Add blogpost about the rr memory debugging saga

[Keno]

No description provided.

[github-actions]

Once the build has completed, you can preview your PR at this URL: https://julialang.netlify.app/previews/PR1002/

[Keno]

Direct link: https://julialang.netlify.app/previews/pr1002/blog/2020/09/rr-memory-magic/

[staticfloat]

This was really fun! Great work!

[staticfloat]

Instantly lost me. 😂

I'm not sure what the C code is supposed to illustrate; if it's supposed to communicate to us that the code is straightforward, it may not be doing its job. Perhaps you can shrink it down to 2-3 lines? I can read a backtrace, so I know that it's focused on jl_svecset, of which there is only one in the above function, but there's a lot going on in that function and none of it is commented, so the casual reader will most likely either completely skip this or be confused by it.

It's also not clear what "getting the type pointer from some array" means. If you could unpack that a bit, (e.g. "getting a pointer to the type information associated with this array, a common operation done by the Julia runtime anytime it needs to access the type information of an object") it would also be helpful.

[staticfloat]

Suggested change

	living under the floors. Here, we were able to fairly convincingly conclude that
	living under the floor. Here, we were able to fairly convincingly conclude that

[Keno]

I like floors, indicating the abstract concept of living under peoples' floors rather than a specific floor.

[fredrikekre]

Mostly formatting and some typos.

[fredrikekre]

Should be updated before merge.

[fredrikekre]

Suggested change

	(note that we masked out the low bits, which julia does use as mentioned above - and
	(note that we masked out the low bits, which Julia does use as mentioned above - and

[Keno]

This is an implementation detail, so the style guide suggests julia (the program/implementation) rather than Julia (the abstract language). I'll code-format it.

[fredrikekre]

Suggested change

	# Conclusion
	## Conclusion

[fredrikekre]

I think this will highlight as Julia code, but it looks pretty good anyway for the rr traces.

[Keno]

I'll use bash highlighting.

[StefanKarpinski]

A few edits:

However, instead of seeing this segfault, we ended up somewhere else. In this case (but instead at ‘write’), during the replay, we instead tried to execute a write system call

would be clearer as

However, instead of seeing this segfault, we ended up somewhere else: instead at 'write'. In this case, during the replay, rr found itself trying to execute a valid write system call rather than segfaulting

Ok, so this is actually a pointer to a julia object (the type Array{Any, 1}) with a corrupted high bit.

maybe:

Barring a massive coincidence, this is an otherwise-valid pointer to a julia object (the type Array{Any, 1}) with a corrupted high bit.

I’m a bit lost in the middle part where you’re hunting down the precise divergence point—maybe a bit more explanation of each step and why you’re doing it.

memory on the users machine

should be:

memory on the user’s machine

great post, especially the conclusion!

[Keno]

Thanks for the feedback. Everyone. Merged. Of course as always, additional feedback can happen via comment here or direct PR.