Tracking Issue for Deprecating Boost

[arcuru]

I believe we should move to make Boost a dependency for only a couple non-core exercises, while leaving the core of the track Boost-free. That would let us avoid having to help configure boost for every single platform and corner case, which just isn't going to happen.

For what we really needed Boost for, the unit test suite, we will switch to a header-only library so we can just distribute it along with the exercises, https://github.com/catchorg/Catch2 seems to be the most popular. Most platforms should then be able to compile problems without even bothering with CMake if necessary, since it'll just be a handful of self-contained files + the stdlib, and many IDEs have support for Catch2 specifically.

We will use version 2.7.0 of Catch, as that is the latest release as of today, located in a test/ subdirectory in each exercise folder.

I initially setup #159 to remove everything at once, but it ended up having to touch too much of the repo (I didn't realize how often we used boost outside of the unit test suite itself) and it's now so old as to be far too difficult to merge.

Also, I'm older and a bit smarter, so I realize we can just do it in pieces so long as we let the students know in the documentation.

What I'm proposing now is that we take the following actions in roughly this order.

• Remove Boost Regex - done Removing Boost::Regex #223
• Make Gigasecond and Meetup non-core exercises (some discussion in Consider removing dependency on Boost::date_time #224 already). These are the only exercises that pretty much require the use of boost libs, so we should make them non-core with extra documentation noting you'll need to install boost. - See Swap out gigasecond for reverse-string in the core tracks. #240
• Update the documentation to explain we're in the process of switching, so some exercises may use Catch and not require installing boost. - See Updating documentation to explain Boost deprecation. #239
• Update the exercises from Boost to Catch via PRs. A lot of which can be stolen from WIP: Remove Boost as a required dependency #159. - See Boost -> Catch Exercise Conversions #253
• (optionally) Remove the uses of Boost algorithms from our examples (just cherry-pick 8922c1f where I've already done that). This is only to make the examples more relevant for people who cheat and end up here. - See Removing boost/algorithms in example solutions #238
• After all exercises have been updated, update the docs again.

Most of the grunt work is in updating the test suites. I will, at a minimum, put together an example moving the Hello World exercise over to Catch to demonstrate what to do for anyone who wants to help, and figure out what to do with an old script I had that helped me do a lot of the conversion work. There isn't a need to do the problems 1 at a time, but be somewhat reasonable as I do want 2 people to actually look at each of the changes.

If you'd like to grab some work for yourself, go ahead and claim it here or just open a PR.

[arcuru]

Alright, I've setup the initial set of changes. If you'd like to help in the short term, please take a look at #239, #240, and #244 and let me know if you have any issues or thoughts.

Once I get a little feedback I can merge all 3 of those and we'll be on our way :)

If you might be interested in helping to update exercises, please reply here and I can write up a short tutorial on how to do it relatively easily, (it's just unclear right now if anybody is willing to help, so I want to be sure it'd actually be used before I spend the time).

[arcuru]

I'll leave these open for feedback for another week. If nothing comes up I'll merge in those initial 3 PRs on April 7th/8th, which will switch over all of the core exercises to Catch.

The only real downside I've encountered is that the Catch builds take longer since it needs to be compiled every time, to the point that it adds ~7 seconds on each compile. Some header only test suites are faster (doctest, for instance) but Catch is by far the most used. I think it's a fair tradeoff to make for an easier setup.

[elyashiv]

I didn't actually tried, and I'm not sure how portable it will be, but there is an option for pre-compiled headers. Did you look into this?

[arcuru]

Precompiled headers aren't really that portable, and I'm not quite sure how setting them up in CMake would work.

There does seem to be a cmake module that will do some of that for us though, which might be worth looking into at some point - https://github.com/sakra/cotire

Of note, Visual Studio compiled it in about half the time as clang/gcc.

[arcuru]

Alright, the kickoff PRs have been ready for ~2 weeks with no objections, so I have gone ahead and merged them. I'll make a tracking issue for onboarding the rest of the exercises.

Since it seems I'll be doing these myself, I will likely do them in batches and leave each batch up for a week or two for comments, then do the merge if I hear no objections. Will probably take a month+ unless I feel like devoting a ton of time.

With the exception of the already opened PRs, I won't be looking at new problems nor will I support merging new problems that use Boost at all.

[KevinWMatthews]

Hi @patricksjackson ,
I've been keeping abreast of the Catch PR's and things look good from afar. I'm trying things out in detail now :)
Unfortunately I'll be traveling for work for most of next week, but I'll still try to keep an eye on things.

[arcuru]

@KevinWMatthews No problem. I'll leave things up for a week or two just in case you or others are around and want to comment, but unless something comes up in that time or I think something is questionable I'll just push things through myself.