home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.robelle3000.ai 2014
/
2014.06.ftp.robelle3000.ai.tar
/
ftp.robelle3000.ai
/
papers
/
nmdebug.txt
< prev
next >
Wrap
Text File
|
1994-04-28
|
36KB
|
902 lines
The MPE/iX System Debugger
By David J. Greer, Robelle Consulting Ltd.
Copyright Robelle Consulting Ltd. 1991,1994
Permission is granted to reprint this document (but not
for profit), provided that copyright notice is given.
MPE/iX comes with a powerful debugger. For those of us who have
struggled with Debug/V, there are many great features to look
forward to. But, like all new software, there is a learning
curve in understanding the new MPE/iX debugger. Attempting to
find the dozen or so most useful features in the three-inch stack
of paper called the System Debugger Reference Manual is
impossible, unless you have three spare months. In this article,
I intend to summarize the features I've found most useful.
I am indebted to Stan Sieler of Allegro Consultants who taught me
much of what is presented in this article. To obtain the maximum
benefit from this article, you should try all of the examples
that are presented. Having both a CM- and an NM-program
available that calls the FWRITE intrinsic will make following the
examples that much easier. If you are going to be a big user of
Debug/iX, then you really have to have the reference manual.
Order part #32650-90013, System Debug Reference Manual.
Native-Mode or Compatibility-Mode
HP was very kind to write not only a nice debugger for
native-mode programs, but include features for debugging
compatibility-mode programs too. When I first attempted to debug
a CM-program, I got so confused that I returned to my classic HP
3000 where at least I knew all the command names. The next
section will show Debug/iX commands for our old favorite Debug/V
commands.
Debug/V Versus Debug/iX
Just like Debug/V, you can invoke Debug/iX by including the
keyword "Debug" on the :Run Command for your program. Debug/iX
responds with its "CM" prompt:
:run testprog;debug
CM DEBUG Intrinsic: PROG %6.3542
cmdebug >
If you have a Pmap (or Robelle's Qmap), you can set a breakpoint
just as you would in Debug/V -- using segment.offset:
cmdebug >B 0.45
Anyone who has struggled with Pmaps knows how convenient it would
be if the system debugger took advantage of the FPMAP information
stored in program files. With this information it should be
possible to set a breakpoint by procedure name. Debug/iX lets
you either set a symbolic breakpoint at the first logical
instruction in your procedure or at the more useful entry-point:
cmdebug >B open'input'file {first instruction}
cmdebug >B ?open'input'file {? implies entry point}
cmdebug >B open'input'file+255 {octal offset from first}
cmdebug >c {Continue = Resume}
Breakpoints
In the last ten years, I have probably lost over a month of time
from one horrible default in Debug/V. In Debug/V, the Break
Command would only break the first time it encountered the
breakpoint, unless you added ":@" to the Break Command. In
Debug/iX the default is to always break. Occasionally, you only
want the breakpoint to be invoked once. Use ",-1" after the
break location to have the breakpoint removed after one
occurrence:
nmdebug >b ?open'input'file,-1 {break once}
Clearing Breakpoints
In Debug/V, the Clear Command disables breakpoints that have been
set with the Break Command. In Debug/iX, use the BD (Breakpoint
Delete) Command to remove breakpoints:
nmdebug >bd {You will be asked for which one}
nmdebug >bd @ {Delete all breakpoints}
Setting a "Return" Breakpoint
One of the most useful breakpoints is the one immediately after a
procedure call. Suppose that your program calls the procedure
extract_ready. You want to know the result of extract_ready, so
you would like a breakpoint in the calling code immediately after
the call to extract_ready. You do the following:
cmdebug >b ?extract_ready
cmdebug >c
.
.
. {break at extract_ready}
cmdebug >lev 1;b p,-1 {-1 means only break once}
cmdebug >c {continue execution}
The "lev 1" goes back to the previous logical level in the
calling sequence (use "tr,d" to see a complete traceback). The
"b p" sets a breakpoint at the compatibility-mode program
counter. The "lev 1" places the program counter at the
instruction after the one that called the current procedure. The
",-1" tells Debug/iX to execute the breakpoint once and then
throw it away. Note that it's safe to use this breakpoint
anywhere in the extract_ready procedure -- not just at the
beginning.
What if we are in native-mode code (e.g., FWRITE)? Our return
breakpoint won't work, since we called FWRITE from compatibility-
mode. To set a return breakpoint in this case, first switch into
cmdebug:
cmdebug >b ?FWRITE
cmdebug >c
.
.
. {break at NM FWRITE}
nmdebug >cm {switch into CM}
cmdebug >lev 1;b p,-1 {set return breakpoint}
cmdebug >c {continue execution}
Abort Command -- Getting Out of Debug
You can terminate your program with the Abort Command. Use this
any time that Debug/iX is prompting for commands. The Debug/iX
Abort Command is similar to the Debug/V E@ Command.
Displaying Values
When I first used Debug/iX, I became totally confused about how
to display the usual DB-, Q-, and S-relative values. It turns
out to be very simple. In Debug/V the Display Command takes the
register as a parameter. In Debug/iX there are separate command
names for displaying values relative to each register. Here are
the Debug/V and Debug/iX Display Commands:
Debug/V Debug/iX
D DB DDB
D S DS
D Q DQ
The Debug/iX Display Command has a count as its second parameter
(just like Debug/V), but the display attribute is different.
Here is the comparison:
Debug/V Debug/iX Description
,I ,# or D Decimal
,O ,% or O Octal {default in CM}
,H ,$ or H Hexadecimal {default in NM}
,A ,S Ascii/String
Instead of S, you can also use A for displaying string values.
The A-option is closer to the A-option of Debug/V, but we find
the S-option more useful.
By default, the S-option displays all characters you request and
only displays the virtual address of the string once. If you
want to see as many characters per line as possible, with each
new line starting with the virtual address of the characters
displayed, use this command:
cmdebug >dq 104,200,s,e {"e" shows addresses}
Symbolic Machine Code
Our list above doesn't show you how to display the actual
run-time machine instructions (commonly called decompiling).
That's because Debug/iX has many excellent features to
symbolically display code. While you can use the DC (Display
Code) Command to show symbolic code, we have found a better
method -- windows.
Symbolic Traceback
The Debug/V Trace Command was almost useless. You had to
manually work through the segment numbers and offsets to figure
out the true procedure names. The Debug/iX Trace Command
produces a proper symbolic traceback of procedure names.
You can also use the traceback to observe switches from
native-mode to compatibility-mode. For example, if you have SM
capability you can set a breakpoint in any system SL or system XL
routine. KSAM files come in two flavors: CM and NM. If you
access a CM KSAM file from a NM program, MPE/iX calls the CM
FWRITE intrinsic. You can easily prove this to yourself by
setting a breakpoint in the CM FWRITE intrinsic:
:run testprog;debug {NM program to read CM KSAM file}
nmdebug >cmdebug {switch to CM}
cmdebug >b ?FWRITE {question-mark for entry point}
cmdebug >c {continue execution}
.
.
. {note the ",d" on the TR Command}
cmdebug >tr,d {traceback showing switches}
Compatibility-Mode Windows
The WON Command is the real power of Debug/iX. WON is short for
Windows On. When you turn windows on, the top portion of the
screen is reserved for a symbolic display of the currently
executing code, another portion displays register and/or stack
values, and the bottom of the screen is used to enter commands.
This is a very powerful feature.
CM Window Example
The following is an example compatibility-mode window. We first
set a breakpoint, continue to that breakpoint, and finally we
turn windows on.
cmdebug >b ?input'command {break at the entry point}
cmdebug >c {continue execution}
.
.
.
cmdebug >won {turn windows on}
The top three lines of the display show the register information:
R % Regs DB=001200 DBDST=001632 X=000002 STATUS=(mITroC CCG 007) PIN=051
SDST=001627 DL=177450 Q=023620 S=023620 CMPC=PROG 000006.006711
CIR=035004 MAPFLAG=0 MAPDST=000000
FcmP % PROG 6.6711 (?) SETUP CSTX 7 Level 0
006707: INPUT'COMMAND+%437 031031 2. PCAL ?ERRX
006710: INPUT'COMMAND+%440 032000 4. SXIT 0
006711: [1]> ?INPUT'COMMAND 035004 :. ADDS 4
006712: INPUT'COMMAND+%442 171700 .. LRA S-0
006713: INPUT'COMMAND+%443 051401 S. STOR Q+1
006714: INPUT'COMMAND+%444 035023 :. ADDS %23
006715: INPUT'COMMAND+%445 041401 C. LOAD Q+1
Q % (DB mode) QDST=001627 Level 0
023610: 000000 047420 061006 000006 177600 D000002 D00364
023620:Q>D000014 <S
023630:
S % (DB mode) SDST=001627 Level 0
023610: 000000 047420 061006 000006 177600 D000002 D00364
023620:Q>D000014 <S
Commands
%47 (%103) cmdebug >
For most of us, only the DL=, Q=, S=, and X= values are
interesting. If the DBDST and the SDST (the DB- and S- data
segments) are different, you are in split-stack mode. Line four
shows that we are currently at location 6.6711 in the program.
The PROG would change if the breakpoint was inside an SL. Next
we see seven instructions. The "[1]" means breakpoint number 1.
The ">" symbol next to "?INPUT'COMMAND" shows the next
instruction to be executed. The bottom of the display shows the
values around the Q- and S- registers. In our example, the Q and
S registers are the same so the Q- and S-displays are identical.
Finally, you are prompted for more Debug/iX commands.
Single-Stepping
One other command adds a lot of power to windows: S --
single-stepping. The S Command executes the next instruction,
then returns control to Debug/iX. After the execution, register
and stack values are updated and any changed values are
highlighted. Because the compatibility-mode window shows the top
few words of the stack, you can often get an instant picture of
what is going on. Here is the first window after executing one
single-step:
cmdebug >s {single-step}
R % Regs DB=001200 DBDST=001632 X=000002 STATUS=(mITroC CCG 007) PIN=1
SDST=001632 DL=177450 Q=023620 S=023624 CMPC=PROG 000006
CIR=171700 MAPFLAG=0 MAPDST=000000
cmP % PROG 6.6712 (?) SETUP CSTX 7 Level 0
006707: INPUT'COMMAND+%437 031031 2. PCAL ?ERRX
006710: INPUT'COMMAND+%440 032000 4. SXIT 0
006711: [1] ?INPUT'COMMAND 035004 :. ADDS 4
006712: > INPUT'COMMAND+%442 171700 .. LRA S-0
006713: INPUT'COMMAND+%443 051401 S. STOR Q+1
006714: INPUT'COMMAND+%444 035023 :. ADDS %23
006715: INPUT'COMMAND+%445 041401 C. LOAD Q+1
Q % (DB mode) QDST=001632 Level 0
023610: 000000 047420 061006 000006 177600 000002 00364
023620:Q>000014 000002 006712 062007 d
023630:
S % (DB mode) SDST=001632 Level 0
023610: 000000 047420 061006 000006 177600 000002 00364
023620:Q>000014 000002 006712 062007<S
Commands
%47 (%103) cmdebug >
The ">" symbol has moved forward by one instruction. The
register values have been updated and the top of stack has
changed because we added four to the S-register.
Set CRON
This sounds like the title of a futuristic movie, but when
combined with single-stepping it can be very powerful. Once you
start using the S (Single-Step) Command, you'll find yourself
typing it a lot, especially when debugging NM programs where you
have a lot more instructions per source code statement.
Fortunately, the Debug/iX designers already thought of this.
When you Set CRON, hitting Return tells Debug/iX "execute the
last command that I typed". This is most useful when your last
command was S, but it applies to any command:
cmdebug >set cron {Return = last-command}
cmdebug >s {single-step}
cmdebug > {another single-step!}
cmdebug > {and one more}
cmdebug > {and so on}
Multiple Steps
While single-stepping is useful, it can be very slow. You can
step through a program faster using multiple instructions for
every step. The following example shows how to step through
every seven executed instructions. Note: you must have a space
after the Step Command and before the number of instructions to
execute (e.g., "S7" is invalid):
cmdebug >set cron {Return = last-command}
cmdebug >s 7 {execute seven instructions}
cmdebug > {another seven!}
cmdebug > {and seven more}
cmdebug > {and so on}
Native-Mode Debugging
Much of what has been discussed applies to native-mode. There
are a few minor differences:
1. You don't need to specify Fpmap (or any other magic parameter)
on the :Link Command. Procedure name and location information
is automatically included in all NM program files.
2. Since the first instruction of a procedure and its entry point
are the same, you never need to use a question mark. If you
happen to type a question mark, Debug/iX may not print an
error. In this case, you will have set a breakpoint in a stub
procedure. Since you almost never want to do this, it's
important to remember not to type the question mark before the
procedure name.
3. In most programming languages, any separators (e.g.,
apostrophes) used in procedure names will now become
underbars.
Here is our previous breakpoint example in native-mode:
nmdebug >b open_input_file {break at procedure entry}
Case Sensitivity
It is easy to see that portions of MPE/iX were affected by UNIX
and the C programming language. In UNIX and C, case is
significant (i.e., upper-case and lower-case are not the same).
When setting breakpoints in native-mode code, it is important to
remember this. Most MPE/iX routine names are in upper-case. The
most well-known exceptions are all of the IMAGE and VPLUS
intrinsics which are in lower-case. The following example
results in a Debug/iX error:
nmdebug >b fwrite {not found; lower-case}
Switching Modes
Sometimes you want to switch between CM-debug and NM-debug. For
example, the NM-FWRITE intrinsic calls the CM-FWRITE intrinsic
for certain types of files (e.g., circular). These commands
would set breakpoints in both the CM- and NM-FWRITE intrinsics:
nmdebug >b FWRITE {NM-FWRITE breakpoint}
nmdebug >cm {switch into cmdebug}
cmdebug >b ?FWRITE {CM-FWRITE breakpoint}
cmdebug >nm {switch back into nmdebug}
nmdebug >c {continue execution}
Native-Mode Windows
The WON (Windows On) Command is just as powerful in native-mode
as in compatibility-mode. The display is different -- instead of
the old familiar DB, S, and Q registers, there is a strange group
of 32 "general-purpose" registers. The code looks a lot
different too -- those famous RISC instructions instead of our
old faithful Classic 3000 ones.
NM Window Example
We will show an example native-mode window, by setting a
breakpoint for the FWRITE intrinsic:
nmdebug >b FWRITE {requires SM capability}
nmdebug >c {continue execution}}
.
.
.
nmdebug >won {turn windows on}
GR$ ipsw=0006fe0f=jthlnxbCVmrQPDI priv=0 pc=0000000a.004a5fc0 pin=0000007a
r0 00000000 40100e20 004aee30 00000001 r4 c0000000 0000ffff 4033292a 00000000
r8 00000001 00000009 00000004 4034a880 r12 00000000 00000000 00000000 00000000
r16 00000000 00000000 00000000 c0000000 r20 c0000000 00000001 85240000 00000314
r24 40332604 000000d0 00000001 c0202008 r28 00000001 ffffffff 4034afd8 004aee30
nmP$ SYS a.4a5fb8 NL.PUB.SYS/FSPACE+$5a4 Level 0,0
004a5fb8: FSPACE+$5a4 e840c000 BV 0(2)
004a5fbc: FSPACE+$5a8 4fc33d31 LDWM -360(0,30),3
004a5fc0: [1]> FWRITE 6bc23fd9 STW 2,-20(0,30)
004a5fc4: FWRITE+$4 6fc30340 STWM 3,416(0,30)
004a5fc8: FWRITE+$8 6bc43cc9 STW 4,-412(0,30)
004a5fcc: FWRITE+$c 6bc53cd1 STW 5,-408(0,30)
004a5fd0: FWRITE+$10 6bc63cd9 STW 6,-404(0,30)
Commands
$7 ($1d) nmdebug >
The first line contains general information about the process
(e.g., the pin number). The pc= is the program counter (notice
it's a full 64-bit address in space.offset format). Lines two
through four of the display show all 32 general-purpose
registers. The fifth line shows where the first instruction in
the window is located (in NL.Pub.Sys @ FSPACE+$5a4). The
native-mode instructions are shown, along with the breakpoint
number "[1]" and the next instruction to be executed is marked
with the ">".
There are two commands that can be a big benefit in examining the
code "around" a breakpoint: PB (Program Back) and PF (Program
Forward).
Paging
Debug/iX windows have to display all their information in the
twenty-four lines on a standard terminal screen. By default, the
size of the symbolic instruction list is seven instructions.
Especially when you are single-stepping through instructions, it
is very useful to see the previous seven instructions or the next
seven. The PB (Program Back) Command displays the previous seven
instructions and the PF (Program Forward) Command shows the next
seven. While seven instructions is the default, there are
commands to change the size of the program window. If you have
changed the size, Program Back and Program Forward adjust
themselves to the new window size.
nmdebug >pf {program forward}
nmdebug >pb {program back}
PL Command
If you want to change the number of program instructions on the
screen, use the PL Command (Program List). The PL Command
assumes that the number of lines you want is in the current base.
Therefore, PL 10 means 16 instructions in NM Debug and 8
instructions in CM Debug. To get around the problem, we always
specify the number of instructions in decimal:
nmdebug >pl #10 {show "ten" instructions}
Native-Mode Procedure Parameters
Long-time users of Debug/V know how to anticipate where procedure
parameters will be located. For example, if we had a procedure
with this declaration:
integer procedure convint(buf,len); !result = Q-6
value len;
integer len; !len = Q-4
byte array buf; !@buf = Q-5
In CM-debug, we would look at the parameters as follows:
cmdebug >dq -6 {result of Convint procedure}
cmdebug >dq -4 {length of buffer}
cmdebug >dq -5 {address of buffer}
Q-%5 % 000104 {must use this value below}
cmdebug >ddb 104/2,10,s {print actual buffer contents}
Notice that we had to divide the value at Q-5 (i.e., %104) by
two, since the buffer was passed as a byte address. In
native-mode, this irritation disappears (except for those using
SPLash! to emulate Classic byte addressing).
Native-Mode Calling Conventions
With the power to set breakpoints symbolically, by just knowing
the name of a procedure, there is even more incentive to be able
to guess the location of procedure parameters. NM procedures are
allocated registers for the first four parameters, but they are
allocated left-to-right -- the opposite of CM procedures. The
first parameter is assigned to Register-26, the second to
Register-25, the third to Register-24, the fourth to Register-23,
and any remaining parameters are stored on the NM stack. The
return value is in Register-28 (and Register-29 for 64-bit
values). For native-mode, you would think of the declaration for
Convint as:
integer procedure convint(buf,len); !result = R28
value len;
integer len; !len = R25
byte array buf; !@buf = R26
If you have windows on, the 32 general-purpose registers are
always displayed. The only problem area is the buffer parameter:
nmdebug >b convint {note lower-case}
nmdebug >won {windows on}
nmdebug >c {continue execution}
.
.
. {debug breaks @ convint}
nmdebug >=r25 {display the length}
nmdebug >dv r26,10,s {display virtual uses the contents}
{of register 26 as an address}
Variables
Debug/iX contains a programming language. We won't try and cover
all of the features of this language, but variables are so
powerful that they are worth knowing about. In our example with
the Convint procedure, suppose that the buffer you are passing to
Convint is a global variable. Setting the breakpoint at Convint
gives you a convenient method to find and save the address of
your buffer so that you can use it at any breakpoint.
nmdebug >var buf_var=r26 {save address of buffer}
nmdebug >dv buf_var,10,s {display buffer contents}
nmdebug >c {continue execution}
.
.
. {sometime much later ...}
nmdebug >dv buf_var,10,s {display the buffer contents}
The final Display Virtual Command displays the contents of the
buffer using the address that we saved. When the breakpoint
takes place, we may have no convenient way of finding the program
variable that has the address of our buffer. Because we have
saved the address in the Debug/iX variable "buf_var", we display
the buffer contents without knowing where the address is stored.
Virtual Addresses
So far, we have assumed that all addresses are 32-bits. In
MPE/iX, addresses are actually 64-bits. Debug/iX shows these
addresses as space.offset. If you are working with mapped files,
you will find that the full 64-bit address suddenly becomes
important. The following example opens a file with mapped
access, saves the virtual address of the file into a variable,
and then displays the actual contents of the file.
nmdebug >map "file1.suprtest" {open an mpe file mapped}
nmdebug >var fileaddr = mapva("file1.suprtest")
nmdebug >=fileaddr {display the virtual address}
Debug/iX has a built-in calculator that accepts any Debug/iX
expression. You invoke the calculator with an equal sign "=".
Debug/iX evaluates the calculator expression and prints the
result. The calculator will display the full 64-bit address of
"file1.suprtest" as space.offset.
You can display the actual contents of the file:
nmdebug >dv fileaddr,20,s {first 20-bytes of file}
Warning: Due to a very serious bug in MPE/iX, never, never,
never do this on the file Catalog.Pub.Sys. If you open
Catalog.Pub.Sys with mapped access, you will cause a system
failure.
The map command displays the virtual address of a file in
space.offset format. You can use the DV (Display Virtual)
Command to display the file contents or you can use our method.
We prefer using a variable and mapva function, since typing in a
full 64-bit address correctly is quite difficult.
Cseq.Pub.Nuggets
While it is easy to predict the layout of parameters in our
simple example, things can get more complicated in MPE/iX. For
example, addresses can be passed as 64-bit quantities instead of
the default 32-bit values. The best way I've found to determine
parameter location is to use the Cseq (calling sequence} utility
in the Nuggets collection (available from Software Research
Northwest 206-463-3030). Here is the Cseq output for the FWRITE
intrinsic:
Procedure FWRITE (
Parm 1: int16 ; {R26, bits = 16}
Parm 2: anyvar record ; {(skip 25) R23, R24}
{bits = 65536}
{Address type = LongAddr}
Parm 3: int16 ; {SP-$0032, bits = 16}
Parm 4: UInt16 ) {SP-$0036, bits = 16}
uncheckable_anyvar
Note that the buffer parameter is a "LongAddr" that is passed in
both R23 and R24 (the first is the space and the second is the
offset). Fortunately, it is still easy to see the contents of
the buffer. If we were at a breakpoint at the start of FWRITE,
we would display the buffer with:
nmdebug >dv r23.r24,20,s {display buffer contents}
Integers: 16-bit versus 32-bit
Cmdebug displays integers in octal as 16-bit quantities. Nmdebug
displays integers in hex as 32-bit quantities. In our FWRITE
example, it is easy to see the value of the length parameter.
nmdebug >dv sp-32,1 {display the length}
$ 00005f00
We used the DV (Display Virtual) Command to display the stack
contents. The ",1" is not necessary - it's the default, but we
have shown it to make the following examples a little clearer.
The "dv sp-32" displays the value at sp-32 as a 32-bit quantity,
but we know that the actual value of FWRITE's length parameter is
a 16-bit quantity. You can display two 16-bit integers using the
following:
nmdebug >dv sp-32,1,,,2 {display two 16-bit integers}
$ 0000 5f00
nmdebug >dv sp-32,1,#,,2 {display two integers in decimal}
# 0 24320
Display Virtual always rounds down to a virtual address that is a
multiple of four and then displays one or more 32 bit words.
Miscellaneous Tips
Setting a "Return" Breakpoint
We showed how to set a return breakpoint in compatibility-mode.
You use a similar method to set a return breakpoint in
native-mode code:
nmdebug >b extract_ready
nmdebug >c
.
.
. {break at extract_ready}
nmdebug >lev 1;b pc,-1 {-1 means only break once}
nmdebug >c {continue execution}
The only difference between a CM return breakpoint, and an NM
one, is the name of the program counter. In native-mode it's
called "pc". This sets a return breakpoint immediately after the
code that called extract_ready. Note that it's safe to use this
breakpoint anywhere in the extract_ready procedure -- not just at
the beginning.
Debugging Batch Programs
In Debug/V, there was no practical way to debug a program running
in batch. In Debug/iX, you can debug a batch program on the
console, although it's a bit messy to set up. You have to do
these steps:
1. Obtain the pin number of the program you want to debug.
You'll need to use a program like Shot.Pub.Nuggets. You can
use the Showproc Command, if you have MPE/iX version 2.1 or
later versions.
2. Go to the console and insure that there will be no output on
the console. The easiest way to do this is to initiate a
:Restore on the console. This assumes that your tape drive is
not configured for auto-reply. Do not reply to the tape
request.
{On the console ...}
:hello user.acct
:restore
3. On another terminal, log on with SM capability and enter
debug. For example,
:hello manager.sys
:debug
4. Once you are inside Debug, you must set an environment
variable and force a breakpoint in the batch program. Our
example assumes that the batch program will call the FWRITE
intrinsic:
nmdebug >env job_debug true {set special variable}
nmdebug >b FWRITE:pin# {don't forget the pin#}
You don't actually type "b FWRITE:pin#" when setting the
breakpoint. You substitute the actual pin# that you obtained in
step 1 (e.g., "b FWRITE:103").
When the batch program encounters the breakpoint, Debug/iX is
invoked and all Debug/iX input/output is done via the console.
On the console you can type any of the usual Debug/iX commands.
When you finish your debugging session, you'll need to remember
to abort the :Restore that you initiated. You must also return
to the Manager.Sys session and disable job debugging:
cmdebug >env job_debug false
Macros
Debug/iX contains a small programming language that lets you
create your own macros. Debug/iX has no command to skip over
procedure calls, although almost all PC-based debuggers have this
feature. When single-stepping through a program, you rarely want
to single-step through external procedures (e.g., the Print
intrinsic). Use the j macro to jump over the next native-mode BL
instruction. Macros use braces for the body of the macro (i.e.,
as begin/end), so don't interpret the braces as comments. Here
is how to declare the macro:
nmdebug >mac j {b pc+$8,-1; c}
Macros are declared with the Mac Command. The first parameter to
the Mac Command is the macro name (in this case it's j). The
body of the macro follows and is surrounded with braces. Macros
can take several lines. The j macro sets a breakpoint at the
next native-mode instruction after a branch-and-link "pc+$8".
The breakpoint is only executed once ",-1". Multiple commands
are separated by semi-colons ";". The last step of the macro is
to execute the Continue Command "c". Note that the j macro is
only useful around branch-and-link instructions which is why we
jump eight bytes ahead of the program counter instead of four.
You execute the macro as if it were a built-in Debug/iX command:
nmdebug >j
Vfilepages Macro
When doing any performance measurements with disc files, you need
to know what portion of the file is in memory. This macro takes
advantage of many features of Debug/iX. The macro displays the
number of pages of a file that are currently present in virtual
memory.
/* Macro: Vfilepages
/*
/* Purpose: Display the number of pages (and corresponding
/* sectors) of a file that are actually in memory.
/*
/* Warning: Never use this macro on catalog.pub.sys.
/*
mac vfilepages (filename:str) {
map !filename;
w !filename " contains ";
w vainfo(mapva(!filename),"pages_in_mem"):"D";
w " pages in memory = ";
w vainfo(mapva(!filename),"pages_in_mem")*#16:"D";
w " sectors";
wl;
unmap(mapindex(!filename));
}
Lines starting with "/*" are treated as comments. The "filename"
is a parameter to the macro and it's of string type. To
understand the rest of the macro requires looking up the
description of the Map, Mapva, W, WL, and Unmap Commands and an
understanding of the Vainfo and Mapindex Function. We'll leave
that up to you. To invoke this macro, you would do the following
(note the quotes around the filename):
vfilepages "file50.suprtest"
file50.suprtest contains 8 pages in memory = 128 sectors
Warning: Because this macro uses the Debug/iX Map Command, do
not use it on the file catalog.pub.sys. If you do, you will
cause a system failure.
DBUGINIT File
Once you start writing macros, you will want to have them
automatically loaded when you enter Debug/iX. Debug/iX always
executes a use-file called DBUGINIT. Debug/iX first looks for
this file in the same group and account as the program, then it
looks in the logon group and account. Rather than fill our
DBUGINIT file with macros, we fill it with Use Commands for
different files that contain useful macros: You can use :file
commands for the DBUGINIT file, but you must use a fully
qualified filename. For example:
:hello david.dev,david
:print dbuginit.macro.dev
use splash.macro.splash
use macros.macro.dev
:file dbuginit.david.dev=debuginit.macro.dev
:run testprog;debug {Debug/iX will use debuginit.macro}
:Setdump Command
Classic MPE contains a :Setdump Command, but I believe most of us
ignored it because the traceback it printed was not symbolic. If
you enable :setdump in MPE/iX, you not only get an excellent
symbolic traceback, but in native-mode you are placed into
Debug/iX (certain exceptions apply to privileged-mode programs).
MPE Commands
You can enter almost any MPE command by preceding it with a
colon. This includes UDCs and the :Run Command. Often in the
middle of a debugging session, you need to examine your source
code. An easier way to do this is to run your editor from within
Debug/iX. One word of caution -- Debug/iX, like many HP
products, fails to see if a son process has terminated or
suspended.
We also find it useful to invoke Cseq.Pub.Nuggets when we are
debugging a program. This lets us determine the location of the
parameters for any MPE intrinsic:
nmdebug >:cseq.pub.nuggets {obtain parameter addresses}
Running Qedit from Debug/iX
If you invoke Qedit from Debug/iX, be sure to run it with Parm=32
(this tells Qedit not to suspend on exit). The most likely
reason to invoke Qedit from Debug/iX is to examine your source
code. If you do not /Shut your file before running your program,
you will get "Error: Busy file" when you try to open your file
inside Qedit (inside Debug/iX). To get around the problem, you
can either /List your source code or /Text a copy.
Conclusion
If you are going to make heavy use of Debug/iX, I strongly
recommend getting the System Debug Reference Manual. While it's
not helpful for learning Debug/iX, it's invaluable in looking up
specific commands and their syntax. That part number again is
#32650-90013.
When I first set out to write this article, I thought that it
would only take me a few paragraphs to convey what I'd learned
about Debug/iX. If you've got this far, you realize that I
underestimated the amount of material -- not surprising given the
rich feature set of Debug/iX.