Wondering if record save should be deprecated

Wondering if record save should be deprecated

[Guinevere Larsen via Gdb]

Hi everyone!

I decided to try a bit of a different workflow for reverse debugging and 
noticed that using "record save", then loading that file in a new gdb 
session, and I noticed this generated an assert. The assert itself is 
not too hard to solve (Hannes had a patch ready when I mentioned it in 
IRC), but it had me wondering: this bug is present since GDB 10, and 
there are no bugs in bugzilla about it.

In fact, searching for "record save" only has a stack overflow question 
from 2.5 years ago that uses precsave, but is about a separate topic. 
Everything else, including the few reverse debugging tutorials I've 
seen, never mention this option.

Does it make sense for us to continue supporting this feature that was 
nigh unusable for 5 releases without anyone noticing? Does anyone 
reading this makes use of this feature and managed to somehow avoid that 
crash?

-- 
Cheers,
Guinevere Larsen
She/Her/Hers

Re: Wondering if record save should be deprecated

[Keith Seitz via Gdb]

On 7/10/24 12:58 PM, Guinevere Larsen via Gdb wrote:
> Hi everyone!

Warning: I am not a record/replay user.

> Does it make sense for us to continue supporting this feature that was 
> nigh unusable for 5 releases without anyone noticing? Does anyone 
> reading this makes use of this feature and managed to somehow avoid that 
> crash?

 From the User Manual:

record save filename

     Save the execution log to a file filename. Default filename is 
gdb_record.process_id, where process_id is the process ID of the inferior.

     This command may not be available for all recording methods.


This certainly sounds like a useful feature to have/keep/make work.
I can envision some use cases where this would really be useful.

That said, I would not be quick to confuse utility with awareness.
GDB has so many features nowadays that even as a developer/packager
for it, I feel I barely use more than 25-30% of GDB's available
features.

My vote is to fix it, test it, and keep it.

Keith

Re: Wondering if record save should be deprecated

[Luis Machado via Gdb]

On 7/10/24 20:58, Guinevere Larsen via Gdb wrote:
> Hi everyone!
> 
> I decided to try a bit of a different workflow for reverse debugging and noticed that using "record save", then loading that file in a new gdb session, and I noticed this generated an assert. The assert itself is not too hard to solve (Hannes had a patch ready when I mentioned it in IRC), but it had me wondering: this bug is present since GDB 10, and there are no bugs in bugzilla about it.
> 
> In fact, searching for "record save" only has a stack overflow question from 2.5 years ago that uses precsave, but is about a separate topic. Everything else, including the few reverse debugging tutorials I've seen, never mention this option.
> 
> Does it make sense for us to continue supporting this feature that was nigh unusable for 5 releases without anyone noticing? Does anyone reading this makes use of this feature and managed to somehow avoid that crash?
> 

If it is easy enough to fix, sounds like that is the way then. As for whether people are using this,
it is hard to tell. I've been hearing more and more about people using RR for their
reverse debugging needs. Does it also save state?

Re: Wondering if record save should be deprecated

[Guinevere Larsen via Gdb]

On 7/16/24 9:24 AM, Luis Machado wrote:
> On 7/10/24 20:58, Guinevere Larsen via Gdb wrote:
>> Hi everyone!
>>
>> I decided to try a bit of a different workflow for reverse debugging and noticed that using "record save", then loading that file in a new gdb session, and I noticed this generated an assert. The assert itself is not too hard to solve (Hannes had a patch ready when I mentioned it in IRC), but it had me wondering: this bug is present since GDB 10, and there are no bugs in bugzilla about it.
>>
>> In fact, searching for "record save" only has a stack overflow question from 2.5 years ago that uses precsave, but is about a separate topic. Everything else, including the few reverse debugging tutorials I've seen, never mention this option.
>>
>> Does it make sense for us to continue supporting this feature that was nigh unusable for 5 releases without anyone noticing? Does anyone reading this makes use of this feature and managed to somehow avoid that crash?
>>
> If it is easy enough to fix, sounds like that is the way then. As for whether people are using this,
> it is hard to tell. I've been hearing more and more about people using RR for their
> reverse debugging needs. Does it also save state?
>
Yes. RR has 2 completely separate phases, one recording the execution of 
the inferior, and another opening a gdbserver that uses that recording 
for execution. So it's not like GDB is the only way to get a recording 
of a broken execution to be debugged later. The difference is that with 
RR you decide at the start that you want to record things, and with GDB 
you can decide mid-debugging session. And that RR only works for a few 
x86_64 CPUs, while GDB supports many more.

I've been convinced, let's keep it :)

-- 
Cheers,
Guinevere Larsen
She/Her/Hers