Programming

Just yesterday 😅 there’s a bug in the main branch of Lemmy itself that I was trying to pinpoint (introduced after 0.18.0 was tagged). Instead of walking through all recent commits manually, I used bisect. Bisect is not a magic bullet, and you could do the same manually, but it’s a good tool in the toolbox to know sometimes.

4-5 times now. When confronted with more than a hundred commits between latest known working version and the one you've observed the bug (which was not catched by any of the unit tests) it can save some time to find the fishy commit.

In such a case I create a testcase on top to reproduce the bug. Then bisect and for each stage add the testcase, build, run tests. FYI: this only works if all (or at least most) of the commits in the chain are compilable - if you've done a big messy refactoring with several commits breaking the build, bisect can get you only so far.

For those (like me) who were not aware of git bisect:

The git bisect command is a powerful tool that quickly checks out a commit halfway between a known good state and a known bad state and then asks you to identify the commit as either good or bad. Then it repeats until you find the exact commit where the code in question was first introduced.

Quite a few times, sure. git bisect is a specific case of a more general technique -- binary search fault localization -- which comes in handy every once in a while (you can go a long while without needing it, but when you do need it, you'll be thankful for it). If you can't otherwise trace where in the code something is going wrong, bisect the code: comment or remove half of it out, see if it reproduces (therefore localizing it to either the removed or the remaining half), and repeat. If you're working with some software that's breaking on your config after a major version bump, bisect your config. Don't have an idea what introduced a bug into your branch? git bisect.

Yes, many times. And I recall using the technique manually back when I was working with Subversion many, many years ago. No fun stories though, sorry, it's just another tool in a toolbox.

If you know about bisect and don't find it useful, my first guess is you aren't making enough commits (or you're squashing too many of them). Keep your git history granular, make intermediary commits when possible (as long as the intermediary step doesn't make it completely broken), and you'll be using bisect for a good chunk of regressions.

Yeah, I don't need it often but the sort of problem it solves is very hard to solve otherwise. It is useful for when you know something is wrong and you have an easy way to check if the problem is there but when you don't know what the actual cause of the problem is.

The most obvious usage is for checking if the build passes. This is something easy to do with tools but hard to know why (or when) it began to fail.

Seems like most of the answers are on another instance and my client does not show them here. For those having the same problem: check here.

Federation is fun and all, but having “the same” sub on multiple instances does not make it easier atm.

Yeah it’s saved my bacon a bunch of times

The few times I have used git bisect, it has been on projects with no automated tests where the reported issue last worked far in the past. It isn't my first option, but it is what I turn to if I can't figure out what the correct internal state for some portion of what triggers the issue is supposed to be, saving a lot of time I could have spent banging my head into a wall.

I used it a few times to find a bug, introduced some commit ago.

It's mostly relevant for a project you're not familiar with (perhaps it is/was someone else's project, or perhaps a project that's too large for a single user to be familiar with the entirety of), since it helps you figure out where a bug came from.

If you're familiar with the entire project you usually don't need it IME.

I certainly have! Although I can count on two hands how many times I've used it, it has proven to be extremely powerful when trying to pin-point changes that introduced non-trivial bugs. Specifically if the time-span of when the bug was introduced it is very useful to quickly sift through commit history.

No, but I once wrote my one git bisect without realizing git bisect existed.

I used it to track down a commit that caused an old bug. The commit didn't pop out on a cursory check, and I was getting lost in tracking the test results.

git bisect is probably one of the most revolutionary parts of git, you can even automatize regression testing with it (if you have a regression test, you can run git bisect script.sh and it will bisect automatically the problem for you). I have used it lately to automate performance testing & regression finding.

Usually when theres a "mystery" bug on something that already worked before. Spending half a day running the build script over and over again, is still faster than debugging in different parts of the codebase with the hope of finding some hint.