Go offline with the Player FM app!
419: What's New in Your World? (Extended Edition)
Manage episode 407681067 series 1401614
Stephanie introduces her ideal setup for enjoying coffee on a bike ride. Joël describes his afternoon tea ritual. Exciting news from the hosts: both have been accepted to speak at RailsConf! Stephanie's presentation, titled "So, Writing Tests Feels Painful. What now?" aims to tackle the issues developers encounter with testing while offering actionable advice to ease these pains. Joël's session will focus on utilizing Turbo to create a Dungeons & Dragons character sheet, combining his passion for gaming with technical expertise.
Their conversation shifts to artificial intelligence and its potential in code refactoring and other applications, such as enhancing the code review process and solving complex software development problems. Joël shares his venture into combinatorics, illustrating how this mathematical approach helped him efficiently refactor a database query by systematically exploring and testing all potential combinations of query segments.
Transcript:
JOËL: Hello and welcome to another episode of The Bike Shed, a weekly podcast from your friends at thoughtbot about developing great software. I'm Joël Quenneville.
STEPHANIE: And I'm Stephanie Minn, and together, we're here to share a bit of what we've learned along the way.
JOËL: So, Stephanie, what's new in your world?
STEPHANIE: So, today I went out for a coffee on my bike, and I feel like I finally have my perfect, like, on-the-go coffee setup. We have this thoughtbot branded travel mug. So, it's one of the little bits of swag that we got from the company. It's, like, perfectly leak-proof. I'll link the brand in the show notes. But it's perfectly leak-proof, which is great. And on my bike, I have a little stem bag, so it's just, like, a tiny kind of, like, cylindrical bag that sits on the, like, vertical part of my handlebars that connects to the rest of my bag. And it's just, like, the perfect size for a 12-ounce coffee.
And so, I put my little travel mug in there, and I just had a very refreshing morning. And I'd gone out on my bike for a little bit, stopping by for coffee and headed home to work. And I got to drink my coffee during my first meeting. So, it was a wonderful way to start the day.
JOËL: Do you just show up at the coffee shop with your refillable mug and say, "Hey, can you pour some coffee in this?"
STEPHANIE: Yeah. I think a lot of coffee places are really amenable to bringing your own travel mugs. So yeah, it's really nice because I get to use less plastic. And also, you know, when you get a to-go mug, it is not leak-proof, right? It could just slosh all over the place and spill, so not bike-friendly. But yeah, bring your own mug. It's very easy.
JOËL: Excellent.
STEPHANIE: So, Joël, what's new in your world?
JOËL: Also, warm beverages. Who would have thought? It's almost like it's cold in North America or something. I've been really enjoying making myself tea in the afternoons recently. And I've been drinking this brand of tea that is a little bit extra. Every flavor of tea they have comes with a description of how the tea feels.
STEPHANIE: Ooh.
JOËL: I don't know who came up with these, but they're kind of funny. So, one that I particularly enjoy is described as feels like stargazing on an empty beach.
STEPHANIE: Wow. That's very specific.
JOËL: They also give you tasting notes. This one has tastes of candied violet, elderberry, blackberry, and incense.
STEPHANIE: Ooh, that sounds lovely. Are you drinking, like, herbal tea in the afternoon, or do you drink caffeinated tea?
JOËL: I'll do caffeinated tea. I limit myself to one pot of coffee that I brew in the morning, and then, whenever that's done, I switch to tea. Tea I allow myself anything: herbal, black tea; that's fine.
STEPHANIE: Yeah, I can't have too much caffeine in the afternoon either. But I do love an extra tea. I wish I could remember, like, what even was in this tea or what brand it was, but once I had a tea that was a purplish color. But then, when you squeeze some lemon in it, or I guess maybe anything with a bit of acid, it would turn blue.
JOËL: Oh, that's so cool.
STEPHANIE: Yeah, I'll have to find what this tea was [laughs] and update the podcast for any tea lovers out there. But yeah, it was just, like, a little bit of extra whimsy to your regular routine.
JOËL: I love adding a little whimsy to my day, even if it's just seeing a random animated GIF that a coworker has sent or Tuple has some of the, like, reactions you can send if you're pairing with someone. And I don't use those very often, so whenever one of those comes through, and it's like, ship it or yay, that makes me very happy.
STEPHANIE: Agreed.
JOËL: This week is really fun because as we were prepping for this episode, we both realized that there is a lot that's been new in our world recently. And Stephanie, in particular, you've got some pretty big news that recently happened to you.
STEPHANIE: Yeah, it turns out we're making the what's new in your world segment the entire episode today [laughs]. But my news is that I am speaking at RailsConf this year, so that is May 7th through 9th in Detroit. And so, yeah, I haven't spoken at a RailsConf before, only a RubyConf. So, I'm looking forward to it. My talk is called: So, Writing Tests Feels Painful. What now?
JOËL: Wait, is writing tests ever painful [laughs]?
STEPHANIE: Maybe not for you, but for the rest of us [laughs].
JOËL: No, it absolutely is. I, right before this recording, came from a pairing session where we were scratching our heads on an, like, awkward-to-write test. It happens to all of us.
STEPHANIE: Yeah. So, I was brainstorming topics, and I kind of realized, especially with a lot of our consulting experience, you know, we hear from developers or even maybe, like, engineering managers a lot of themes around like, "Oh, like, development is slowing down because our test suite is such a headache," or "It's really slow. It's really flaky. It's really complicated." And that is a pain point that a lot of tech leaders are also looking to address for their teams.
But I was really questioning this idea that, like, it always had to be some effort to improve the test suite, like, that had to be worked on at some later point or get, like, an initiative together to fix all of these problems, and that it couldn't just be baked into your normal development process, like, on an individual level. I do think it is really easy to feel a lot of pain when trying to write tests and then just be like, ugh, like, I wish someone would fix this, right? Or, you know, just kind of ignore the signals of that pain because you don't know, like, how to manage it yourself.
So, my talk is about when you do feel that pain, really trying to determine if there's anything you can do, even in just, like, the one test file that you're working in to make things a little bit easier for yourself, so it doesn't become this, like, chronic issue that just gets worse and worse. Is there something you could do to maybe reorganize the file as you're working in it to make some conditionals a little bit clearer?
Is there any, like, extra test setup that you're like, "Oh, actually, I don't need this anymore, and I can just start to get rid of it, not just for this one example, but for the rest in this file"? And do yourself a favor a little bit. So yeah, I'm excited to talk about that because I think that's perhaps, like, a skill that we don't focus enough on.
JOËL: Are you going to sort of focus in on the side of things where, like, a classic TDD mantra is that test pain reflects underlying code complexity? So, are you planning to focus on the idea of, oh, if you're feeling test pain, maybe take some time to refactor some of the code that's under test, maybe because there's some tight coupling? Or are you going to lean a little bit more into maybe, like, the Boy Scout rule, you know, 'Leave the campsite cleaner than you found it' for your test files?
STEPHANIE: Ooh, I like that framing. Definitely more of the former. But one thing I've also noticed working with a lot of client teams is that it's not always clear, like, how to refactor. I think a lot of intermediate developers start to feel that pain but don't know what to do about it. They don't know, like, maybe the code smells, or the patterns, or refactoring strategies, and that can certainly be taught. It will probably pull from that. But even if you don't know those skills yet, I'm wondering if there's, like, an opportunity to teach, like, developers at that level to start to reflect on the code and be like, "Hmm, what could I do to make this a little more flexible?"
And they might not know the names of the strategies to, like, extract a class, but just start to get them thinking about it. And then maybe when they come across that vocabulary later, it'll connect a lot easier because they'll have started to think about, you know, their experiences day to day with some of the more conceptual stuff.
JOËL: I really like that because I feel we've probably all heard that idea that test pain, especially when you're test driving, is a sign of maybe some anti-patterns or some code smells in the underlying code that you're testing. But translating that into something actionable and being able to say, "Okay, so my tests are painful. They're telling me something needs to be refactored. I'm looking at this code, and I don't know what to refactor." It's a big jump. It's almost the classic draw two circles; draw the rest of the owl meme. And so, I think bridging that gap is something that is really valuable for our community.
STEPHANIE: Yeah, that's exactly what I hope to do in my talk. So, Joël, you [chuckles] also didn't quite mention that you have big news as well.
JOËL: So, I also got accepted to speak at RailsConf. I'm giving a talk on Building a Dungeons & Dragons Character Sheet Using Turbo.
STEPHANIE: That's really awesome. I'm excited because I want to learn more about Turbo. I want someone else to tell me [laughs] what I can do with it. And as a person with a little bit of Dungeons & Dragons experience, I think a character sheet is kind of the perfect vehicle for that.
JOËL: Building a D&D character sheet has been kind of my go-to project to experiment with a new front-end framework because it's something that's pretty dynamic. And for those who don't know, there's a bunch of fields that you fill in with stats for different attributes that your character has, but then those impact other stats that get rendered. And sometimes there can be a chain two or three long where different numbers kind of combine together. And so, you've got this almost dependency tree of, like, a particular number.
Maybe your skill at acrobatics might depend on a number that you entered in the dexterity field, but it also depends on your proficiency bonus, and maybe also depends on the race that you picked and a few other things. And so, calculating those numbers all of a sudden becomes not quite so simple. And so, I find it's a really fun exercise to build when trying out a new interactive front-end technology.
STEPHANIE: Have you done this with a different implementation or a framework?
JOËL: I've done this, not completely, but I've attempted some parts of a D&D character sheet, I think, with Backbone.js with Ember. I may have done an Angular one at some point in original Angular, so Angular 1. I did this with Elm. Somehow, I skipped React. I don't think I did React to build a D&D character sheet. And now I'm kind of moving a little bit back to the backend. How much can we get done just with Turbo? Or do we need to pull in maybe Stimulus? These are all things that are going to be really fun to demonstrate.
STEPHANIE: Yeah. Speaking of injecting some whimsy earlier, I think it's kind of like just a little more fun than a regular to-do app, you know, or a blog to show how you can build, you know, something that people kind of understand with a different technology.
JOËL: Another really fun thing that I've been toying with this week has been using AI to help me refactor code. And this has been using just sort of a classic chat AI, not a tool like Copilot. And I was dealing with a query that was really slow, and I wanted to restructure it in a different way. And I described to the AI how I wanted it to refactor and explicitly said, "I want this to be the same before and after." And I asked it to do the refactor, and it gave me some pretty disappointing results where it did some, like, a couple of really obvious things that were not that useful.
And I was talking to a colleague about how I was really disappointed. I was thinking, well, AI should be able to do something better than this. And this colleague suggested changing the way I was asking for things and specifically asking for a step-by-step and asking it to prove every step using relational algebra, which is the branch of math that deals with everything that underlies relational databases, so the transformations that you would do where you keep everything the same, but you're saying, "Hey, these equations are all equivalent." And it sure did. It gave me a, like, 10-step process with all these, like, symbols and things.
My relational algebra is not that strong, and so I couldn't totally follow along. But then I asked it to give me a code example, like, show me the SQL at every step of this transformation and at the end. And, you know, it all kind of looked all right. I've not fully tested the final result it gave me to see if it does what it says on the tin. But I'm cautiously optimistic. I think it looks very similar to something that I came up with on my own. And so, I'm somewhat impressed, at least, like, much better than things were in the beginning with that first round. So, I'm really curious to see where I can take this.
STEPHANIE: Yeah, I think that's cool that you were able to prompt it differently and get something more useful. One of the reasons why I personally have been a little bit hesitant to get into the large language models is because I would love to see the AI show its work, essentially, like, tell me a little bit more about how it got from question to answer. And I thought that framing of kind of step-by-step show me code was a really interesting way, even to just, like, get some different results that do the same thing.
But you can kind of evaluate that a little bit more on your own rather than just using that first result that it gave you that was like, eh, like, I don't know if this really did anything for me. So, it would be cool, even if you don't end up using, like, the final one, right? If something along the way also is an improvement from what you started with that would be really interesting.
JOËL: Honestly, I think you kind of want the same thing if you're chatting with an AI chatbot or having a conversation in Slack with a colleague. They're just like, "Hey, can you help me refactor this?" And then a sort of, like, totally different chunk of code. And it's just like, "Trust me, it works."
STEPHANIE: [laughs].
JOËL: And maybe it does. Maybe you plug it into your codebase and run the tests against it, and the tests are still green. And so, you trust that it works, but you don't really understand where it came from. That doesn't always feel good, even when it comes from a human. So, what I've appreciated with colleagues has been when they've given me a step-by-step. Sometimes, they give me the final product. They just say, "Hey. Try this. Does this work?" Plug it in to the test. It does pass. It's green. Great. "Tell me what black magic you did to get to that."
And then they give me the step-by-step and it's like, oh, that's so good because not only do I get a better understanding of what happens at every step, but now I'm equipped the next time I run into this problem to apply the same technique to figure it out on my own.
STEPHANIE: Yeah. And I liked, also, that relational algebra pro tip, right? It kind of ensures that what you're getting makes sense or is equivalent along the way [laughs].
JOËL: We think, right? I don't know enough relational algebra to check its work. It is quite possible that it is making some subtle mistakes along the way, or, like, making inferences that it shouldn't be. I'm not going to say I trust that. But I think, specifically, when asking for SQL transformations, prompting it to do so using relational algebra in a step-by-step way seemed to be a way to get it to do something more reliably or at least give more interesting results.
STEPHANIE: Cool.
JOËL: I was interested in trying this out in part because I've been more curious about AI tools recently, and also because we're hoping to do a deeper dive into AI on a Bike Shed episode at some point later, so very much still in the gathering information phase. But this was a really cool experience. So, having an AI refactor a query for me using relational algebra, definitely something that's new in my world this week.
STEPHANIE: Speaking of refactoring and this idea of making improvements to your code and trying to figure out how to get from what you currently have to something new, I have been thinking a lot about how to make code reviews more actionable. And that's because, on my current client project, our team is struggling a little bit with code reviews, especially when you kind of want to give feedback on more of a design change in the code or thinking about some different abstractions. I have found that that is really hard to communicate async and also in a, like, a GitHub code review format where you can really just comment, like, line by line.
And I've found that, you know, when someone is leaving feedback, that's like, "I'm having a hard time reading this. And I'm imagining that we could organize the code a bit differently in these three different layers or abstractions," there's a lot of assumptions there, right [laughs]? That your message is being communicated to the author and that they are able to, like, visualize, or have a mental model for what you're explaining as well.
And then kind of what I've been seeing in this dynamic is, like, not really knowing what to do with that and to kind of just, like, I don't know where to go from here. So, I guess the next step is just to, like, merge it. Is that something you've experienced before or encountered when it comes to feedback?
JOËL: Broader changes are often challenging to explain, especially when they're...sometimes you get so abstract you can just write a quick paragraph. And sometimes it's like, hey, what if we, like, totally change our approach? I've definitely done the thing where I'll just ping someone and say, "Hey, can we talk about this synchronously? Can we get on a call and have a deeper conversation?"
How do you tend to approach if you're not going to hop on a call with someone and, like, have a 20 or 30-minute conversation? How do you approach doing that asynchronously on a pull request? Are you the type of person to put, like, a ton of, like, code blocks, like, "Here's what I was thinking. We could instead have this class and this thing"? And, like, pretty soon, it's, like, a page and a half of text. Or do you have another approach that you like to use?
STEPHANIE: Yeah. And I think that's where it can get really interesting. Because my process is, I'll usually just start commenting and maybe if I'm seeing some things that can be done differently. If it's not just, like, a really obvious change that I could just use English to describe, I'll add a little suggested change. But I also don't want to just rewrite this person's code [laughs] in a code review.
JOËL: That's the challenge, right?
STEPHANIE: Yeah. And I've definitely seen that be done before, too. Once I notice I'm at, like, four plus comments, and then they're not just, like, nitpicks about, like, syntax or something like that, that helps me clue into the idea that there is some kind of bigger change that I might be asking of the author. And I don't want to overwhelm them with, like, individual comments that really are trying to convey something more holistic.
JOËL: Right. I wonder if having a, like, specialized yet more abstract language is useful for these sorts of things where a whole paragraph in English or, you know, a ton of code examples might be a bit much. If you're able to say something like, "Hey, how would you feel about using a strategy pattern approach here instead of, you know, maybe a template object or some custom thing that we've built here?" that allows us to say a lot in a fairly sort of terse way. And it's the thing that you can leave more generically on the PR instead of, like, individually commenting in a bunch of places. And that can start a broader conversation at more of an architecture level.
STEPHANIE: Yes, I really like that. That's a great idea. I would follow that up with, like, I think at the end of the day, there are some conversations that do need to be had synchronously. And so, I like the idea of leaving a comment like that and just kind of giving them resources to learn what a strategy pattern is and then offering support because that's also a way to shorten that feedback loop of trying to communicate an idea. And I like that it's kind of guiding them, but also you're there to add some scaffolding if it ends up being, like, kind of a big ask for them to figure out what to do.
JOËL: There's also oftentimes, I think, a tone thing to manage where, especially if there's a difference in seniority or experience between the two people, it can be very easy for something to come across as an ask or a demand rather than a like, "Hey, let's think about some alternatives here." Or, like, "I have some concerns with your implementation. Let's sort of broadly explore some possible alternatives. Maybe a strategy pattern works."
But the person reading that who wrote the original code might be, like, receiving that as "Your code is bad. You should have done a strategy pattern instead." And that's not the conversation I want to have, right? I want to have a back-and-forth about, "Hey, what are the trade-offs involved? Do you have a third architecture you'd like to suggest?" And so, that can be a really tricky thing to avoid.
STEPHANIE: Yeah, I like that what you're saying also kind of suggested that it's okay if you don't have an idea yet for exactly how it should look like. Maybe you just are like, oh, like, I'm having a hard time understanding this, but I don't think just leaving it at that gives the author a lot to go on.
I think there's something to it about maybe the action part of actionable is just like, "Can you talk about it with me?" Or "Could you explain what you're trying to do here?" Or, you know, leave a comment about what this method is doing. There's a lot of ways, I think, that you can reach some amount of improvement, even if it doesn't end up being, like, the ideal code that you would write.
JOËL: Yes. There's also maybe a distinction in making it actionable by giving someone some code and saying, "Hey, you should copy-paste this code and make that..." or, you know, use a GitHub suggested code or something, which works on the small. And in the big, you can give some maybe examples and say, "Hey, what if you refactored in this way?"
But sometimes, you could even step back and let them do that work and say, "Hey, I have some concerns with the current architecture. It's not flexible in the ways that we need to be flexible. Here's my understanding of the requirements. And here's sort of how I see maybe this architecture not working with that. Let's think of some different ways we could approach this problem." And oftentimes, it's nice to give at least one or two different ideas to help start that. But it can be okay to just ask the person, "Hey, can you come up with some alternate implementations that would fulfill these sets of requirements?"
STEPHANIE: Yeah, I like that. And I can even see, like, maybe you do that work, and you don't end up pursuing it completely in addressing that feedback. But even asking someone to do the exercise itself, I think, can then spark new ideas and maybe other improvements.
In general, I like to think about...I'm a little hesitant to use this metaphor because I'm not actually giving code, like, letter grades when I review them, but the idea that, like, not all code has to get, like, an A [chuckles], but maybe getting it, like, from one letter grade up to, like, half a letter grade, like, higher, that is valuable, even if it's not always practical to go through multiple rounds of code review. And I think just making it actionable enough to be a little bit better, like, that is, in my opinion, the sweet spot.
JOËL: That's true. The sort of over-giving feedback to someone to try to get code perfect, rather than just saying, "Hey, can we make it slightly better?" And, you know, there are probably some minimum standards you need to hit. But at some point, it's a trade-off of like, how much time do we need to put polishing this versus shipping something?
STEPHANIE: Yeah, and I think that it is cumulative over time, right? That's how people learn. Yeah, it's like one of the biggest opportunities for developers to level up is from that feedback. And that's why I think it's important that it's actionable because, you know, and you put the time into, like, giving that review, and it's not just to make sure the code works, but it's also, like, one of the touch points for collaboration.
JOËL: So, if you had to summarize what makes code review comments actionable, do you have, like, top three tips that make a comment really actionable as opposed to something that's not helpful? Or maybe that's more of the journey that you're on, and you've not distilled it down to three pithy tips that you can put in a listicle.
STEPHANIE: Honestly, I think it does kind of just distill down to one, which is for every comment, you should have an idea of what you would like the author to do about it. And it's okay if it's nothing, but then tell them that it's nothing. You could just be expressing, "I thought this was kind of weird, [laughs]" or "This is not my favorite thing, but it's okay."
JOËL: And it can be okay for the thing you want the author to do. It doesn't have to be code. It could be a conversation.
STEPHANIE: Yeah, exactly. It could be a conversation. It could be asking for information, too, right? Like, "Did you consider alternatives, and could you share them with me?" But that request portion, I think is really important because, yeah, I think there's so much miscommunication that can happen along the way. So, definitely still trying to figure out how to best support that kind of code review culture on my team.
JOËL: This week's episode has been really fun because it's just been a combination of a lot of things that are new in our world, things that we've been trying, things that we've been learning. And kind of in an almost, like, a meta sense, one of the things I've been digging into is combinatorics, the branch of math that looks at how things combine and particularly how it works with combining a bunch of ActiveRecord query fragments where there's potential branching, so things like doing a union of two sort of sub queries or doing an or where you're combining two different where queries and trying to figure out what are the different paths through that.
STEPHANIE: Wow, what a great way to combine what we were talking about, Joël [laughs]. Did you apply combinatorics to this podcast episode [laughs]?
JOËL: Somehow, topics multiply with each other, something, something.
STEPHANIE: Yeah, that makes sense to me [laughs]. Okay. Will you tell me more about what you've been using it for in your queries?
JOËL: So, one thing I'm trying to do is because I've got these different branching paths through a query, I want to see sort of all the different ways because these are defined as ActiveRecord scopes, and I'm chaining them together. And it looks linear because I'm calling scope1 dot scope2 dot scope3. But each of those have branches inside of them. And so, there's all these different ways that data could get used or not.
And one way that I figured out, like, what are the different paths here, was actually drawing out a matrix, just putting together a table. In this case, I had two scopes, each of which had a two-way branch inside, and so I made a two by two matrix. And that gave me all of the combinations of, oh, if you go down one branch in one scope and down another branch in the other scope. And what I went through is then I went in in each square and filled in how many records I would expect to get back from the query from some basic set that I was working on in each of these combinations.
And one thing that was really interesting is that some of those combinations were sort of mutually exclusive, where a scope further down the line was filtering on the same field as an earlier one and would overwrite it or not overwrite it, but the two would then sort of you can't have both of those things be true at the same time. So, I'm looking for something that has a particular manager ID, and then I'm looking for something that has a particular different manager ID. And the way Rails combines these, if you just change scopes with where, is to and them together. There are no records that have both manager ID 1 and manager ID 2. You can only have one manager ID.
And so, as I'm filling out my matrix, there's some sections I can just zero out and be like, wait, this will always return zero record. And then I can start focusing on the parts that are not zeroed out. So, I've got two or three squares. What's special about those? And that helped me really understand what the combination of these multiple query fragments together were actually trying to do as a holistic whole.
STEPHANIE: Wow, yeah, that is really interesting because I hear you when you say it looks linear. And it would be really surprising to me for there to be branching paths. Like, that's not really what I think about when I think about SQL. But that makes a lot of sense that it could get so complicated that it's just impossible to get a certain kind of result. Like, what's going to be the outcome of applying combinatorics to this? Is there a refactoring opportunity, or is it really just to even understand what's going on?
JOËL: So, this was a refactoring that I was trying to do, but I didn't really understand the underlying behavior of the chain of scopes. I just knew that they were doing some complex things that were inefficient from a SQL perspective. And so, I was looking at ways to refactor, but I also wanted to get a sense of what is this actually trying to do other than just chaining a bunch of random bits of code together? So, the matrix really helped for that.
The other way that I used it was to write some tests because this query I was trying to refactor, this chain of scopes, was untested. And I wanted to write tests that were very thorough because I wanted to make sure that my refactor didn't break any edge cases. And I'm, you know, writing a few tests. Okay, well, here's a record that I definitely want to get returned by this query, and maybe here are a couple of records I don't want to get returned.
And the more I was, like, going into this and trying to write test cases, the more I was finding more edge cases that I didn't want to and, oh, but what about this? And what about the combination of these things? And it got to the point where it was just messing with my mind. I was, like, confusing myself and really struggling to write tests that would do anything useful.
STEPHANIE: Wow. Yeah. Honestly, I have already started to become a little bit suspicious of complex scopes, and this further pushes me in that direction [laughs] because yeah, once you start to...like, the benefit of them is that you can chain them, but it really hides a lot of the underlying behavior. So, you can easily just turn yourself around or, like, go, you know, kind of end up [laughs] in a little bit of a bind.
JOËL: Definitely, especially once it grows a little bit harder to hold in your head. And I don't know exactly where that level is for me. But in this particular situation, I identified, I think, five different dimensions that would impact the results of this query. And then each dimension had maybe three or four different values that we might care about. And, eventually, I just took the time to write this out.
So, I created five arrays and then just said, "Hey, here are the different managers that we care about. Here are the different project types we care about. Here are the different..." and we had, like, five of these, and each array had three or four elements in it. And then, in a series of nested loops, I iterated through all of these arrays and at the innermost loop, created the data that I wanted that matched that particular set of values.
Now, we're often told you should not be doing things in nested loops because you end up sort of multiplying all of these together, but, in this case, this is actually what I wanted to do. You know, it turns out that I had a hundred-ish records I had to create to sort of create a data set that would be all the possible edge cases I might want to filter on. And creating them all by hand with all of the different variations was going to be too much. And so, I ended up doing this with arrays and nested loops. And it got me the data that I needed. And it gave me then the confidence to know that my refactor did indeed work the way I was expecting.
STEPHANIE: Wow. That's truly hero's work [chuckles]. I'm, like, very excited because it sounds like that's a huge opportunity for some performance improvements as well.
JOËL: For the underlying code, yes. The test might be a little bit slow because I'm creating a hundred records in the database. And you might say, "Oh, do you really need to do that? Can you maybe collapse some of these cases?" In this particular case, I really wanted to have high confidence that the refactor was not changing anything. And so, I was okay creating a hundred records over a series of nested iterations. That was a price I was willing to pay. The refactored query, it turns out, I was able to write it in a way that was significantly faster.
STEPHANIE: Yeah, that's what I suspected.
JOËL: So, I had to rewrite it in a way that didn't take advantage of all the change scopes. I had to just sort of write something custom from scratch, which is often the case, right? Performance and reusability sometimes fight against each other, and it's a trade-off. So, I'm not reusing the scopes. I had to write something from scratch, but it's multiple hundreds of times faster.
STEPHANIE: Wow. Yeah. That seems worth it for a slow test [laughs] for the user experience to be a lot better, especially when you just reach that level of complexity. And it's a really awesome strategy that you applied to figure that out. I think it's a very unique one [laughs]. That's for sure.
JOËL: I've had an interest in sort of analytical tools to help me understand domain models, to help understand problems, to help understand code that I'm working with for a while now, and I think an understanding of combinatorics fits into that. And then, particular tools within that, such as drawing things out in a table, in a two by two matrix, or an end-by-end matrix to get something visual, that's a great tool for debugging or understanding a problem.
Thinking of problems as data that exists in multiple dimensions and then asking about the cardinality of that set it's the kind of analysis I did a lot when I was modeling using algebraic data types in Elm. But now I've sort of taken some of the tools and analysis I use from that world into thinking about things like SQL records, things like dealing with data in Ruby. And I'm able to bring those tools and that way of thinking to help me solve some problems that I might struggle to solve otherwise.
For any of our listeners who this, like, kind of piques their interest, combinatorics falls under a broader umbrella of mathematics called discrete math. And within that, there's a lot that I think is really useful, a lot of tools and techniques that we can apply to our day-to-day programming. We have a Bike Shed episode where we talked about is discrete math relevant to day-to-day programmers and what are the ways it's so? We'll link that in the show notes. I also gave a talk at RailsConf last year diving into that titled: The Math Every Programmer Needs. So, if you're looking for something that's accessible to someone who's not done a math degree, those are two great jumping-off points.
STEPHANIE: Yeah. And then, maybe you'll start drawing out arrays and applying combinatorics to figure out your performance problems.
JOËL: On that note, shall we wrap up?
STEPHANIE: Let's wrap up. Show notes for this episode can be found at bikeshed.fm.
JOËL: This show has been produced and edited by Mandy Moore.
STEPHANIE: If you enjoyed listening, one really easy way to support the show is to leave us a quick rating or even a review in iTunes. It really helps other folks find the show.
JOËL: If you have any feedback for this or any of our other episodes, you can reach us @_bikeshed, or you can reach me @joelquen on Twitter.
STEPHANIE: Or reach both of us at hosts@bikeshed.fm via email.
JOËL: Thanks so much for listening to The Bike Shed, and we'll see you next week.
ALL: Byeeeeeeeeee!!!!!!!!
AD:
Did you know thoughtbot has a referral program? If you introduce us to someone looking for a design or development partner, we will compensate you if they decide to work with us.
More info on our website at: tbot.io/referral. Or you can email us at: referrals@thoughtbot.com with any questions.
439 episodes
Manage episode 407681067 series 1401614
Stephanie introduces her ideal setup for enjoying coffee on a bike ride. Joël describes his afternoon tea ritual. Exciting news from the hosts: both have been accepted to speak at RailsConf! Stephanie's presentation, titled "So, Writing Tests Feels Painful. What now?" aims to tackle the issues developers encounter with testing while offering actionable advice to ease these pains. Joël's session will focus on utilizing Turbo to create a Dungeons & Dragons character sheet, combining his passion for gaming with technical expertise.
Their conversation shifts to artificial intelligence and its potential in code refactoring and other applications, such as enhancing the code review process and solving complex software development problems. Joël shares his venture into combinatorics, illustrating how this mathematical approach helped him efficiently refactor a database query by systematically exploring and testing all potential combinations of query segments.
Transcript:
JOËL: Hello and welcome to another episode of The Bike Shed, a weekly podcast from your friends at thoughtbot about developing great software. I'm Joël Quenneville.
STEPHANIE: And I'm Stephanie Minn, and together, we're here to share a bit of what we've learned along the way.
JOËL: So, Stephanie, what's new in your world?
STEPHANIE: So, today I went out for a coffee on my bike, and I feel like I finally have my perfect, like, on-the-go coffee setup. We have this thoughtbot branded travel mug. So, it's one of the little bits of swag that we got from the company. It's, like, perfectly leak-proof. I'll link the brand in the show notes. But it's perfectly leak-proof, which is great. And on my bike, I have a little stem bag, so it's just, like, a tiny kind of, like, cylindrical bag that sits on the, like, vertical part of my handlebars that connects to the rest of my bag. And it's just, like, the perfect size for a 12-ounce coffee.
And so, I put my little travel mug in there, and I just had a very refreshing morning. And I'd gone out on my bike for a little bit, stopping by for coffee and headed home to work. And I got to drink my coffee during my first meeting. So, it was a wonderful way to start the day.
JOËL: Do you just show up at the coffee shop with your refillable mug and say, "Hey, can you pour some coffee in this?"
STEPHANIE: Yeah. I think a lot of coffee places are really amenable to bringing your own travel mugs. So yeah, it's really nice because I get to use less plastic. And also, you know, when you get a to-go mug, it is not leak-proof, right? It could just slosh all over the place and spill, so not bike-friendly. But yeah, bring your own mug. It's very easy.
JOËL: Excellent.
STEPHANIE: So, Joël, what's new in your world?
JOËL: Also, warm beverages. Who would have thought? It's almost like it's cold in North America or something. I've been really enjoying making myself tea in the afternoons recently. And I've been drinking this brand of tea that is a little bit extra. Every flavor of tea they have comes with a description of how the tea feels.
STEPHANIE: Ooh.
JOËL: I don't know who came up with these, but they're kind of funny. So, one that I particularly enjoy is described as feels like stargazing on an empty beach.
STEPHANIE: Wow. That's very specific.
JOËL: They also give you tasting notes. This one has tastes of candied violet, elderberry, blackberry, and incense.
STEPHANIE: Ooh, that sounds lovely. Are you drinking, like, herbal tea in the afternoon, or do you drink caffeinated tea?
JOËL: I'll do caffeinated tea. I limit myself to one pot of coffee that I brew in the morning, and then, whenever that's done, I switch to tea. Tea I allow myself anything: herbal, black tea; that's fine.
STEPHANIE: Yeah, I can't have too much caffeine in the afternoon either. But I do love an extra tea. I wish I could remember, like, what even was in this tea or what brand it was, but once I had a tea that was a purplish color. But then, when you squeeze some lemon in it, or I guess maybe anything with a bit of acid, it would turn blue.
JOËL: Oh, that's so cool.
STEPHANIE: Yeah, I'll have to find what this tea was [laughs] and update the podcast for any tea lovers out there. But yeah, it was just, like, a little bit of extra whimsy to your regular routine.
JOËL: I love adding a little whimsy to my day, even if it's just seeing a random animated GIF that a coworker has sent or Tuple has some of the, like, reactions you can send if you're pairing with someone. And I don't use those very often, so whenever one of those comes through, and it's like, ship it or yay, that makes me very happy.
STEPHANIE: Agreed.
JOËL: This week is really fun because as we were prepping for this episode, we both realized that there is a lot that's been new in our world recently. And Stephanie, in particular, you've got some pretty big news that recently happened to you.
STEPHANIE: Yeah, it turns out we're making the what's new in your world segment the entire episode today [laughs]. But my news is that I am speaking at RailsConf this year, so that is May 7th through 9th in Detroit. And so, yeah, I haven't spoken at a RailsConf before, only a RubyConf. So, I'm looking forward to it. My talk is called: So, Writing Tests Feels Painful. What now?
JOËL: Wait, is writing tests ever painful [laughs]?
STEPHANIE: Maybe not for you, but for the rest of us [laughs].
JOËL: No, it absolutely is. I, right before this recording, came from a pairing session where we were scratching our heads on an, like, awkward-to-write test. It happens to all of us.
STEPHANIE: Yeah. So, I was brainstorming topics, and I kind of realized, especially with a lot of our consulting experience, you know, we hear from developers or even maybe, like, engineering managers a lot of themes around like, "Oh, like, development is slowing down because our test suite is such a headache," or "It's really slow. It's really flaky. It's really complicated." And that is a pain point that a lot of tech leaders are also looking to address for their teams.
But I was really questioning this idea that, like, it always had to be some effort to improve the test suite, like, that had to be worked on at some later point or get, like, an initiative together to fix all of these problems, and that it couldn't just be baked into your normal development process, like, on an individual level. I do think it is really easy to feel a lot of pain when trying to write tests and then just be like, ugh, like, I wish someone would fix this, right? Or, you know, just kind of ignore the signals of that pain because you don't know, like, how to manage it yourself.
So, my talk is about when you do feel that pain, really trying to determine if there's anything you can do, even in just, like, the one test file that you're working in to make things a little bit easier for yourself, so it doesn't become this, like, chronic issue that just gets worse and worse. Is there something you could do to maybe reorganize the file as you're working in it to make some conditionals a little bit clearer?
Is there any, like, extra test setup that you're like, "Oh, actually, I don't need this anymore, and I can just start to get rid of it, not just for this one example, but for the rest in this file"? And do yourself a favor a little bit. So yeah, I'm excited to talk about that because I think that's perhaps, like, a skill that we don't focus enough on.
JOËL: Are you going to sort of focus in on the side of things where, like, a classic TDD mantra is that test pain reflects underlying code complexity? So, are you planning to focus on the idea of, oh, if you're feeling test pain, maybe take some time to refactor some of the code that's under test, maybe because there's some tight coupling? Or are you going to lean a little bit more into maybe, like, the Boy Scout rule, you know, 'Leave the campsite cleaner than you found it' for your test files?
STEPHANIE: Ooh, I like that framing. Definitely more of the former. But one thing I've also noticed working with a lot of client teams is that it's not always clear, like, how to refactor. I think a lot of intermediate developers start to feel that pain but don't know what to do about it. They don't know, like, maybe the code smells, or the patterns, or refactoring strategies, and that can certainly be taught. It will probably pull from that. But even if you don't know those skills yet, I'm wondering if there's, like, an opportunity to teach, like, developers at that level to start to reflect on the code and be like, "Hmm, what could I do to make this a little more flexible?"
And they might not know the names of the strategies to, like, extract a class, but just start to get them thinking about it. And then maybe when they come across that vocabulary later, it'll connect a lot easier because they'll have started to think about, you know, their experiences day to day with some of the more conceptual stuff.
JOËL: I really like that because I feel we've probably all heard that idea that test pain, especially when you're test driving, is a sign of maybe some anti-patterns or some code smells in the underlying code that you're testing. But translating that into something actionable and being able to say, "Okay, so my tests are painful. They're telling me something needs to be refactored. I'm looking at this code, and I don't know what to refactor." It's a big jump. It's almost the classic draw two circles; draw the rest of the owl meme. And so, I think bridging that gap is something that is really valuable for our community.
STEPHANIE: Yeah, that's exactly what I hope to do in my talk. So, Joël, you [chuckles] also didn't quite mention that you have big news as well.
JOËL: So, I also got accepted to speak at RailsConf. I'm giving a talk on Building a Dungeons & Dragons Character Sheet Using Turbo.
STEPHANIE: That's really awesome. I'm excited because I want to learn more about Turbo. I want someone else to tell me [laughs] what I can do with it. And as a person with a little bit of Dungeons & Dragons experience, I think a character sheet is kind of the perfect vehicle for that.
JOËL: Building a D&D character sheet has been kind of my go-to project to experiment with a new front-end framework because it's something that's pretty dynamic. And for those who don't know, there's a bunch of fields that you fill in with stats for different attributes that your character has, but then those impact other stats that get rendered. And sometimes there can be a chain two or three long where different numbers kind of combine together. And so, you've got this almost dependency tree of, like, a particular number.
Maybe your skill at acrobatics might depend on a number that you entered in the dexterity field, but it also depends on your proficiency bonus, and maybe also depends on the race that you picked and a few other things. And so, calculating those numbers all of a sudden becomes not quite so simple. And so, I find it's a really fun exercise to build when trying out a new interactive front-end technology.
STEPHANIE: Have you done this with a different implementation or a framework?
JOËL: I've done this, not completely, but I've attempted some parts of a D&D character sheet, I think, with Backbone.js with Ember. I may have done an Angular one at some point in original Angular, so Angular 1. I did this with Elm. Somehow, I skipped React. I don't think I did React to build a D&D character sheet. And now I'm kind of moving a little bit back to the backend. How much can we get done just with Turbo? Or do we need to pull in maybe Stimulus? These are all things that are going to be really fun to demonstrate.
STEPHANIE: Yeah. Speaking of injecting some whimsy earlier, I think it's kind of like just a little more fun than a regular to-do app, you know, or a blog to show how you can build, you know, something that people kind of understand with a different technology.
JOËL: Another really fun thing that I've been toying with this week has been using AI to help me refactor code. And this has been using just sort of a classic chat AI, not a tool like Copilot. And I was dealing with a query that was really slow, and I wanted to restructure it in a different way. And I described to the AI how I wanted it to refactor and explicitly said, "I want this to be the same before and after." And I asked it to do the refactor, and it gave me some pretty disappointing results where it did some, like, a couple of really obvious things that were not that useful.
And I was talking to a colleague about how I was really disappointed. I was thinking, well, AI should be able to do something better than this. And this colleague suggested changing the way I was asking for things and specifically asking for a step-by-step and asking it to prove every step using relational algebra, which is the branch of math that deals with everything that underlies relational databases, so the transformations that you would do where you keep everything the same, but you're saying, "Hey, these equations are all equivalent." And it sure did. It gave me a, like, 10-step process with all these, like, symbols and things.
My relational algebra is not that strong, and so I couldn't totally follow along. But then I asked it to give me a code example, like, show me the SQL at every step of this transformation and at the end. And, you know, it all kind of looked all right. I've not fully tested the final result it gave me to see if it does what it says on the tin. But I'm cautiously optimistic. I think it looks very similar to something that I came up with on my own. And so, I'm somewhat impressed, at least, like, much better than things were in the beginning with that first round. So, I'm really curious to see where I can take this.
STEPHANIE: Yeah, I think that's cool that you were able to prompt it differently and get something more useful. One of the reasons why I personally have been a little bit hesitant to get into the large language models is because I would love to see the AI show its work, essentially, like, tell me a little bit more about how it got from question to answer. And I thought that framing of kind of step-by-step show me code was a really interesting way, even to just, like, get some different results that do the same thing.
But you can kind of evaluate that a little bit more on your own rather than just using that first result that it gave you that was like, eh, like, I don't know if this really did anything for me. So, it would be cool, even if you don't end up using, like, the final one, right? If something along the way also is an improvement from what you started with that would be really interesting.
JOËL: Honestly, I think you kind of want the same thing if you're chatting with an AI chatbot or having a conversation in Slack with a colleague. They're just like, "Hey, can you help me refactor this?" And then a sort of, like, totally different chunk of code. And it's just like, "Trust me, it works."
STEPHANIE: [laughs].
JOËL: And maybe it does. Maybe you plug it into your codebase and run the tests against it, and the tests are still green. And so, you trust that it works, but you don't really understand where it came from. That doesn't always feel good, even when it comes from a human. So, what I've appreciated with colleagues has been when they've given me a step-by-step. Sometimes, they give me the final product. They just say, "Hey. Try this. Does this work?" Plug it in to the test. It does pass. It's green. Great. "Tell me what black magic you did to get to that."
And then they give me the step-by-step and it's like, oh, that's so good because not only do I get a better understanding of what happens at every step, but now I'm equipped the next time I run into this problem to apply the same technique to figure it out on my own.
STEPHANIE: Yeah. And I liked, also, that relational algebra pro tip, right? It kind of ensures that what you're getting makes sense or is equivalent along the way [laughs].
JOËL: We think, right? I don't know enough relational algebra to check its work. It is quite possible that it is making some subtle mistakes along the way, or, like, making inferences that it shouldn't be. I'm not going to say I trust that. But I think, specifically, when asking for SQL transformations, prompting it to do so using relational algebra in a step-by-step way seemed to be a way to get it to do something more reliably or at least give more interesting results.
STEPHANIE: Cool.
JOËL: I was interested in trying this out in part because I've been more curious about AI tools recently, and also because we're hoping to do a deeper dive into AI on a Bike Shed episode at some point later, so very much still in the gathering information phase. But this was a really cool experience. So, having an AI refactor a query for me using relational algebra, definitely something that's new in my world this week.
STEPHANIE: Speaking of refactoring and this idea of making improvements to your code and trying to figure out how to get from what you currently have to something new, I have been thinking a lot about how to make code reviews more actionable. And that's because, on my current client project, our team is struggling a little bit with code reviews, especially when you kind of want to give feedback on more of a design change in the code or thinking about some different abstractions. I have found that that is really hard to communicate async and also in a, like, a GitHub code review format where you can really just comment, like, line by line.
And I've found that, you know, when someone is leaving feedback, that's like, "I'm having a hard time reading this. And I'm imagining that we could organize the code a bit differently in these three different layers or abstractions," there's a lot of assumptions there, right [laughs]? That your message is being communicated to the author and that they are able to, like, visualize, or have a mental model for what you're explaining as well.
And then kind of what I've been seeing in this dynamic is, like, not really knowing what to do with that and to kind of just, like, I don't know where to go from here. So, I guess the next step is just to, like, merge it. Is that something you've experienced before or encountered when it comes to feedback?
JOËL: Broader changes are often challenging to explain, especially when they're...sometimes you get so abstract you can just write a quick paragraph. And sometimes it's like, hey, what if we, like, totally change our approach? I've definitely done the thing where I'll just ping someone and say, "Hey, can we talk about this synchronously? Can we get on a call and have a deeper conversation?"
How do you tend to approach if you're not going to hop on a call with someone and, like, have a 20 or 30-minute conversation? How do you approach doing that asynchronously on a pull request? Are you the type of person to put, like, a ton of, like, code blocks, like, "Here's what I was thinking. We could instead have this class and this thing"? And, like, pretty soon, it's, like, a page and a half of text. Or do you have another approach that you like to use?
STEPHANIE: Yeah. And I think that's where it can get really interesting. Because my process is, I'll usually just start commenting and maybe if I'm seeing some things that can be done differently. If it's not just, like, a really obvious change that I could just use English to describe, I'll add a little suggested change. But I also don't want to just rewrite this person's code [laughs] in a code review.
JOËL: That's the challenge, right?
STEPHANIE: Yeah. And I've definitely seen that be done before, too. Once I notice I'm at, like, four plus comments, and then they're not just, like, nitpicks about, like, syntax or something like that, that helps me clue into the idea that there is some kind of bigger change that I might be asking of the author. And I don't want to overwhelm them with, like, individual comments that really are trying to convey something more holistic.
JOËL: Right. I wonder if having a, like, specialized yet more abstract language is useful for these sorts of things where a whole paragraph in English or, you know, a ton of code examples might be a bit much. If you're able to say something like, "Hey, how would you feel about using a strategy pattern approach here instead of, you know, maybe a template object or some custom thing that we've built here?" that allows us to say a lot in a fairly sort of terse way. And it's the thing that you can leave more generically on the PR instead of, like, individually commenting in a bunch of places. And that can start a broader conversation at more of an architecture level.
STEPHANIE: Yes, I really like that. That's a great idea. I would follow that up with, like, I think at the end of the day, there are some conversations that do need to be had synchronously. And so, I like the idea of leaving a comment like that and just kind of giving them resources to learn what a strategy pattern is and then offering support because that's also a way to shorten that feedback loop of trying to communicate an idea. And I like that it's kind of guiding them, but also you're there to add some scaffolding if it ends up being, like, kind of a big ask for them to figure out what to do.
JOËL: There's also oftentimes, I think, a tone thing to manage where, especially if there's a difference in seniority or experience between the two people, it can be very easy for something to come across as an ask or a demand rather than a like, "Hey, let's think about some alternatives here." Or, like, "I have some concerns with your implementation. Let's sort of broadly explore some possible alternatives. Maybe a strategy pattern works."
But the person reading that who wrote the original code might be, like, receiving that as "Your code is bad. You should have done a strategy pattern instead." And that's not the conversation I want to have, right? I want to have a back-and-forth about, "Hey, what are the trade-offs involved? Do you have a third architecture you'd like to suggest?" And so, that can be a really tricky thing to avoid.
STEPHANIE: Yeah, I like that what you're saying also kind of suggested that it's okay if you don't have an idea yet for exactly how it should look like. Maybe you just are like, oh, like, I'm having a hard time understanding this, but I don't think just leaving it at that gives the author a lot to go on.
I think there's something to it about maybe the action part of actionable is just like, "Can you talk about it with me?" Or "Could you explain what you're trying to do here?" Or, you know, leave a comment about what this method is doing. There's a lot of ways, I think, that you can reach some amount of improvement, even if it doesn't end up being, like, the ideal code that you would write.
JOËL: Yes. There's also maybe a distinction in making it actionable by giving someone some code and saying, "Hey, you should copy-paste this code and make that..." or, you know, use a GitHub suggested code or something, which works on the small. And in the big, you can give some maybe examples and say, "Hey, what if you refactored in this way?"
But sometimes, you could even step back and let them do that work and say, "Hey, I have some concerns with the current architecture. It's not flexible in the ways that we need to be flexible. Here's my understanding of the requirements. And here's sort of how I see maybe this architecture not working with that. Let's think of some different ways we could approach this problem." And oftentimes, it's nice to give at least one or two different ideas to help start that. But it can be okay to just ask the person, "Hey, can you come up with some alternate implementations that would fulfill these sets of requirements?"
STEPHANIE: Yeah, I like that. And I can even see, like, maybe you do that work, and you don't end up pursuing it completely in addressing that feedback. But even asking someone to do the exercise itself, I think, can then spark new ideas and maybe other improvements.
In general, I like to think about...I'm a little hesitant to use this metaphor because I'm not actually giving code, like, letter grades when I review them, but the idea that, like, not all code has to get, like, an A [chuckles], but maybe getting it, like, from one letter grade up to, like, half a letter grade, like, higher, that is valuable, even if it's not always practical to go through multiple rounds of code review. And I think just making it actionable enough to be a little bit better, like, that is, in my opinion, the sweet spot.
JOËL: That's true. The sort of over-giving feedback to someone to try to get code perfect, rather than just saying, "Hey, can we make it slightly better?" And, you know, there are probably some minimum standards you need to hit. But at some point, it's a trade-off of like, how much time do we need to put polishing this versus shipping something?
STEPHANIE: Yeah, and I think that it is cumulative over time, right? That's how people learn. Yeah, it's like one of the biggest opportunities for developers to level up is from that feedback. And that's why I think it's important that it's actionable because, you know, and you put the time into, like, giving that review, and it's not just to make sure the code works, but it's also, like, one of the touch points for collaboration.
JOËL: So, if you had to summarize what makes code review comments actionable, do you have, like, top three tips that make a comment really actionable as opposed to something that's not helpful? Or maybe that's more of the journey that you're on, and you've not distilled it down to three pithy tips that you can put in a listicle.
STEPHANIE: Honestly, I think it does kind of just distill down to one, which is for every comment, you should have an idea of what you would like the author to do about it. And it's okay if it's nothing, but then tell them that it's nothing. You could just be expressing, "I thought this was kind of weird, [laughs]" or "This is not my favorite thing, but it's okay."
JOËL: And it can be okay for the thing you want the author to do. It doesn't have to be code. It could be a conversation.
STEPHANIE: Yeah, exactly. It could be a conversation. It could be asking for information, too, right? Like, "Did you consider alternatives, and could you share them with me?" But that request portion, I think is really important because, yeah, I think there's so much miscommunication that can happen along the way. So, definitely still trying to figure out how to best support that kind of code review culture on my team.
JOËL: This week's episode has been really fun because it's just been a combination of a lot of things that are new in our world, things that we've been trying, things that we've been learning. And kind of in an almost, like, a meta sense, one of the things I've been digging into is combinatorics, the branch of math that looks at how things combine and particularly how it works with combining a bunch of ActiveRecord query fragments where there's potential branching, so things like doing a union of two sort of sub queries or doing an or where you're combining two different where queries and trying to figure out what are the different paths through that.
STEPHANIE: Wow, what a great way to combine what we were talking about, Joël [laughs]. Did you apply combinatorics to this podcast episode [laughs]?
JOËL: Somehow, topics multiply with each other, something, something.
STEPHANIE: Yeah, that makes sense to me [laughs]. Okay. Will you tell me more about what you've been using it for in your queries?
JOËL: So, one thing I'm trying to do is because I've got these different branching paths through a query, I want to see sort of all the different ways because these are defined as ActiveRecord scopes, and I'm chaining them together. And it looks linear because I'm calling scope1 dot scope2 dot scope3. But each of those have branches inside of them. And so, there's all these different ways that data could get used or not.
And one way that I figured out, like, what are the different paths here, was actually drawing out a matrix, just putting together a table. In this case, I had two scopes, each of which had a two-way branch inside, and so I made a two by two matrix. And that gave me all of the combinations of, oh, if you go down one branch in one scope and down another branch in the other scope. And what I went through is then I went in in each square and filled in how many records I would expect to get back from the query from some basic set that I was working on in each of these combinations.
And one thing that was really interesting is that some of those combinations were sort of mutually exclusive, where a scope further down the line was filtering on the same field as an earlier one and would overwrite it or not overwrite it, but the two would then sort of you can't have both of those things be true at the same time. So, I'm looking for something that has a particular manager ID, and then I'm looking for something that has a particular different manager ID. And the way Rails combines these, if you just change scopes with where, is to and them together. There are no records that have both manager ID 1 and manager ID 2. You can only have one manager ID.
And so, as I'm filling out my matrix, there's some sections I can just zero out and be like, wait, this will always return zero record. And then I can start focusing on the parts that are not zeroed out. So, I've got two or three squares. What's special about those? And that helped me really understand what the combination of these multiple query fragments together were actually trying to do as a holistic whole.
STEPHANIE: Wow, yeah, that is really interesting because I hear you when you say it looks linear. And it would be really surprising to me for there to be branching paths. Like, that's not really what I think about when I think about SQL. But that makes a lot of sense that it could get so complicated that it's just impossible to get a certain kind of result. Like, what's going to be the outcome of applying combinatorics to this? Is there a refactoring opportunity, or is it really just to even understand what's going on?
JOËL: So, this was a refactoring that I was trying to do, but I didn't really understand the underlying behavior of the chain of scopes. I just knew that they were doing some complex things that were inefficient from a SQL perspective. And so, I was looking at ways to refactor, but I also wanted to get a sense of what is this actually trying to do other than just chaining a bunch of random bits of code together? So, the matrix really helped for that.
The other way that I used it was to write some tests because this query I was trying to refactor, this chain of scopes, was untested. And I wanted to write tests that were very thorough because I wanted to make sure that my refactor didn't break any edge cases. And I'm, you know, writing a few tests. Okay, well, here's a record that I definitely want to get returned by this query, and maybe here are a couple of records I don't want to get returned.
And the more I was, like, going into this and trying to write test cases, the more I was finding more edge cases that I didn't want to and, oh, but what about this? And what about the combination of these things? And it got to the point where it was just messing with my mind. I was, like, confusing myself and really struggling to write tests that would do anything useful.
STEPHANIE: Wow. Yeah. Honestly, I have already started to become a little bit suspicious of complex scopes, and this further pushes me in that direction [laughs] because yeah, once you start to...like, the benefit of them is that you can chain them, but it really hides a lot of the underlying behavior. So, you can easily just turn yourself around or, like, go, you know, kind of end up [laughs] in a little bit of a bind.
JOËL: Definitely, especially once it grows a little bit harder to hold in your head. And I don't know exactly where that level is for me. But in this particular situation, I identified, I think, five different dimensions that would impact the results of this query. And then each dimension had maybe three or four different values that we might care about. And, eventually, I just took the time to write this out.
So, I created five arrays and then just said, "Hey, here are the different managers that we care about. Here are the different project types we care about. Here are the different..." and we had, like, five of these, and each array had three or four elements in it. And then, in a series of nested loops, I iterated through all of these arrays and at the innermost loop, created the data that I wanted that matched that particular set of values.
Now, we're often told you should not be doing things in nested loops because you end up sort of multiplying all of these together, but, in this case, this is actually what I wanted to do. You know, it turns out that I had a hundred-ish records I had to create to sort of create a data set that would be all the possible edge cases I might want to filter on. And creating them all by hand with all of the different variations was going to be too much. And so, I ended up doing this with arrays and nested loops. And it got me the data that I needed. And it gave me then the confidence to know that my refactor did indeed work the way I was expecting.
STEPHANIE: Wow. That's truly hero's work [chuckles]. I'm, like, very excited because it sounds like that's a huge opportunity for some performance improvements as well.
JOËL: For the underlying code, yes. The test might be a little bit slow because I'm creating a hundred records in the database. And you might say, "Oh, do you really need to do that? Can you maybe collapse some of these cases?" In this particular case, I really wanted to have high confidence that the refactor was not changing anything. And so, I was okay creating a hundred records over a series of nested iterations. That was a price I was willing to pay. The refactored query, it turns out, I was able to write it in a way that was significantly faster.
STEPHANIE: Yeah, that's what I suspected.
JOËL: So, I had to rewrite it in a way that didn't take advantage of all the change scopes. I had to just sort of write something custom from scratch, which is often the case, right? Performance and reusability sometimes fight against each other, and it's a trade-off. So, I'm not reusing the scopes. I had to write something from scratch, but it's multiple hundreds of times faster.
STEPHANIE: Wow. Yeah. That seems worth it for a slow test [laughs] for the user experience to be a lot better, especially when you just reach that level of complexity. And it's a really awesome strategy that you applied to figure that out. I think it's a very unique one [laughs]. That's for sure.
JOËL: I've had an interest in sort of analytical tools to help me understand domain models, to help understand problems, to help understand code that I'm working with for a while now, and I think an understanding of combinatorics fits into that. And then, particular tools within that, such as drawing things out in a table, in a two by two matrix, or an end-by-end matrix to get something visual, that's a great tool for debugging or understanding a problem.
Thinking of problems as data that exists in multiple dimensions and then asking about the cardinality of that set it's the kind of analysis I did a lot when I was modeling using algebraic data types in Elm. But now I've sort of taken some of the tools and analysis I use from that world into thinking about things like SQL records, things like dealing with data in Ruby. And I'm able to bring those tools and that way of thinking to help me solve some problems that I might struggle to solve otherwise.
For any of our listeners who this, like, kind of piques their interest, combinatorics falls under a broader umbrella of mathematics called discrete math. And within that, there's a lot that I think is really useful, a lot of tools and techniques that we can apply to our day-to-day programming. We have a Bike Shed episode where we talked about is discrete math relevant to day-to-day programmers and what are the ways it's so? We'll link that in the show notes. I also gave a talk at RailsConf last year diving into that titled: The Math Every Programmer Needs. So, if you're looking for something that's accessible to someone who's not done a math degree, those are two great jumping-off points.
STEPHANIE: Yeah. And then, maybe you'll start drawing out arrays and applying combinatorics to figure out your performance problems.
JOËL: On that note, shall we wrap up?
STEPHANIE: Let's wrap up. Show notes for this episode can be found at bikeshed.fm.
JOËL: This show has been produced and edited by Mandy Moore.
STEPHANIE: If you enjoyed listening, one really easy way to support the show is to leave us a quick rating or even a review in iTunes. It really helps other folks find the show.
JOËL: If you have any feedback for this or any of our other episodes, you can reach us @_bikeshed, or you can reach me @joelquen on Twitter.
STEPHANIE: Or reach both of us at hosts@bikeshed.fm via email.
JOËL: Thanks so much for listening to The Bike Shed, and we'll see you next week.
ALL: Byeeeeeeeeee!!!!!!!!
AD:
Did you know thoughtbot has a referral program? If you introduce us to someone looking for a design or development partner, we will compensate you if they decide to work with us.
More info on our website at: tbot.io/referral. Or you can email us at: referrals@thoughtbot.com with any questions.
439 episodes
All episodes
×Welcome to Player FM!
Player FM is scanning the web for high-quality podcasts for you to enjoy right now. It's the best podcast app and works on Android, iPhone, and the web. Signup to sync subscriptions across devices.