different than
run-time evaluation in $N+ mode on numbers ending in .5.  Not fixed.

Arithmetic on the REAL type is sometimes inaccurate in the last few bits
of the result.  Not fixed.

The coprocessor emulator doesn't properly emulate the FINCSTP and
FDECSTP instructions in BASM.  (These aren't used by compiled Pascal
code.)  Not fixed.

The coprocessor emulator doesn't respect the precision control bits. Not
fixed.

The SEG operator in BASM doesn't work on variables declared in the stack
segment. Not fixed.

Referencing fields of records in BASM in a WITH block doesn't add the
base address. Not fixed.

BASM doesn't generate correct code for instructions like MOV AX,[WORD
PTR BX]. (This is TASM Ideal mode syntax; MASM syntax MOV AX,WORD PTR
[BX] is fine, as is MOV AX,[WORD PTR DS:BX].)  Not fixed.

The Graph.TextWidth function returns incorrect results when fonts are
scaled.

October 31, 1991:

The instruction "IMUL AX, 16" (valid on an 80186+) is rejected by BASM.
The equivalent "IMUL AX, AX, 16" does work (in $G+ mode).  Not fixed.

October 26, 1991:

If a byte variable X = 0, then (X-1) DIV 8 produces a zero divide error.
Fixed June 91.

October 23, 1991:

Hitting tab on a help topic with no crossrefs will crash the help
system.  A patch to HELPFILE.PAS is available from Borland.

October 17, 1991:

The coprocessor emulator does not respect the rounding bits internally
or when storing results to Single or Double variables.

October 11, 1991:

Segments in external .OBJ files with no class names (e.g. ABSOLUTE
segments) will crash TURBO, TPC or TPCX.  Not fixed.

September 16, 1991:

The IDE will give incorrect matches when doing word searches on words at
the start of lines (e.g. "Const" matches "onst").  Fixed June 91.

The compiler won't allow the optional identifier in a variant record to
be an already defined scalar type identifer. Not fixed.

The compiler gives spurious errors if you nest loops very deeply (e.g.
256 deep).


August 16, 1991

Dangerous bugs:

You can use typed constants and variables in constant expressions,
giving garbage values.  Fixed June 91.

The coprocessor emulator doesn't check properly for internal stack
overflow; it just wraps around and trashes the top of the stack segment.
Not fixed.

Succ(Seg(v)) = Seg(v)  and Pred(Seg(v)) = Seg(v) !  Fixed June 91.

TBufStream writes garbage to the stream sometimes.  (A patch for this is
included in TP6BUGSn.ZIP as TBUFSTRE.FIX.).  Not fixed.

IN doesn't properly handle out of range values.  (What it does is to
demote out of range scalars to byte, ignoring any high bits.) Not fixed.

Referencing fields of records in inline machine code sometimes makes the
compiler go quietly insane.  Not fixed.

 Other bugs:

Coprocessor errors are often reported at the wrong address, especially
if the error occurs on the first pass through. Not fixed.

GetDir returns a root directory for non-existent drives.  (This is
almost as documented.)

Horizontal text justification to the right side doesn't work if the
writing direction is vertical; this is only in the DefaultFont.

Intr and MsDos don't always work properly in V86 mode under Windows 3,
because they don't use the INT instruction.

The TVEdit unit can't input some digits.  (A patch for this is
included in EDITOR.FIX in TP6BUGSn.ZIP.) Fixed in June 91.

^T in TVEdit sometimes deletes more than the next word.  (This is by
design, but inconsistent with the design of the IDE.)

^QF followed by ^L followed by Space deletes words in TVEdit.  (This is
also by design.)

TProgram doesn't clean up memory properly on exit.

TFileViewer's FileName field is never disposed.

The IDE forgets your snow check setting if you jump to DOS and exit
back.

Writeln wipes out the status word in the coprocessor. Not fixed.

You can allocate an array bigger than 65528 bytes on the heap; it may
wrap around and trash whatever precedes it.  Not fixed.

The coprocessor emulator doesn't propagate NaNs the same way as a real
coprocessor does. Not fixed.

The string conversion routines sometimes print NaNs as INF.

Complicated expressions involving Reals compiled $N+ will unnecessarily
overflow the coprocessor stack (and trigger the dangerous error above).
Not fixed.

The $G+ option makes the inline assembler use inline floating point
code, rather than the CD forms that allow emulation. Not fixed.

Graph3.Plot leaves a word on the stack; if you call it repeatedly, you
may get a stack overflow; if you call it from a routine that doesn't
have any locals, you're dead.  Fixed in June 91.

Some weird problems with F7 Trace; sometimes it just refuses to do it.

There are certain illegal Extended encodings that crash the debugger in
TURBO. Not fixed.

Range checking shouldn't apply to the limits in a for loop.

Method arguments shouldn't be at the same scoping level as method names.

The compiler exports private identifiers in .TPU files (though you can't
use them).