[Libreoffice-qa] rr - lightweight recording & deterministic debugging

[Terrence Enger]

On Thu, 2014-10-30 at 15:10 -0700, bfoman wrote:
> Hi!
> Today I stumbled upon rr - lightweight recording & deterministic debugging
> tool. 
> More infos at http://rr-project.org, source https://github.com/mozilla/rr. 
> This potentially could be great tool for developers to debug. Just imagine
> that .rr file is attached to bug report and dev simply downloads it and
> replays the crash in his environment. 
> Any volunteer using Linux to have time to check if that tool works with
> LibreOffice?
> Best regards.

This sounds really interesting, but I was not able to install the
package.  I forget where I got to before giving up.

Beside the uses that others foresee, I had dreams of another use:
(*) The overhead during data collection is said to be 15 to 20%.  Thus
    I could afford to use it all the time.
(*) I would then have a record of what I did in the all-too-common
    case where LibreOffice crashes while I am futzing around *after*
    giving up the attempt to recreate a reported bug; of course I
    have stopped making careful notes of what I was doing.
(*) Maybe(!) I could learn to read the recorded data and extract user
    inputs.
(*) And then, of course, there is the exercise of cutting out parts of
    the user actions until we have more-or-less minimal crasher.

On Mon, 2014-12-15 at 12:26 +0100, Michael Stahl wrote:
> On 14.12.2014 22:26, Robinson Tryon wrote:
> > Auto-replay of bug repro steps sounds pretty awesome for various
> > reasons, especially for drudge work such as re-testing MAB's and
> > moving them from, say, 4.2 -> 4.3. Even if we have a real person do
> 
> well that's not what rr does.  rr records a single execution and can
> replay it, which is useful to track down bugs that are very hard to
> reproduce - but the replay only works with identical program binaries.

I suspect that a lot of things must be the same.  Like, for example,
the full names of files being opened or saved, configuration options,
and what-not.  So, transferring a test to another system might be
quite a lot of work.  It might be easier to figure out how give a
developer access to my machine.

All the best,
Terry.