|
Swift generics are cool.
|
#
?
Sep 29, 2018 23:05
|
|
|
|
| # ? May 17, 2024 15:15 |
|
|
eth0.n posted:Rust has great generics. As easy to use as Java's, but get compiled and optimized like C++ templates. They're modeled after Haskell, but not as flexible. It can't be a good generics system without HKT. 
|
|
#
?
Sep 30, 2018 00:13
|
|
|
Plorkyeran posted:Seems pretty obvious what the error is? It really isn't. In fact, I still don't know what the problem with the original example is. 
|
|
#
?
Sep 30, 2018 04:05
|
|
|
Presto posted:It really isn't. In fact, I still don't know what the problem with the original example is. It was a vector of vectors of ints, and it was trying to find a particular int in it; it failed because int is not comparable to a vector of ints.
|
|
#
?
Sep 30, 2018 04:06
|
|
|
Plorkyeran posted:This is a 7-year-old compiler. Let's take a look at the error with something a bit more recent:
|
|
#
?
Sep 30, 2018 04:16
|
|
|
SupSuper posted:I'm glad in 7 years the actual source line is still just a footnote buried under template errors, how far we've come. Mods change my name to "template error footnote"
|
|
#
?
Sep 30, 2018 04:17
|
|
|
I mean, it's the same thing you do when diagnosing an exception in library code - ignore the stuff at the top, and look down the stack trace until you see a filename that's part of your code. What it really needs is a way to output errors that's more than just "a massive dump of plain text", so that the compiler can immediately show you the failing line and then let you drill down into the details if you need them in order to figure out what's up.
|
|
#
?
Sep 30, 2018 04:48
|
|
|
Bruegels Fuckbooks posted:i like generics in C# the best. C#/java create generic objects on runtime and deal with all this poo poo using the best language feature in the history of language features, reflection. No they don't. C# generics are real generics and the runtime is aware of them. The IL JIT/AOT compiler does in fact generate generic instantiations for value types, though as an optimization all reference types share an implementation. The runtime reflection model is also aware of generics. You can dynamically create instances of generic types and inspect the types of generic parameters at runtime. Java generics are garbage because the runtime isn't aware of them at all. The compiler throws it all away and everything is Object at runtime, which is also why Java generics don't support value types (or more accurately primitive types since Java doesn't have user-defined value types).
|
|
#
?
Sep 30, 2018 04:54
|
|
|
Presto posted:It really isn't. In fact, I still don't know what the problem with the original example is. The actual error message is error: invalid operands to binary expression ('std::vector<int, std::allocator<int> >' and 'const int'). In then prints the stack trace of the error, and the first line that isn't inside library code is the call to std::find(). This tells you that the things you passed to std::find are not comparable.
|
|
#
?
Sep 30, 2018 06:23
|
|
|
Simulated posted:No they don't. C# generics are real generics and the runtime is aware of them. The IL JIT/AOT compiler does in fact generate generic instantiations for value types, though as an optimization all reference types share an implementation. The runtime reflection model is also aware of generics. You can dynamically create instances of generic types and inspect the types of generic parameters at runtime. ok yeah, I was loose about how this stuff works in java because frankly i don't care about java and not surprisingly it did something asinine. let me try to explain how i think C++/C# work. say you're using templates in C++. whenever you use a type in a template, the code has to be generated for that specific use. so there will be a copy of the code in the binary for type1, type 2, type 3 that is used in a template (of course there's a feature called template folding that will identify templates with common implementation and use that same implementation from every template) in c#, there are generic "types", and the substitution of these types is done at JIT - but the generic type actually exists. you may have to provide constraints on what you an pass in as a type. the effect is that in C#, the generics aren't rocket science to use. Bruegels Fuckbooks fucked around with this message at 14:21 on Sep 30, 2018 |
|
#
?
Sep 30, 2018 14:14
|
|
|
Simulated posted:Java generics are garbage because the runtime isn't aware of them at all. The compiler throws it all away and everything is Object at runtime, which is also why Java generics don't support value types (or more accurately primitive types since Java doesn't have user-defined value types). Bruegels Fuckbooks posted:ok yeah, I was loose about how this stuff works in java because frankly i don't care about java and not surprisingly it did something asinine. Generics were retrofitted into Java 8 years after the original release, so it was significantly easier to make them a compiler-only feature.
|
|
#
?
Sep 30, 2018 14:31
|
|
|
Plorkyeran posted:The actual error message is error: invalid operands to binary expression ('std::vector<int, std::allocator<int> >' and 'const int'). In then prints the stack trace of the error, and the first line that isn't inside library code is the call to std::find(). This tells you that the things you passed to std::find are not comparable. I love how even once you vector in on the most useful line of the error spew, you still have to learn to ignore the "std::allocator" junk to read it.
|
|
#
?
Sep 30, 2018 16:23
|
|
|
Simulated posted:Java generics are garbage because the runtime isn't aware of them at all. The compiler throws it all away and everything is Object at runtime, which is also why Java generics don't support value types (or more accurately primitive types since Java doesn't have user-defined value types). Of course, Scala supports value types no problem. If you start from scratch you can do bytecode basically anything the java compiler can't because of backwards compatibility. You'll just need to read up on covariant and contravariant types and upper and lower type bounds and all that poo poo because Scala generics are so much more powerful than Java ones, but that also means the more powerful features are more complicated to use.
|
|
#
?
Sep 30, 2018 16:32
|
|
|
eth0.n posted:Rust has great generics. As easy to use as Java's, but get compiled and optimized like C++ templates. They're modeled after Haskell, but not as flexible. Eela6 posted:Rust. I'd say Rust has very good generics, but atm they fall short of "great" because there aren't type-level numerics yet. This may not seem like a big deal, but it causes some issues with tuples and arrays, as well as makes any attempt to implement matrices and vectors really clunky to work with. This mainly affects math code, but I've seen it bleed into other arenas as well on occasion. There's also a need for some subset of HKTs, especially wrt lifetimes, because you can't express a few things that would be very helpful. For instance, you can't create create streaming iterators where the borrow duration is tied to the loop rather than the entire iterator. Obviously all generics systems have limitations, even the really great ones, but it's pretty easy to hit a wall in Rust. The first time I hit one was when trying to unroll a recursive tree search into a loop and had to go full-unsafe to do it. Trees and graphs in Rust are a pain in general, but it was really the limitations on lifetimes in iterators that got me here. Linear Zoetrope fucked around with this message at 17:15 on Sep 30, 2018 |
|
#
?
Sep 30, 2018 17:08
|
|
|
Generics in C# seem so limited that in a lot of cases I just resort to downcasting liberally 
|
|
#
?
Sep 30, 2018 17:47
|
|
Linear Zoetrope posted:I'd say Rust has very good generics, but atm they fall short of "great" because there aren't type-level numerics yet. This may not seem like a big deal, but it causes some issues with tuples and arrays, as well as makes any attempt to implement matrices and vectors really clunky to work with. This mainly affects math code, but I've seen it bleed into other arenas as well on occasion. There's also a need for some subset of HKTs, especially wrt lifetimes, because you can't express a few things that would be very helpful. For instance, you can't create create streaming iterators where the borrow duration is tied to the loop rather than the entire iterator. I just want to say that I appreciate this post and agree. Rust is not perfect.
|
|
|
#
?
Sep 30, 2018 18:17
|
|
|
How do I make microservices and microrepos not a coding horror? My current gig has broken it down to nearly one repo for one file (one project has 85 lines of code and 200 lines of packaging and boilerplate  ). There’s no integration tests to speak of, People make changes to libraries, and then some schmuck gets to work out what broke what when they go work on something that depends on it. As far as I can tell, people mostly work on one repo at a time. But I’m working on three or four at a time as I’ve been tasked with expanding their coding-horror-of-it’s-own of an api gateway (one of three?). I don’t really have the clout to roll in and change everything. But there’s got to be something I can suggest to make it less of a clusterfuck of making GBS threads code into a fan and hoping you miss the blades, and certainly something more worthy of developer time than their current plan to rename all the repos and code  .I am building a suite of BDD style integration tests for the API gateway, but I’d wager they’ll never get run once I leave, let alone expanded to test other components.
|
|
#
?
Sep 30, 2018 18:29
|
|
|
Personally I'd make a repo called ci that contains scripts you can run generically from any of the repos to fetch the latest release version of every service (except the current repo, which is modified obv) and then run tests. Then you can add a small CI file (er... I hope CI is available to you), where on push it pulls the CI repository and runs the suite. This ensures the integration tests for the entire drat thing are sorted out whenever a change is made. Obviously breaking changes can happen, but that's easier to handle intentionally and should be the rare case. Linear Zoetrope fucked around with this message at 18:38 on Sep 30, 2018 |
|
#
?
Sep 30, 2018 18:34
|
|
|
Are the powers that be open to process improvements and at least a little bit sane? If not, save your energy. If so, you need to have some concept of a released version of a library. Instead of building against source, you build against the copy in some location that has versions of libraries that have been through a release process. And then you put all of the tests you want into the release process.
|
|
#
?
Sep 30, 2018 18:36
|
|
|
Libraries and services are versioned, so there’s no rolling breakages. It’s not a total shitshow. But running tests against the latest, and not the pinned version makes sense. If not on each build, at least nightly. This would point out breaking changes in any dependencies. ultrafilter posted:Are the powers that be open to process improvements and at least a little bit sane? If not, save your energy.
|
|
#
?
Sep 30, 2018 18:58
|
|
|
Horse Clocks posted:Libraries and services are versioned, so there’s no rolling breakages. It’s not a total shitshow. If all else fails, you can go the Netflix way. Netflix has a test server running the latest versions of their microservices. It has a special setup, where it only runs those services deemed essential (whitelisted) and blocks connections to anything else. Every couple minutes, a test script on that server tries to login to the local Netflix instance and start a movie. That's it. If, at any point it fails, it's most likely that someone added functionality that makes something that isn't whitelisted essential (or someone just broke a thing). When this happens, the ops team responsible for the test server finds the dev who made the change, and gives them a simple choice: Either make your service non-essential, for instance by building in a graceful fallback... or prepare your team for 24/7 devops customer support because any downtime would cost too much. If you do the latter, we'll add your thing to the test server as a whitelisted essential service. For some reason, in over 90% of cases, devs manage to find a way to make their thing non-essential.
|
|
#
?
Sep 30, 2018 19:22
|
|
|
Plorkyeran posted:Seems pretty obvious what the error is? Stockholm syndrome strikes again 
|
|
#
?
Sep 30, 2018 19:31
|
|
|
Haskell's generic programming features are basically perfect. My only complaint is the lack of some sort of variadic type parameter support for tuples. Haskell libraries always seem to end up with a series of functions/types called foo2, foo3, foo4, one for each tuple size. Look at this nonsense: https://hackage.haskell.org/package/lens-4.17/docs/Control-Lens-Tuple.html.
|
|
#
?
Sep 30, 2018 21:11
|
|
|
Is there a term for building a lovely replacement for an existing platform on top of that platform? For example, building your own web portal in front of a service that already provides a fully functional web portal? I can't match it up to any of the common anti-patterns like the inner platform effect. I guess it's a form of "not invented here".
|
|
#
?
Sep 30, 2018 21:19
|
|
|
Pretty sure it's just called "hiring a project manager"
|
|
#
?
Sep 30, 2018 21:22
|
|
|
New Yorp New Yorp posted:Is there a term for building a lovely replacement for an existing platform on top of that platform? For example, building your own web portal in front of a service that already provides a fully functional web portal? I've used "buying a car to crush into a wheel-shaped object" before. Oh god, rolling their own AOP.
|
|
#
?
Sep 30, 2018 21:22
|
|
|
LOOK I AM A TURTLE posted:Haskell's generic programming features are basically perfect. My only complaint is the lack of some sort of variadic type parameter support for tuples. Haskell libraries always seem to end up with a series of functions/types called foo2, foo3, foo4, one for each tuple size. Look at this nonsense: https://hackage.haskell.org/package/lens-4.17/docs/Control-Lens-Tuple.html. If you have nineteen elements in a tuple you should probably be using a record type. Maybe two.
|
|
#
?
Sep 30, 2018 21:34
|
|
|
New Yorp New Yorp posted:Is there a term for building a lovely replacement for an existing platform on top of that platform? For example, building your own web portal in front of a service that already provides a fully functional web portal? I think that's called "software development." For example, every REST or GraphQL web app right now would have been faster and more secure as a database that properly uses authentication, authorization, triggers, relations and constraints. But once a problem is solved, you need to start coming up with more problems. So let's just use the database as a key-value store and rewrite all the rest in slow-as-hell, insecure-as-poo poo Rails. xtal fucked around with this message at 01:20 on Oct 1, 2018 |
|
#
?
Sep 30, 2018 23:40
|
|
|
xtal posted:I think that's called "software development." For example, every REST or GraphQL web app right now would have been faster and more secure as a database that properly uses authentication, authorization, triggers, relations and constraints. But once a problem is solved, you need to start coming up with more problems. So let's just use the database as a key-value store and rewrite all the rest in slow-as-hell, insecure-as-poo poo Rails. I'm thinking more along the lines of something like the Azure portal. Everything you can do with the portal, you can do with REST APIs. I keep seeing people rebuilding their own Azure portal that calls the same REST APIs as the existing portal. I can't fathom why, other than "not invented here" and I was looking for a term to use to describe it when I'm telling people "that's dumb, don't do it".
|
|
#
?
Oct 1, 2018 01:16
|
|
|
New Yorp New Yorp posted:I'm thinking more along the lines of something like the Azure portal. Everything you can do with the portal, you can do with REST APIs. I keep seeing people rebuilding their own Azure portal that calls the same REST APIs as the existing portal. I can't fathom why, other than "not invented here" and I was looking for a term to use to describe it when I'm telling people "that's dumb, don't do it". Sounds precisely like NIH in that case!
|
|
#
?
Oct 1, 2018 01:20
|
|
|
Absurd Alhazred posted:It was a vector of vectors of ints, and it was trying to find a particular int in it; it failed because int is not comparable to a vector of ints. Ok, see, now that's an error message I can get behind.
|
|
#
?
Oct 1, 2018 03:02
|
|
|
Plorkyeran posted:The actual error message is error: invalid operands to binary expression ('std::vector<int, std::allocator<int> >' and 'const int'). In then prints the stack trace of the error, and the first line that isn't inside library code is the call to std::find(). This tells you that the things you passed to std::find are not comparable.
|
|
#
?
Oct 1, 2018 03:06
|
|
|
New Yorp New Yorp posted:I'm thinking more along the lines of something like the Azure portal. Everything you can do with the portal, you can do with REST APIs. I keep seeing people rebuilding their own Azure portal that calls the same REST APIs as the existing portal. I can't fathom why, other than "not invented here" and I was looking for a term to use to describe it when I'm telling people "that's dumb, don't do it". the second dipshit azure portal was entirely due to microsoft infighting. the last company i was at scored an ex-microsoftie manager from that clusterfuck and then, given impetus by a second azure portal, other fools started makin third party ones
|
|
#
?
Oct 1, 2018 03:06
|
|
|
Presto posted:All right, but why doesn't it just say "The things you passed to std::find are not compatible"? That is what it's saying.
|
|
#
?
Oct 1, 2018 03:06
|
|
|
ultrafilter posted:That is what it's saying. Yes, you just need to scroll all the way to the right. quote:invalid operands to binary expression ('std::vector<int, std::allocator<int> >' and 'const int')
|
|
#
?
Oct 1, 2018 03:18
|
|
|
Presto posted:All right, but why doesn't it just say "The things you passed to std::find are not compatible"? C++ currently doesn't have a way to express "the two types passed to this function have to be comparable" (that's maybe finally coming in C++20 after over a decade of delays). As a result, the compiler doesn't know what point in the call stack is incorrect, and just prints the whole thing. As with any time you get an error stack starting in library code, you usually want to start with looking at the first one not in the library. xtal posted:Yes, you just need to scroll all the way to the right. That's mostly just a result of compiler explorer having really long paths.
|
|
#
?
Oct 1, 2018 03:45
|
|
|
bob dobbs is dead posted:the second dipshit azure portal was entirely due to microsoft infighting. the last company i was at scored an ex-microsoftie manager from that clusterfuck Wait what? Where's the second azure portal?
|
|
#
?
Oct 1, 2018 03:58
|
|
|
brap posted:Wait what? Where's the second azure portal? i think they successfully migrated off but portal.azure.com and manage.windowsazure.com had two separate existing portals. they didn't have proper iam in the manage.windowsazure.com, which was the old one, and it was also poo poo in lots of other ways but the deprecation was a big fight https://stackoverflow.com/questions/32902089/manage-windowsazure-com-vs-portal-azure-com
|
|
#
?
Oct 1, 2018 04:15
|
|
|
xtal posted:I think that's called "software development." For example, every REST or GraphQL web app right now would have been faster and more secure as a database that properly uses authentication, authorization, triggers, relations and constraints. But once a problem is solved, you need to start coming up with more problems. So let's just use the database as a key-value store and rewrite all the rest in slow-as-hell, insecure-as-poo poo Rails. I don't understand what you're saying here. Are you saying REST and GraphQL do not use a database to store their data?
|
|
#
?
Oct 1, 2018 04:19
|
|
|
|
| # ? May 17, 2024 15:15 |
|
|
Thermopyle posted:I don't understand what you're saying here. I think he is saying that they are useless layers on top of a database. Which is, uh, an interesting argument if I'm reading it right.
|
|
#
?
Oct 1, 2018 06:12
|
|
