Remove/fix `prettytable-rs` dependency

[cschwan]

Recently, ubuntu-latest was switched from ubuntu-20.04 to ubuntu-22.04, which seems to break pineappl. It segfaults, see https://github.com/NNPDF/pineappl/actions/runs/3646167214/jobs/6156975988. My bet is that it's not caused by pineappl itself but rather by a dependency.

[alecandido]

I can tell you that I could compile PineAPPL on my local Ubuntu 22.04, I did it many times. And run the CLI as well, of course.

[cschwan]

Yes, it's not quite that easy. What I noticed is that it started breaking when the switch happened, but it's still there with ubuntu-20.04.

It seems that pineappl [subcommand] --help is fine, while operations involving convolutions segfault: https://github.com/NNPDF/pineappl/actions/runs/3646494375/jobs/6157637600.

[alecandido]

I had a look to the logs, but it is highly non-trivial. Especially, I do not see what they all have in common: consider that also commands like pineappl obl are failing, and no convolution should be involved there.

Maybe the more precise statements is about grid loading: --help does not load any grid.

[cschwan]

True, that is even weirder.

[cschwan]

I can reproduce a segfault with Rust nightly, which segfaults in prettytable::Table::printstd - that makes sense with the observations above, they all print a table using prettytable-rs.

[alecandido]

I agree it matches the observation, but is the CI running with Nightly?

[cschwan]

Yes, it was. It works with stable, see here: https://github.com/NNPDF/pineappl/actions/runs/3646808485/jobs/6158307245. However, I think the problem is prettytable-rs, and Rust nightly is probably just more strict than stable (in some way I don't know). Since you can't produce segfaults in Rust unless you use unsafe, my bet is on

https://github.com/phsym/prettytable-rs/blob/e159600d999f55f49fa9141e1a014d776a67c836/src/lib.rs#L513-L518

which is the only piece marked 'unsafe'. I can confirm that this piece of code is called.

[alecandido]

Interestingly enough, the guy wrote:

// All this is a bit hacky. Let's try to find something else

on May 29, 2017, and it's still there.

It looks a good candidate to break with new versions. Essentially, it is recasting an immutable reference to a table to a mutable one, to perform a single operation. But that's an entry point for plenty of issues.

(and it is the only unsafe in the whole library in master, even though we should check the tag for the version used...)

[alecandido]

Ok, it seems like there is already an issue open mentioning it https://github.com/phsym/prettytable-rs/issues/145. Is this matching the usage in pineappl_cli? @cschwan

[cschwan]

Yes, I'm pretty sure that's the cause.

[pinkforest]

You can just bump prettytable-rs to 0.10 - I've pushed a release :)

[cschwan]

@pinkforest I'm already on it, thanks!