From: Guinevere Larsen <guinevere@redhat.com>
To: Eli Zaretskii <eliz@gnu.org>, Simon Marchi <simark@simark.ca>
Cc: gdb-patches@sourceware.org
Subject: Re: [PATCH] gdb: add tutorial command
Date: Wed, 3 Dec 2025 10:07:53 -0300 [thread overview]
Message-ID: <b8f85690-64c1-4ead-9315-861199389119@redhat.com> (raw)
In-Reply-To: <867bv33fd1.fsf@gnu.org>
[-- Attachment #1: Type: text/plain, Size: 7438 bytes --]
On 12/3/25 9:43 AM, Eli Zaretskii wrote:
>> Date: Tue, 2 Dec 2025 14:33:08 -0500
>> Cc:gdb-patches@sourceware.org
>> From: Simon Marchi<simark@simark.ca>
>>
>>> First, writing this command in Python means that GDB without Python
>>> will not have a tutorial, which is IMO a pity. Is it so complicated
>>> to have this implemented in C++ instead? The text and the basic
>>> script of the tutorial session could be on a text file that code
>>> accesses when the tutorial is running, so that only the necessary
>>> stuff needs to be in code.
>> If possible, it would indeed be nice, but I'll take a tutorial written
>> in Python over nothing.
> If the C++ implementation is not feasible or not practical, I agree.
> My point is that we should consider that possibility before deciding
> on a Python implementation.
>
>>> Next, I'm not sure we need to compile a program for this purpose. We
>>> could use the GDB executable itself instead: that would allow us to
>>> show the basic commands without the need for the user to have a
>>> compiler and a working development environment to go with it.
>> I don't see how that's feasible or practical. The GDB binary users are
>> most likely to have are optimized (bad debugging experience, especially
>> if you're learning) and won't have debug info anyway. And like
>> Guinevere said, GDB is way too complicated to throw at newbies anyway.
>> I like Guinevere's idea of giving the user a small program they can toy
>> with.
> This is a _GDB_ tutorial. It doesn't need to include a complete
> debugging problem described in its entirety, right up to finding the
> bug and fixing it. It needs to show the use of important GDB
> commands, under the assumption that the person using the tutorial
> knows how to debug programs, but is not familiar with GDB commands to
> do that. (Yes, teaching debugging techniques in addition is helpful,
> but IMO the minimal requirements from a useful tutorial are to teach
> the commands.)
>
> For the purpose of teaching the useful commands, it doesn't matter
> much that GDB is optimized, the only possible problem could be that
> it's stripped. If it is not stripped, I don't see why we couldn't
> show use of important commands on GDB itself, explaining their purpose
> in plain text. There's no need to solve an actual bug, just let users
> tinker with commands when they have a live program at their disposal.
> Whatever the complexity of GDB, users definitely can step it, step
> into functions, examine its data, set breakpoints and watchpoints and
> see them break, etc. We could also show how to change values of
> variables and call functions from the program.
It does matter that GDB is optimized, though. Assuming that the user
could get debug info, they could use "next" and end up in a previous
line because the compiler decided to reorder instructions/lines, they
could try to step into a function that was entirely optimized away, they
could try to read the value of a variable only to find that it was
optimized away
And, most importantly, if they aren't familiar with GDB, it isn't
unreasnable to think they never used a debugger before and only did
printf debugging. Having those confusing reactions would only cement in
such user's brains that debuggers are actually not as useful as printf,
because variables can disappear when they never do in a printf, for example.
Lastly, we aren't arguing about still needing to come up with a small
toy problem. I already did that, the code is already here. The decision
is "should we scrap what has already been done or not" rather than
"should we come up with something very good from scratch or not".
> Moreover, if we limit
> ourselves to teaching commands, we can show advanced commands and
> techniques, like working with threads and scheduling them, following
> 'fork', handling signals, etc. -- these are hard to impossible to do
> with toy programs, but they are needed in most real-life debugging
> sessions.
These are most definitely out of scope for a tutorial that opens with
"welcome to GDB" and assumes minimal knowledge of programming in
general. But that's ok, because this doesn't have to be the only
tutorial. What I'm presenting here is a quickstart tutorial, that'll get
users acquainted with general purpose tools, and this framework can
easily be expanded to have specific tutorials, like multi-threaded
debugging, reverse debugging, linker namespace debugging and other
advanced cases I haven't ran into.
Also, I think it's worth to note that I haven't used any of the other
commands. So unless my last 5 years of work experience (4 of them in
GDB) aren't real life debugging sessions, I think you're overestimating
the commonality of the issues you deal with.
>
>>> Also, a tutorial doesn't have to teach people how to debug, it could
>>> only teach them the important GDB commands to use for debugging.
>>> Doing both makes the tutorial more complex because it teaches two
>>> non-trivial subjects instead of just one.
>>>
>>> What do others think?
>> I don't understand your last point. Do you mean that the tutorial
>> should just be some text that says "you can use command X to do this,
>> command Y to do that, etc"? Seems way less interesting and interactive
>> than what is proposed here.
> Yes, I mean just describe the important commands and let the user try
> them and see the results.
Then I don't see the difference between using "tutorial" or telling the
user to use "help essential". And if that is your suggestion, we already
have the C++ implementation of a tutorial.
>
> Indeed, teaching both debugging techniques and GDB commands is better,
> but that job is much more complex if taken seriously.
But I already did the job. This is already implemented. Your suggestion
is that, because the task is hard, we shouldn't use the solution I
already came up with?
> And I question
> the actual value of showing how to debug a toy program anyway, because
> it is nothing like debugging real-life programs.
I am unfamiliar with any starter tutorial that, both, starts with a real
life example showing all the complexities of real life example and is
well liked at the same time.
> The latter
> frequently requires advanced techniques, some of them special to the
> program in question (e.g., see etc/DEBUG in the Emacs source tree).
Again, this isn't meant to be the tutorial that teaches all of GDB; to
quote the tutorial's own opening words: "This quick tutorial should be
enough to get ou acquainted with essential commands for basic debugging."
Once a first tutorial goes through and we have a basic framework for how
to teach things to users, you're welcome to try your hand at teaching
the advanced techniques with a tutorial of your own, say "tutorial
multithreaded" for example.
> So I suggest to start from a modest task of showing and teaching the
> important commands, leaving the rest to the users, who will need to
> learn much more, if they never debugged a real-life program. We may
> decide in the future to have a separate tutorial about debugging
> techniques.
>
> Please note that the above is just MO; I will not object and won't
> argue if you-all think differently. I just wanted to present an
> alternative POV, so that people could think about this and make up
> their minds. After all, it is unlikely that I will use this tutorial
> myself...
>
--
Cheers,
Guinevere Larsen
It/she
[-- Attachment #2: Type: text/html, Size: 9604 bytes --]
next prev parent reply other threads:[~2025-12-03 13:08 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-12-01 17:08 Guinevere Larsen
2025-12-02 14:20 ` Eli Zaretskii
2025-12-02 17:02 ` Guinevere Larsen
2025-12-02 19:33 ` Simon Marchi
2025-12-03 12:43 ` Eli Zaretskii
2025-12-03 13:07 ` Guinevere Larsen [this message]
2025-12-03 17:27 ` Simon Marchi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=b8f85690-64c1-4ead-9315-861199389119@redhat.com \
--to=guinevere@redhat.com \
--cc=eliz@gnu.org \
--cc=gdb-patches@sourceware.org \
--cc=simark@simark.ca \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox