The canonical list of +3 oddities
(first posted on comp.sys.sinclar on 15th March 1994)
Note: This list applies to ROM version 4.0. The version number may be
seen from the system test message (see 1(b) below). I'm told that most +3
machines have version 4.0 while most +2A machines have version 4.1. I have
never examined this updated ROM and would be interested to know how many of
the following things apply to it also.
1. Strange Features
(a) In ROM1
- At 2AA8 is the message "You should never see this". It is right,
because error messages are generated from a message address table,
and the address of this message isn't in that table.
- At 2AC1 is the message "Hello there!"
- At 3283 is the name of the temporary file used by COPY:
"M:VAXNSUZ.$$$". Don't save a file of this name and expect
it to stay there while you copy a file from disk to disk!
- At 35C4 is the COPY RANDOMIZE routine. Between the time you press
enter after typing this command and the appearance of the red error
cursor, you have about a quarter of a second in which to press
simultaneously the keys C, J and L. Don't use this command when
you have valuable data in memory!
(b) In ROM0
- From the test screen, press keys Q, A, Z, P, L and M together.
This brings up the system test (which starts at 21DF and occupies
most of the rest of the ROM).
- From the test screen, press keys E, U and A together and then
play a tape (this routine is at 22D0-22FD).
- The test screen can be ended by pressing B and V together. The
machine resets.
- The editor has a bunch of editing keys (table at 05E6):
TAB up 10
AT down 10
CODE left word
VAL$ right word
INKEY$ top
RND bottom
FN start of line
PI end of line
SCREEN$ delete char right
COS delete word right
TAN delete word left
VAL delete line right
LEN delete line left
POINT screen (same as "Screen" in +3 basic menu)
2. Strange Feechurs (aka buglets)
(a) In ROM1
046D: The erase message goes to a recently-used stream, not always to
the lower screen. For example, PRINT : ERASE "*.bak" prints
the "Erase *.bak?" message on the main screen. Also, the
lower screen is cleared only if you press Y. If you press N
then the message stays there even if it was printed in the
right place.
0769: "SUB +40" should be "SUB +3F". The effect of this is that the
CAT command will only give up to 63 file names, even if there
are more files on the disk.
083E: The jump is made to an incorrect location if the load-bytes
routine is called to verify. This is mostly harmless, because
the BASIC VERIFY command doesn't attempt to verify at all.
0998: A file name such as "T:file" is correctly diverted to tape, but
does not have its leading "T:" stripped off.
1236: CLEAR may leave the GOSUB stack without an end marker if it was
non-empty before the CLEAR [example 1].
1883: A volume instruction in a PLAY command attempts to output the
volume to the sound chip even for channels 4 to 8 [example 2].
It should not output the volume at all, since that will be
done at 1C74. Not only is it one time-unit early to output
the volume [example 3], but it also means that PLAY "V10&"
and PLAY "&","V10" both make spurious sounds.
189B: The U instruction in a PLAY command sets the stored volume
number to 1F even for channels 4 to 8. The programmer appears
to have intended to output this value to the sound chip, but
forgotten. This is a bonus (see above). Note that notes sent
to a MIDI channel will be of volume 15 after a U instruction,
even on channels 4 to 8.
1BFA: A key-up signal is sent to the MIDI port even after a rest.
This signal will be for the most recently played note in the
current string. For instance, PLAY "Y1a&","Y1N9a" will sound
only for a minim.
1F08: The delay to wait for stop bits after an input is (BAUD)-7.
This value is not checked for overflow, so this delay
effectively prevents communication at 19200 baud, which
would otherwise be OK.
2017: TVPARS is set to accept two parameters after a colour control
code, rather than just 1. This means that the A in LPRINT INK
3;"A" is absorbed. This routine also exits with carry and
zero reset unless the character was an INK control (see 2B60
and example 4).
2020: Any control byte is stored in TVDATA, which might change in
certain unusual circumstances [example 5].
202B: When all the bytes of a colour control routine have been
received, this routine returns with carry and zero both
reset (see 2B60 and example 4).
219C: A "LD (TVFLAG),1" should probably be "LD (DFSZ),1" (harmless).
21FA: COPY capitalises the drive letter in each file name. This
has a permanent effect if the file name was written in the
BASIC program or in a variable.
25D6: The error reporting routine contains HALT, as does the normal
spectrum one.
2624: If the error number in 23610 is out of range, the error
reporting routine crashes because it does not check the range
before using it as an index to a message address table.
2B09: There is a special case for RS232 input which occurs when the
48K BASIC editor is active, for example in INPUT #3;a. It
traps both input and output, but does not check before entering
an input loop (this is probably OK since the editor does not
print anything before this loop gets control). However, before
making a return it calculates the wrong value to be output to
port 1FFD, namely (BANKM)|#10. This will cause a crash.
2B60: The routine which is called for both input and output on
channel "P" checks the flags before returning. If both
carry and zero are reset, then it reports "End of file".
This should only happen on input (see 2017 and 202B).
2DF3: The copy-disk-to-disk routine appears to check both source
and destination to see that they are not "M:" (if they are
then they will be treated as normal files and hence give a
bad filename error). However the test is done in upper case
on a lower case letter and so always fails. The routine opens
source and destination drives and then, assuming that "M:" is
not involved, it tries to open the temporary file unless the
RAM drive is too full. This gives "File already in use" -
but apparently not before the contents of the destination
drive have been erased. In any case, copying between "M:"
and "A:" should not be allowed because they have different
formats.
2ECD: A buffer starting at ED11 and of length 0800 is used by the
COPY command. This is almost certainly supposed to be in
page 7, but instead it is in page 0 where it will overwrite an
area of user memory. If RAMTOP (and hence the machine stack)
is in that area when the COPY command is executed then the
machine will crash. This error also occurs at 2F12, 2F61,
2FA7, 3024 and 3043.
3150: The length of the string ".HED#" (where # is an end marker,
code FF) as used in COPY TO SPECTRUM FORMAT is given as 4.
This may result in "Bad filename" unless the original file
had a file extension of the same length (in which case the
original name's end marker is in the right place for the new
name).
(b) In ROM2
016C: BANK678 is not initialised. It should be set equal to #10
so that the STROBE of the centronics port is set high. Some
printers cope with an inverted STROBE, but some don't.
079E: The key beep routine (which is also used for the longer beep
when a key press is not accepted) uses 0C80 for the frequency,
whereas the 48K spectrum uses 00C8. This means that the value
stored in PIP has 16 times as much effect on the note duration
in editor mode as it does during the command INPUT.
0A27: The "line has altered" flag is not cleared when a null line is
entered (for example, if you type a space then delete it and
press enter). This means that if you move the cursor up or
down on to a program line then the program line is considered
to have been changed.
0E13: The "alternate cursor column" is corrupted after a syntax
error. If the cursor was moved from its original position
into its error position then the alternate column is zero and
the next up or down movement will cause the cursor to jump to
the left-hand side of the screen. If it stayed in the same
place then the alternate column is random and so moving up
or down will cause unpredictable results [example 6].
0E6F: "LD HL,(F9DB)" should be "LD HL,F9DB". This causes several
undesirable effects with lines that go off the top of the
screen (this is only a problem when using the lower-screen
editing mode). One possible effect is that one or more
complete lines are deleted from your input when you press
enter. Another is that there is a delay before the line is
successfully entered and when the line is displayed there is a
beep and a red cursor on the line.
10E4: After del-word-left and del-word-right, an incorrect "alternate
cursor column" is stored.
142C: "RET NC" should be "RET" (harmless).
14AF: On the 48K spectrum, leading spaces are only printed for token
codes >=197 starting with a letter. On the +3, all tokens have
spaces before and after.
153D: A program which has a line 0 will display that line ad
infinitum if the stored current line number is zero. This
leads to an infinite loop if you try to add a program line.
156E: A line number of 0 is displayed as all blanks.
1A95: When calculating the highest theoretical line number for
"renumber", the program ignores the carry flag. If the line
increment is unusually large this might result in an erroneous
renumber being accepted and carried out. The result of doing
this will probably crash the machine (in the routine at 1C12)
if it is renumbered again.
1B48: If a GOTO cannot be renumbered, the routine just returns
instead of warning the user (similarly at 1B63).
1B6C: When the renumber routine fetches the number from after a
GOTO, it ignores the flags from the "get integer" routine.
Thus, if the number is over 65535 then the routine will fetch a
(seemingly) random value. If this value, or indeed any number
following GOTO, is too large (for example, try GOTO 65535)
then the program will crash in routine 196E, LINE-ADDR in the
spectrum ROM because this routine does not cope with too-big
line numbers.
1B73: The programmer intends a GOTO which points past the end of
the program to be renumbered to 9999. If the program has no
variables, this will happen. Otherwise, such a line will be
numbered to the lowest line number not used by the program.
This is due to the test "CP #80:JR NZ,notend" to test for
the end of the program, which should be either "AND #C0:
JR Z,notend" or "CP #28:JR C,notend".
1F71: "CALL 202F" should be "CALL 1FA9". If the character ">" or "<"
is followed (without a space) by a sequence of letters and then
anything except "$", "#" or a space and some more letters, then
the character and the sequence of letters is exchanged when
you press enter. For example, "IF foo1<bar2 THEN" changes
to "IF foo1bar<2 THEN".
20AD: Following a syntax error, the editor goes through some complex
manoeuvres to translate the spectrum's error marker (positioned
at (XPTR)) into a position within the line you typed. Then it
louses it all up by using (XPTR)-(ELINE) instead - except on
the rare occasion when that value is further to the right
than the position it calculated. This unnecessary piece
of code means that the error marker is almost always wildly
inaccurate.
20D3: An "INC DE" is missing. This means that, on the rare occasion
when the bug at 20AD doesn't mess up the position of the error
marker, the error marker is displaced rightwards one character
for each hidden floating-point form which was inserted to the
left of it while the line was being checked.
Stop press... Matthew Wilson has just pointed out that the
"renumber" routine incorrectly renumbers "FORMAT LINE xxxx"
because it expects all "LINE" keywords to be part of "SAVE ...
LINE xxxx".
Examples
1. 10 GOSUB 100 This program does not fail at line 110, as one
20 RETURN would expect, but instead keeps going and, when it
100 CLEAR 32767 reaches line 20, tries to return to a non-existent
110 RETURN statement. It also alters memory locations 32768
and 32769, which should be forbidden.
2. 10 PLAY "X8192W0U1&9c" Lines 10 and 20 should sound
20 PLAY "X8192W0U1&9c","","","","V15" identical, but they don't since
V15 affects the "X" register.
3. 10 PLAY "T240N7cccc" Between each pair of notes there is a
20 PLAY "T60N3cccc" silent gap of one time-unit, which is
30 PLAY "T60N3cV15cV15cV15c" about 1/24 of a crotchet. Thus the
notes in line 10 are much closer together
than those in line 20. However, those in line 30 are joined together
because the V command before each note makes it sound one time unit
early.
4. 10 LPRINT CHAR$ 16;CHR$ 1;"ABC" Each line reports an error. Lines 10
20 LPRINT INK 1;"ABC" and 30 say "End of File" while line 20
30 LPRINT PAPER 1;"ABC" says "Nonsense in BASIC". The reason
for this latter report is that the
INK control causes 2017 to set the carry flag, which remains set
through 1F89 (which receives the second control character), and
an oversight in the Spectrum ROM at 2210 causes it to think that
the colour control routine did not find a colour control. It then
searches for an expression but finds a ';' and reports the error.
5. 10 LPRINT CHR$ 16; The control byte 16 is stored in TVDATA, but
20 PRINT AT 0,0; so is the control 22 from PRINT AT. When the
30 LPRINT CHR$ 0;"ABC" earlier control is finished, it is treated as
PRINT AT instead of INK.
6. Type a syntax error which is just over one line long on the screen.
Press enter twice. Press the down-arrow. The cursor probably jumps
into the middle of the screen. Press an arrow again and the computer
will probably crash.
Ian Collier
Ian.Collier@comlab.ox.ac.uk | imc@ecs.ox.ac.uk