This blog has been quiet recently because, for the next month, I’m focused on my imminent 1,000 mile cycle ride from Land’s End to John O’Groats. I started a separate blog to track my training and progress if you’re interested!
Categories
Categories
Static checking java
As a distraction from learning about denotational semantics, I took a look at ESCJava2, which is an extended static checker for Java. ESCJava2 lets you add annotations to your source code which describe invariants and pre/post-conditions which you expect to be true. Then it goes away and tries to verify that they are true. Often, you have to go back and add extra annotations elsewhere (anyone who has const-ified a C++ program knows this feeling). But eventually you get to a point where you can be fairly confident that your program is (statically) free from all sorts of badness – eg. you’ll never try to pop from an empty stack, you’ll never get a null pointer exception.
This is all done declaratively, which makes it much tastier than unit tests.
Does it work? Well, I got out the source code for my trusty java raytracer, which was literally the first java program I ever wrote .. back in 1996 or something. At about 1,000 lines of code, it’s both “real world” enough to act as a decent test case, yet simple enough to grok quickly.
Initially, I had to spend a while adding ‘non_null’ annotations everywhere – every field in every object had that property. Dull, but useful to have it all checked.
Next was various bound-checking stuff. I had an image class with width/height members and a 2d array of pixels. I ‘knew’ that the array was of the right size, but by asserting some invariants to that effect, I could get ESCJava to statically bounds-check all the code that reads/writes pixels. The invariant were moderately complex things like “forall i < height; pixels[i].length == width". These "forall" quantifiers are what makes this approach tastier than unit tests. With unit tests, you have to pick "representative" data points and then trust that you chose well. In the haskell world, QuickCheck is a big step above that. Perhaps ESCJava brings some hint of universal tastiness to Java? Well, except that it didn't work in practise. My key 'raytrace all the pixels' method couldn't be processed by ESCJava. It gave an error which effectively said "the theorem which I'd have to prove to be sure that the code is safe is just too complex". Disappointing, considering that it was little more than some nested loops plus a bit of logic. ESCJava had a few other nice features though. Since my code is old, it the plain old Vector class. ESC can add a 'phantom' elementType field to your vector, which gives you 1.5-like static type checking on pre-1.5 collections. But in the end, the final nail in the coffin was that ESCJava does not support java 1.5+ features. So it doesn't understand generics. Oh well, no use for the real code that I work with. I'd love to have the time to understand this area more deeply. I like this kind of "unsound and incomplete but useful in practise" part of the statically-checked spectrum.
Categories
HAppS-state mistake
I’m grappling with HAppS-State at the moment, and thought it useful to capture some work-in-progress notes. My toy webapp allows you to view and edit information about people, places and things. The webapp state just consists of several identifier->entity maps.
HAppS-state requires that you write your state query/update functions as normal MonadState or MonadReader computations. But you also must process each of these functions using the mkMethods template haskell function. This generates some “behind the scenes” machinery to turn your vanilla state-updating monads into something which additionally maintains a write-ahead disk log to make the change durable. If your update function was called “modifyPersonName”, the call to mkMethod generates a datatype/constructor called ModifyPersonName which, when used like “update ModifyPersonName newName” has the richer durable behaviour.
I have lots of different entities, and they all have lots of different attributes. It quickly gets boring writing seperate “modifyEntityX” functions for each attribute. Haskell’s rather lousy record syntax doesn’t help out much either.
Fortunately, there’s a nice library called data-accessor which provides a more pleasant way to handle haskell record types. The idea is that you build up a getter/setter pair for each record member. These are first class values, and are consequently much more flexible than the builtin haskell record update syntax.
This seemed to be the answer to my problem – I can make a generic “modifyPersonAttribute” function which takes one of these accessors as an argument in order to select the field to update.
Unfortunately, this doesn’t work. I get a type error effectively stating that happs-state requires that all of the arguments to update/query function must themselves be Serializable.
This confused me. I can see that the application state type (and all of its constituent subparts) need to be serializable. But I was surprised that all the arguments for state-updating functions needed to be Serializable.
Then I realized what my false assumption was. I had assumed that happs was persisting the result of running the update operation to the logfile, similar to what mysql does for redo logs. In other words, I thought the logfile consisted of things like “the new value for row 42 is ‘foo'”.
However, a quick look at the contents of the _local directory (where happs stores its state) shows that this isn’t the case. Happs stores a description of the computation itself – ie. the name of the update operation and the (serialized) arguments it took.
This has got me somewhat stuck. Firstly, my generic ‘modifyPersonAttribute’ doesn’t work because the “accessor” values are not serializable. I’m now wondering if perhaps I can bypass data-accessors and instead write some template haskell to generate the happs machinery for all my entity types and all their attribute values.
But more importantly, this means that you need to be super-careful not to change the behavior of your state-modifying functions if there are any uncheckpointed changes in the logfile. Let’s say you have a createPerson function which takes a name and stores the name straight into the application state. But some days later, you decide that you want to make names have an initial capital letter before storing them. You change the code and restart the application – but unless you were careful to checkpoint the application state, the log will be replayed and you’ll end up with a different application state from before (some existing people will have the initial-caps logic applied to their name, not just new people).
Categories
Hacking with HAppS – what type, handler?
(These posts are prettified versions of my notes that I made whilst stumbling around HAppS for the first time. I hope they’ll be of use to future HAppS explorers!)
Note 1 .. In which we figure out what ServerPartT and WebT really are.
When we start HAppS running with ..
simpleHTTP config list_of_handlers :: IO ()
… what type do the handlers have? Let’s explore some possible universes!
In a simple world, a handler could have type
Request -> Maybe Response
This allows handlers to be selective about which Requests they handle, which is useful. But it also means that handlers must be pure functions so there’d be no way of maintaining any ‘application state’ between requests. Not very useful!
Let’s try to remedy that. If the handlers had type:
Request -> State S Response
.. then we’d be able to retain and modify some application state (of type S) between handler invocations. But that’s all we’d be able to do. We still couldn’t write logs to disk, read file contents or talk to databases.
To allow handlers to do that, handlers would need to have type ..
Request -> IO Response
(As an aside, note that whilst simpleHTTP has type “IO ()” it’s up to simpleHTTP whether or not it exposes the full power of the IO monad to the handlers. However, as we’ve just seen, not doing so would pretty much cripple the handlers)
So is this really what HAppS gives you? Let’s see. In the real world HAppS handlers have type “ServerPartT IO a” which is defined to be …
newtype ServerPartT m a = ServerPartT { unServerPartT :: Request -> WebT m a }
In other words, if you have a value of type “Request -> WebT m a”, you can tag it with the ServerPartT constructor to say “this ain’t just any function of that type, it’s part of a web server, dogammit!”.
But, apart from the tag, our handlers really just have type “Request -> WebT m a”. So what’s a WebT?
newtype WebT m a = WebT { unWebT :: m (Result a) }
So, ignoring the tag, it looks kinda like our handlers can produce any monadic computation that, when run, yields a “Result a” (the “Result a” type lets us say “I handled this request, here’s my answer” or “I didn’t handle this request”)
Hmm, there’s something not right there. If that was the case, it would be possible to use, for example, both “Result -> IO (Result a)” and “Result -> State S (Result a)” as handlers. But how would simpleHTTP ‘run’ such monadic computations? To run a State computation you use “evalState”. To run an IO computation, you have to return it from main. There’s no single way to run all possible monads.
Did I miss something? Yes, sort of – I just blindly assumed ‘m’ was a monad because of the way it was used. But there’s no “Monad m =>” constraints anywhere. And, in fact, if we go back to the start ..
simpleHTTP :: ToMessage a => Conf -> [ServerPartT IO a] -> IO ()
.. we can see that, right from the top, simpleHTTP only deals with ServerPartT’s and WebT’s where the ‘m’ type parameter is IO.
So, after all that, the handlers in HAppS handlers can be anything of type: Request -> IO (Response a). The ‘a’ type needs to be something that HApps can figure out a content-type and be able to make an HTTP message from (the ToMessage type class). The “IO (Response a)” part gets dressed up as a “WebT” (think: computation that produces the response). And the whole thing things gets dressed up as a “ServerPartT” (think: request handler).
What does each bit do? The IO part gives you free reign to do side-effecting computations, which get sequentially executed by the runtime. The Response part allows you to say ‘I handled this’ or ‘didn’t handle it’ and the ‘a’ part can be your XML or HTML data type (so long as its an instance of ToMessage).
So that’s it.
Man, that was a lot of complexity for such a simple end result. I don’t understand why there’s so much generality in the types of ServerT and WebT – I’d love to know!
Categories
Hacking with HAppS
I’m always on the lookout for more fun ways of programming – I hate doing dull boilerplate work. So I’ve been trying out HAppS, a webserver framework for Haskell. It’s interestingly different from most web frameworks. I want to capture my early observations before I forget them. To give some context, my “toy app” I’m building is an “IMDB for Computer Scientists” type website which gives structured information about people (eg. Guy Steele), what papers they’ve published (eg. Growing a Language), what events they’ve participated in (eg. a talk about “Growing a Language” at OOPSLA, available via Google video).
So the biggest thing about HAppS it that it doesn’t have to use a database to persist state. It takes the Prevayler approach, which means all your data lives in RAM as normal in-memory data structures. When the data is about to be modified, a delta is first persisted to a fast log on disk (giving durability) and then the in-memory version is updated. Periodically, checkpoints are taken.
This is a huge win because it means you avoid the whole tedious ‘impedance mismatch’ involved with persisting to a database – O/R mappings, mismatches between DB types and language types. It means that you can create new datatypes easily, and get on with the real business of adding functionality rather than futzing with databases.
Of course, there are downsides to this approach which I will discuss next. But first, the big upside: I can develop features quicker. Life being what it is, I’m almost certainly making lots of mistakes whilst building my app. Without a DB layer to burn time on, I can find out that I’m (inevitably) building the Wrong Thing sooner and make a course correction.
Downside #1: you normally rely on a database to get fast indexed lookups. With HAppS, you do this by constructing an in-memory hash table to act as an index. Actually, that’d be quite boring to do by hand so HApps uses metaprogramming (template haskell) to build the index data structures for you. They are moderately smart, with lazy update strategies.
Downside #2: scalability and availability. A big benefit of the usual “stateless web frontend, stateful database” split is that you can trivially scale the web fleet (no state) and fairly easily scale the database (replication/partitioning). With the HAppS approach it sounds like you can only have a single webserver/state-store machine – which means no horizontal scalability and bad availability. Actually, that’s not the whole story – the HAppS team are working on a system whereby the state gets shared across multiple machines (permitting scalable reads) and sharded (each machine is the master for some subset of the state).
That should be a killer downside, right? The argument is looks like this: “all successful webapps are popular, and popular apps must scale”. However, I think that to become a successful webapp, you have to firstly be a webapp. My “projects” directory is littered with half-finished projects where I got bored before I finished it. Quite frankly, if using HAppS means that I can avoid a lot of tedious work and actually finish a project then I’ll happily embrace the ghost of future scaling pain. Also, not every successful website gets as much traffic as Flickr/Twitter – my toy app is an “IMDB for Computer Scientists” and there just aren’t that many computer scientists in the world. Not every website needs to be super-scalable from day one. I’m not being naive here – my dayjob is scalability central – but it’s a question of finding the most appropriate hammer for the particular nail you’re trying to hit.
Having said all that, let me qualify it a bit: I would ideally like to create a perfectly scalable webapp from day one if there was no overhead in doing so. Most of the industry is focused on scalability-via-database, and attempt to minimize tedious work in various ways (eg. metaprogramming and reflection in ActiveRecord). However, these invariably ends up being painfully ‘leaky’ abstractions. I’ve seen this many times, and it’s given me the motivation to look for other approaches which can avoid the persistence pain. So perhaps HAppS is the answer, especially if my question is “can I write a reasonable webapp quickly” rather than “can I write a superscalable website slowly?”.
Disadvantage #3: Data migrations. This is really a red herring. I don’t really believe that data migrations are an advantage of having your data in a database. If you are using an O/R mapping, you surely want to migrate old objects to new object, not old relations to new relations. The approach which HAppS takes is to version your datatypes (mostly done via metaprogramming) and you write haskell code to migrate from old versions.
Disadvantage #4: Language dependence. If you use a database, you can access your data from any language. You can produce reports, do adhoc SQL queries and such like. These are indeed useful; I do this a lot. However, language-independence comes at a cost (impedance mismatch between database/language data types, business rules usually not applied at the database layer). Often, language independent access can be better provided by service layers which wrap the database layer, providing insulation from the schema changes and allowing fine grained access control and throttling. I’ve never been able to query Flickr’s database directly, but I’ve accessed their data via their API many times. HAppS provides more metaprogramming support for exposing data types via XML.
So that’s why I’m investing the time into learning about HAppS. It’s good to challenge your assumptions about how to do things. In the next post, I’ll write a bit about how HAppS works under the hood.