Re: How do you use GDB to debug GDB

Re: How do you use GDB to debug GDB

[Eli Zaretskii]

On Tue, 20 Mar 2001, Andrew Cagney wrote:

> > One thing puzzles me, though: why is the section "Debugging GDB with
> > itself" in gdbint.texinfo instead of gdb.texinfo?  Is there any reason
> > not to move (or copy) it to gdb.texinfo?  Can some of the elders
> > please shed some light on this?
> 
> To expand a little on Michael's comment.
> 
> gdbint.texinfo is for internals - the things you see when you lift the
> bonnet (1) and start poking around with the engine.

Exactly.  And, at least in my opinion, a collection of user-level
tricks useful for debugging a specific program called `gdb' is not
exactly ``what you see under the hood''.

> As a complete asside, it has also been suggested that the two documents
> be merged.  From memory GCC did this.

How about if we do that for 5.1?  Shouldn't be too hard, I think.

OTOH, the GCC manual leaves a lot to be desired (IMHO), so perhaps we
shouldn't follow their lead without thinking real hard.

Re: Merging manuals (was Re: How do you use GDB to debug GDB)

[Eli Zaretskii]

On Tue, 20 Mar 2001, Stan Shebs wrote:

> The arguments for merging are that it slightly simplifies document
> maintenance, and more importantly that it facilitates the transition
> from being a user of the tool to being a developer of it.

There's another important consideration: the ease of finding
information when you use the manual as a reference (as opposed to a
textbook which is read in its entirety).

The most efficient means of using a manual as a reference is the `i'
command of the Info browser.  (For those who don't know: that command
looks up a string you type in the manual's indices, and in a
well-indexed manual, will usually land you on the right spot in a
matter of seconds.)

If that fails, the next best alternative is `s' which searches the
manual's text for a string or a regexp.

By contrast, the facilities presented by today's Info browsers for
searching multiple documents are abysmally inadequate.  In the
stand-alone Info reader, we have the --apropos option (which I'm quite
sure some of the people here have never heard about); and we have
nothing similar in Emacs.

Also, please note that the chapters which describe annotations and
GDB/MI are already very technical in nature and come very close to
documenting the internals.

So I think, on balance, merging the two documents would be a good
thing.

Re: How do you use GDB to debug GDB

[Michael Elizabeth Chastain]

Hello Stephen,

This is how I do it:

  % gdb powerpc-motorola-elf-gdb
  (gdb) set prompt [top] 
  [top] set args --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe
  [top] break help_command
  [top] run

The first command helps me keep track of which gdb I'm in.  There is
a space at the end of the command: "set prompt [top] ".

The second command tells the superior gdb what the arguments are for
the inferior gdb.  This is the heart of what you want.  Note that you
can re-run the inferior gdb and the arguments will repeat.

The third line sets a convenient breakpoint in the superior gdb so that
you can bounce back up to it.  I usually do "help foo" instead of "help"
for getting to the superior gdb, to reduce the screen output.

The fourth line starts the inferior gdb with the arguments given on the
second line.

I've done this on both Unix/Linux and Cygwin gdb's.

Cheers,

Michael Elizabeth Chastain
<chastain@redhat.com>
"love without fear"

Re: How do you use GDB to debug GDB

[Kevin Buettner]

On Mar 20, 11:17am, Eli Zaretskii wrote:

> One thing puzzles me, though: why is the section "Debugging GDB with
> itself" in gdbint.texinfo instead of gdb.texinfo?  Is there any reason
> not to move (or copy) it to gdb.texinfo?

Presumably there's little reason for the average user to want to
debug GDB (with itself).  Therefore it's more suitable for such
material to appear in the internals document...

Kevin

Re: How do you use GDB to debug GDB

[Stephen Smith]

Michael Elizabeth Chastain wrote:

> Hello Stephen,
>
> This is how I do it:
>
>   [top] break help_command
>   [top] run
>
> The third line sets a convenient breakpoint in the superior gdb so that
> you can bounce back up to it.  I usually do "help foo" instead of "help"
> for getting to the superior gdb, to reduce the screen output.
>

Ok, so how do I active "help_command"

It doesn't appear to be by accessing help

sps

Re: How do you use GDB to debug GDB

[Eli Zaretskii]

On Mon, 19 Mar 2001, Michael Elizabeth Chastain wrote:

>   % gdb powerpc-motorola-elf-gdb
>   (gdb) set prompt [top] 
>   [top] set args --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe
>   [top] break help_command
>   [top] run
> 
> The first command helps me keep track of which gdb I'm in.  There is
> a space at the end of the command: "set prompt [top] ".
> 
> The second command tells the superior gdb what the arguments are for
> the inferior gdb.  This is the heart of what you want.  Note that you
> can re-run the inferior gdb and the arguments will repeat.
> 
> The third line sets a convenient breakpoint in the superior gdb so that
> you can bounce back up to it.  I usually do "help foo" instead of "help"
> for getting to the superior gdb, to reduce the screen output.
> 
> The fourth line starts the inferior gdb with the arguments given on the
> second line.

I think this is important enough to go into the manual.

If anyone has additional tricks pertginent to debugging GDB, please
post them, and I will merge them all together.

One thing puzzles me, though: why is the section "Debugging GDB with
itself" in gdbint.texinfo instead of gdb.texinfo?  Is there any reason
not to move (or copy) it to gdb.texinfo?  Can some of the elders
please shed some light on this?

Re: How do you use GDB to debug GDB

[Andrew Cagney]

> One thing puzzles me, though: why is the section "Debugging GDB with
> itself" in gdbint.texinfo instead of gdb.texinfo?  Is there any reason
> not to move (or copy) it to gdb.texinfo?  Can some of the elders
> please shed some light on this?

To expand a little on Michael's comment.

gdbint.texinfo is for internals - the things you see when you lift the
bonnet (1) and start poking around with the engine.

gdb.texinfo is for the user - the external interfaces to GDB.  If anyone
has ever wondered why the remote protocol is documented in gdb.texinfo
and not gdbint.texinfo then this is it - it is an external interface to
gdb.  To get carried away with the car metaphore, its a bit like the
section of the manual which tries to explain how it is possible to
change a flat using just the spare and tools all hidden somewhere in the
boot (2).

As a complete asside, it has also been suggested that the two documents
be merged.  From memory GCC did this.

	Andrew


(1) Hood for the geographically impaired.
(2) Trunk for the geographically impaired.

Re: Merging manuals (was Re: How do you use GDB to debug GDB)

[J.T. Conklin]

>>>>> "Stan" == Stan Shebs <shebs@apple.com> writes:
>> As a complete asside, it has also been suggested that the two documents
>> be merged.  From memory GCC did this.

Stan> GCC still works this way.  GDB has it as a separate document
Stan> because that's how John Gilmore set it up, and although I wasn't
Stan> there, I bet Roland Pesch, as the only professional tech writer
Stan> at Cygnus at the time, strongly objected to gluing the internals
Stan> documentation into the user manual.  (Probably the same way
Stan> would have happened to GCC if he'd had any input into GCC docs.)

Stan> I've thought about merging them from time to time.  The main
Stan> argument against merging should be obvious; the user manual is
Stan> just that, and should not include anything that might mislead or
Stan> intimidate users.

I think the split users/internals manuals is a good thing.  I've seen
the puzzled expressions of folks when encountering the gcc manual for
the first time.  They were quite relieved when they learned that they
didn't have to learn about all that stuff just to use the compiler.

A further split into a users and a reference manual is also desirable.
I can't recall a single manual where both introductory and reference
information was presented well.  Separating them into two allows each
to focus more clearly on it's target audiance, but the pair would
probably be more difficult to maintain than one.  Because of that, I
can't recommend we go that route.  

But it is something to consider as we continue editing the manual, I
think our current scheme leaves a lot to be desired.

        --jtc

-- 
J.T. Conklin
RedBack Networks

Re: Merging manuals (was Re: How do you use GDB to debug GDB)

[Eli Zaretskii]

On 20 Mar 2001, J.T. Conklin wrote:

> I've seen the puzzled expressions of folks when encountering the gcc
> manual for the first time.

I don't think the GCC manual is something we should consider as an
example: it is IMHO in such a bad shape that it cannot possibly serve
as a useful example of anything.  For starters, its indices, the
single most important means for finding information quickly, omit too
much; even the command-line options are not indexed!

> A further split into a users and a reference manual is also desirable.
> I can't recall a single manual where both introductory and reference
> information was presented well.

Please tell what would you like to see in a typical reference
manual, and why do you think a single manual cannot usefully serve
that purpose along with being an introduction.

The usual philosophy of a good manual written in Texinfo is that you
describe the various package facilities in a logical order, as
appropriate for someone who learns to use the package for the first
time, and leave it up to the indexing to provide the ability of using
the manual as a reference.

> But it is something to consider as we continue editing the manual, I
> think our current scheme leaves a lot to be desired.

If you have specific issues with the current scheme and/or suggestions
for improvement, please tell what they are.

Re: How do you use GDB to debug GDB

[Stephen Smith]

I ignored the break line since I'm not sure what to do with it.  Anyway thanks.  I now will be able to work on the
shared library problem that I keep posting about.

sps

Michael Elizabeth Chastain wrote:

> Hello Stephen,
>
> This is how I do it:
>
>   % gdb powerpc-motorola-elf-gdb
>   (gdb) set prompt [top]
>   [top] set args --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe
>   [top] break help_command
>   [top] run
>

Re: Merging manuals (was Re: How do you use GDB to debug GDB)

[Michael Meissner]

On Tue, Mar 20, 2001 at 04:23:37PM -0800, J.T. Conklin wrote:
> >>>>> "Stan" == Stan Shebs <shebs@apple.com> writes:
> >> As a complete asside, it has also been suggested that the two documents
> >> be merged.  From memory GCC did this.
> 
> Stan> GCC still works this way.  GDB has it as a separate document
> Stan> because that's how John Gilmore set it up, and although I wasn't
> Stan> there, I bet Roland Pesch, as the only professional tech writer
> Stan> at Cygnus at the time, strongly objected to gluing the internals
> Stan> documentation into the user manual.  (Probably the same way
> Stan> would have happened to GCC if he'd had any input into GCC docs.)
> 
> Stan> I've thought about merging them from time to time.  The main
> Stan> argument against merging should be obvious; the user manual is
> Stan> just that, and should not include anything that might mislead or
> Stan> intimidate users.
> 
> I think the split users/internals manuals is a good thing.  I've seen
> the puzzled expressions of folks when encountering the gcc manual for
> the first time.  They were quite relieved when they learned that they
> didn't have to learn about all that stuff just to use the compiler.

Note, in the case of the GCC manual, you really can't use the extended asm
extension, without knowing what the valid constraint letters are, and those of
course are in the implementation section.

-- 
Michael Meissner, Red Hat, Inc.  (GCC group)
PMB 198, 174 Littleton Road #3, Westford, Massachusetts 01886, USA
Work:	  meissner@redhat.com		phone: +1 978-486-9304
Non-work: meissner@spectacle-pond.org	fax:   +1 978-692-4482

Merging manuals (was Re: How do you use GDB to debug GDB)

[Stan Shebs]

Andrew Cagney wrote:
> 
> gdbint.texinfo is for internals - the things you see when you lift the
> bonnet (1) and start poking around with the engine.
> 
> gdb.texinfo is for the user - the external interfaces to GDB.  If anyone
> has ever wondered why the remote protocol is documented in gdb.texinfo
> and not gdbint.texinfo then this is it - it is an external interface to
> gdb.  To get carried away with the car metaphore, its a bit like the
> section of the manual which tries to explain how it is possible to
> change a flat using just the spare and tools all hidden somewhere in the
> boot (2).

A good analogy!

> As a complete asside, it has also been suggested that the two documents
> be merged.  From memory GCC did this.

GCC still works this way.  GDB has it as a separate document because
that's how John Gilmore set it up, and although I wasn't there, I bet
Roland Pesch, as the only professional tech writer at Cygnus at the time,
strongly objected to gluing the internals documentation into the user
manual.  (Probably the same way would have happened to GCC if he'd had
any input into GCC docs.)

I've thought about merging them from time to time.  The main argument
against merging should be obvious; the user manual is just that, and
should not include anything that might mislead or intimidate users.
The arguments for merging are that it slightly simplifies document
maintenance, and more importantly that it facilitates the transition
from being a user of the tool to being a developer of it.  (How often
have you looked at a GCC manual to find an option, then idly flipped
through the porting section and thought about trying your hand at it?
All the info you need is right there, right?....)

To me the arguments on each side seemed to have about equal weight, and
so I opted for the status quo.  But if people felt strongly that the GCC
approach was better, I'd be agreeable to merging the GDB manuals.

Stan

Re: How do you use GDB to debug GDB

[Michael Elizabeth Chastain]

Eli Zaretskii writes:

> I think this is important enough to go into the manual.

I'm up for writing a manual section on "debugging gdb with gdb".
It's a good task for my Sunday Project.

I just added a yellow sticky to my monitor to that effect.

Michael

Re: How do you use GDB to debug GDB

[Christopher Faylor]

On Mon, Mar 19, 2001 at 05:44:12PM -0500, Frank Ch. Eigler wrote:
>Fernando Nasser <fnasser@cygnus.com> writes:
>: [...]
>: Start your "top" gdb in a separate shell.  Use the command attach passing the Windows 
>: PID (as opposed to the Unix PID) of the other GDB.
>: [...]
>
>By the way, why does cygwin gdb use Windows PIDs only instead of
>cygwin PIDs, or some fall-back defaulting from one to the other?

GDB uses Windows pids because it can debug more than just cygwin
applications.  In older DLLs it was possible to have the same number
indicate both a Windows PID and a Cygwin PID, so there was ambiguity.
You don't want to have a mechanism which disallows the ability to
debug a program or ends up debugging the wrong program.

In current net releases of Cygwin it is (or should be) impossible
to have the cygwin pid == the windows pid except for when they
are the same process.

So, I guess you could have a fallback mechanism if somebody wanted
to implement it.

cgf

Re: How do you use GDB to debug GDB

[Michael Elizabeth Chastain]

Hi Eli,

> Note that there's already such a section in gdbint.texinfo (but it 
> doesn't tell much).

Okay, I see it.  I'll be expanding that section.

> I just asked if somebody knew of any reason not to 
> move that into gdb.texinfo.

Mmmm, that is a separate issue.

gdbint.texinfo is the right file for people who are working on gdb.
Someone running gdb on another copy of gdb is working on gdb.
So I'm in favor of leaving this section in gdbint.texinfo.

Michael

Re: How do you use GDB to debug GDB

[Fernando Nasser]

Stephen Smith wrote:
> 
> I have a gdb debugger configured to for a remote target (host i686-pc-cygwin target=powerpc-motorola-elf).  normally I
> run gdb as:
> 
>      powerpc-motorola-elf-gdb --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe
> 
> Now I want to run the gdb against the above process.  Since I am running this in a Windows NT command shell I can't use
>     gdb   /usr/local/bin/powerpc-motorola-elf-gdb  some_process_id
> 
> Can someone help
> 

Use the "ps -l" command.  It will give you the Windows PID in addition to the Unix (Cygwin) one.


Start your "top" gdb in a separate shell.  Use the command attach passing the Windows PID (as opposed to the Unix PID) of the other GDB.

You are now debugging your powerpc-motorola-elf-gdb


Good luck.





-- 
Fernando Nasser
Red Hat - Toronto                       E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: How do you use GDB to debug GDB

[J.T. Conklin]

>>>>> "Stephen" == Stephen Smith <ischis2@home.com> writes:
Stephen> I have a gdb debugger configured to for a remote target (host
Stephen> i686-pc-cygwin target=powerpc-motorola-elf).  normally I run
Stephen> gdb as:
Stephen>
Stephen>      powerpc-motorola-elf-gdb --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe
Stephen>
Stephen> Now I want to run the gdb against the above process.  

On a UNIX system, you would first start a native GDB, and then use it
to run and debug the child GDB.

        $ gdb /usr/local/bin/powerpc-motorola-elf-gdb
        GDB is free software and you are ...
        (gdb)

At this point, I typically install breakpoints where I think there the
problem lies.  The .gdbinit file distributed with GDB sets breakpoints
and sets directory paths, etc.; thus creating an initial environment
that are likely to be useful debugging GDB.

Then run your child GDB.

        (gdb) run --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe

I don't have experience with NT to know whether this will work there.

        --jtc

-- 
J.T. Conklin
RedBack Networks

Re: How do you use GDB to debug GDB

[Eli Zaretskii]

On Tue, 20 Mar 2001, Michael Elizabeth Chastain wrote:

> > I just asked if somebody knew of any reason not to 
> > move that into gdb.texinfo.
> 
> Mmmm, that is a separate issue.

Yes.

> gdbint.texinfo is the right file for people who are working on gdb.
> Someone running gdb on another copy of gdb is working on gdb.
> So I'm in favor of leaving this section in gdbint.texinfo.

Thanks for the feedback.  In any case, please modify the section in 
gdbint.texinfo; any administrivia such as moving it to another place can 
then be done separately.

Thanks again!

Re: How do you use GDB to debug GDB

[Frank Ch. Eigler]

Fernando Nasser <fnasser@cygnus.com> writes:

: [...]
: Start your "top" gdb in a separate shell.  Use the command attach passing the Windows 
: PID (as opposed to the Unix PID) of the other GDB.
: [...]

By the way, why does cygwin gdb use Windows PIDs only instead of
cygwin PIDs, or some fall-back defaulting from one to the other?


- FChE

Re: How do you use GDB to debug GDB

[Eli Zaretskii]

On Tue, 20 Mar 2001, Michael Elizabeth Chastain wrote:

> Eli Zaretskii writes:
> 
> > I think this is important enough to go into the manual.
> 
> I'm up for writing a manual section on "debugging gdb with gdb".
> It's a good task for my Sunday Project.

Thanks!

> I just added a yellow sticky to my monitor to that effect.

Note that there's already such a section in gdbint.texinfo (but it 
doesn't tell much).  I just asked if somebody knew of any reason not to 
move that into gdb.texinfo.

How do you use GDB to debug GDB

[Stephen Smith]

I have a gdb debugger configured to for a remote target (host i686-pc-cygwin target=powerpc-motorola-elf).  normally I
run gdb as:

     powerpc-motorola-elf-gdb --command=C:/rti/bin/rtidbug.ini C:/examples/dll/main.exe

Now I want to run the gdb against the above process.  Since I am running this in a Windows NT command shell I can't use
    gdb   /usr/local/bin/powerpc-motorola-elf-gdb  some_process_id

Can someone help

sps