C#, C# 4, Google, Speaking engagements

OS Jam at Google London: C# 4 and the DLR

18 Comments

Last night I presented for the first time at the Google Open Source Jam at our offices in London. The room was packed, but only a very few attendees were C# developers. I know that C# isn’t the most popular language on the Open Source scene, but I was still surprised there weren’t more people using C# for their jobs and hacking on Ruby/Python/etc at night.

All the talks at OSJam are just 5 minutes long, with 2 minutes for questions. I’m really not used to this format, and felt extremely rushed… however, it was still a lot of fun. I used a somewhat different approach to my slides than the normal “bullet points in PowerPoint” – and as it was only short, I thought I might as well effectively repeat the presentation here in digital form. (Apologies if the images are an inconvenient size for you. I tried a few different ones, and this seemed about right. Comments welcome, as I may do a similar thing in the future.)

First slide

Introductory slide. Colleagues forced me to include the askjonskeet.com link.

Second slide

.NET isn’t Open Source. You can debug through a lot of the source code for the framework if you agree to a “reference licence”, but it’s not quite the same thing.
Third slide

.NET isn’t Open Source, but the DLR is. And IronRuby. And IronPython. Yay!

And of course Mono is Open Source: the DLR and Mono play nicely together, and the Mono team is hoping to implement the new C# 4.0 features for the 2.8 release in roughly the same timeframe as Microsoft.
Fourth slide

This is what .NET 4.0 will look like. The DLR will be included in it, despite being open source. IronRuby and IronPython aren’t included, but depend heavily on the DLR. (Currently available versions allow you to use a “standalone” DLR or the one in .NET 4.0b1.)

C# doesn’t really depend on the DLR except for its handling of dynamic. C# is a statically typed language, but C# 4.0 has a new static type called dynamic which you can do just about anything with. (This got a laugh, despite being a simple and mostly accurate summary of the dynamic typing support in C# 4.0.)
Fifth slide

The fundamental point of the DLR is to handle call sites – decide what to do dynamically with little bits of code. Oh, and do it quickly. That’s what the caches are for. They’re really clever – particularly the L0 cache which compiles rules (about the context in which a particular decision is valid) into IL via dynamic methods. Awesome stuff.

I’m sure the DLR does many other snazzy things, but this feels like it’s the core part of it.

Sixth slide

At execution time, the relevant binder is used to work out what a call site should actually do. Unless, that is, the call has a target which implements the shadowy IDynamicMetaObjectProvider interface (winner of “biggest mouthful of a type name” prize, 2009) – in which case, the object is asked to handle the call. Who knows what it will do?
Seventh slide

Beautifully syntax-highlighted C# 4.0 source code showing the dynamic type in action. The method calls on lines 2 and 3 are both dynamic, even though in the latter case it’s just using a static method. Which overload will it pick? It all depends on the type of the actual value at execution time.

If I’d had more time, I’d have demonstrated how the C# compiler preserves the static type information it knows at compile time for the execution time binder to use. This is very cool, but would take far too long to demonstrate in this talk – especially to a bunch of non-C# developers.
Eighth slide There were a couple of questions, but I can’t remember them offhand. Someone asked me afterwards about how all this worked on non-.NET implementations (i.e. Mono, basically). I gather the DLR itself works, but I don’t know whether C# code compiled in the MS compiler will work at the moment – it embeds references to binder types in Microsoft.CSharp.dll, and I don’t know what the story is about that being supported on Mono.

This is definitely the format I want to use for future presentations. It’s fun to write, fun to present, and I’m sure the “non-professionalism” of it makes it a lot more interesting to watch. Although it’s slower to create text-like slides (such as the first and the last one) this way, the fact that I don’t need to find clip-art or draw boxes with painful user interfaces is a definite win – especially as I’m going to try to be much more image-biased from now on. (I don’t want people reading slides while I’m talking – they should be listening, otherwise it’s just pointless.)

Books, C#, C# 4

Dynamic type inference and surprising possibilities

9 Comments

There have been mutterings about the fact that I haven’t been blogging much recently. I’ve been getting down to serious work on the second edition of C# in Depth, and it’s taking a lot of my time. However, I thought I’d share a ghastly little example I’ve just come up with.

I’ve been having an email discussion with Sam Ng, Chris Burrows and Eric Lippert about how dynamic typing works. Sam mentioned that even for dynamically bound calls, type inference can fail at compile time. This can only happen for type parameters where none of the dynamic values contribute to the type inference. For example, this fails to compile:

static void Execute<T>(T item, int value) where T : struct {}
…
dynamic guid = Guid.NewGuid();
Execute("test", guid);

Whatever the value of guid is at execution time, this can’t possibly manage to infer a valid type argument for T. The only type which can be inferred is string, and that’s not a value type. Fair enough… but what about this one?

static void Execute<T>(T first, T second, T third) where T : struct {}
…
dynamic guid = Guid.NewGuid();
Execute(10, 0, guid);
Execute(10, false, guid);
Execute("hello", "hello", guid);

I expected the first call to compile (but fail at execution time) and the second and third calls to fail at compile time. After all, T couldn’t be both an int and a bool could it? And then I remembered implicit typing… what if the vaue of guid isn’t actually a Guid, but some struct which has an implicit conversion from int, bool and string? In other words, what if the full code actually looked like this:

using System;

public struct Foo
{
    public static implicit operator Foo(int x)
    {
        return new Foo();
    }

    public static implicit operator Foo(bool x)
    {
        return new Foo();
    }

    public static implicit operator Foo(string x)
    {
        return new Foo();
    }
}

class Test
{
    static void Execute<T>(T first, T second, T third) where T : struct {}

    static void Main()
    {
        dynamic foo = new Foo();
        Execute(10, 0, foo);
        Execute(10, false, foo);
        Execute("hello", "hello", foo);
    }
}

Then T=Foo is a perfectly valid inference. So yes, it all compiles – and the C# binders even get it all right at execution time. So much for any intuition I might have about dynamic typing and inference…

No doubt I’ll have similar posts about new C# 4 features occasionally… but they’re more likely to be explanations of misunderstandings than deep insights into a correct view of the language. Those end up in the book instead :)

Stack Overflow

Reasons for voting on questions and answers

23 Comments

I’ve recently been involved in a few discussions around voting on Stack Overflow, and I think my own "policy" around it may be different to that of others. I thought it would be worth sharing why I personally vote items up or down, and hear your thoughts too. This blog may not be the ideal venue for such a post, but until such time as we have a real "meta" site for Stack Overflow (such as Stack Overflow Overflow) I can’t think of anywhere better to write about it. Readers who are only interested in coding should move on; I promise not to include anything about code in the rest of this post.

I’m going to assume that anyone who’s read this far is at least somewhat familiar with the logistics of Stack Overflow – in particular, how one votes and the effects on reputation. I’ll use the word "post" here to mean either a question or an answer.

I’d like to stress that this is in no way meant to be seen as an "official" voting guide – just how I happen to think.

Why vote?

There are two "audiences" for a vote in my view: the author of the post and the community who is reading the post and looking at the vote tally. The author can tell what votes they’ve received using the reputation tab of their "recent activity" page, whereas the readers can only tell what the overall tally is. Obviously the author also receives or loses reputation, too. This means the effect on the author and the audience are slightly different.

For the author, the immediate reward or punishment aspect may sound like the most important aspect: but I’d argue that for many users (particularly those with high reputation) the reputation for a single vote isn’t as important as the effect on the vote tally and what it communicates about your post. There’s usually a positive feedback effect on Stack Overflow: if one answer has a couple of votes and another has none, then the higher voted one is likely to get read more and thus garner more votes. The opposite can happen: an answer with a negative score will sometimes receive "sympathy" votes from users who think, "This answer isn’t brilliant, but it’s not bad enough to deserve downvotes."

For me, the important point about a downvote is that it may indicate I’ve got something wrong in an answer. I may have missed the point of the question in the first place, or simply provided a technically incorrect or unhelpful answer. I want to know about that, so I can fix my answer or delete it if I can’t actually provide any more useful information than is contained in other answers.

For the reader, the information communicated by the score can be as simple as "this question is interesting/this answer is helpful" vs "this is a poor question/this answer is harmful". I would hope that non-regular visitors will quickly get the idea that the highest voted answers are likely to be the best ones, and that answers with a negative score really shouldn’t be trusted. It doesn’t always work that way, but it’s a reasonable rule of thumb.

So, how do I vote?

How I react to questions

I generally upvote a question if it’s been well written and I think the problem is sufficiently common that it’s going to help someone else searching for it. I’ll also upvote it if it’s particularly interesting, even if it’s not a very general problem. I suspect I should upvote questions more often – and I also suspect that’s true of many users.

I very rarely downvote a question, however. If the question is inappropriate, I’ll usually vote to close it. If it’s badly written but intelligible, I’ll edit it. If it isn’t precise enough (just lacks information) I’ll ask for more information in a comment. I don’t see much use in downvoting. Now there are various users who don’t support the idea of closing a question at all, of course – and if closing weren’t an option then I would downvote instead. However, I personally support question closing (when appropriate, of course – I’m not saying every closed question deserved it).

How I react to answers

I will generally upvote an answer if I feel it’s correct and helpful. If there are multiple posts which effectively answer a question, I will usually upvote the best one, but others which provide other bits of relevant information may get a vote too. Again, I probably don’t upvote as often as I should.

I downvote if I see an answer as actively unhelpful: this is usually if it’s technically inaccurate, or suggests something which I think is a really bad idea (such as string concatenation in a loop without a small, known limit). I don’t downvote an answer just for being "not as helpful as it could be" or "not as helpful as another answer". I believe that behaviour discourages people from contributing in the first place, and an extra answer has relatively little cost associated with it. If it contains no information which isn’t present in another answer then I’d prefer it to be deleted, but I’d leave that suggestion as a comment rather than a downvote.

Speaking of comments, I practically always leave a comment when I downvote. A downvote means I believe something is really wrong with the answer, and it should be fixed: leaving it as it is makes the world a worse place than if it didn’t exist. A downvote without a comment is fairly pointless – it doesn’t help the poster to fix the answer, because they don’t know what they did wrong in the first place. I find it intensely frustrating when someone downvotes one of my answers without giving any reason: apart from anything else, that downvote could be made on the basis of a mistaken belief, but without that belief being expressed I have no way of correcting it. I always take another look at an answer which has been downvoted, but I’m much more likely to edit it if I’m given a specific reason.

Note that the reputation loss due to a downvote is almost insignificant – particularly if it’s early in the day (i.e. before very active users hit the 200 cap) – the idea that I’ve written something unhelpful is far more disconcerting to me than the loss of a tiny amount of rep.

I can see why users aren’t prompted for a comment on a downvote – it would be very easy to just type garbage, and that would be worse than no comment at all. It would also require the comment to be anonymised in order to keep the vote itself anonymous. Even so, I’d ask courteous readers to add a comment when you downvote one of my answers: I promise not to "retaliate" with a spate of downvotes, but I’d really like to be able to fix the answer!

In terms of editing, I will often edit an answer for formatting reasons or to correct a small typo, but I edit answers (from other people) less often than I edit questions.

How about you?

Enough about me – how do you vote? If you downvote without comments, what effect are you trying to achieve? When would you downvote a question rather than editing it, voting to close or leaving a comment? How do you react to downvotes to your own posts?

C#, Java, Wacky Ideas

A different approach to inappropriate defaults

8 Comments

I’ve had a couple of bug reports about my Protocol Buffers port – both nicely detailed, and one including a patch to fix it. (It’s only due to my lack of timeliness in actually submitting the change that the second bug report occurred. Oops.)

The bug was in text formatting (although it also affected parsing). I was using the default ToString behaviour for numbers, which meant that floats and doubles were being formatted as "50,15" in Germany instead of "50.15". The unit tests caught this, but only if you ran them on a machine with an appropriate default culture.

Aaargh. I’ve been struggling with a similar problem in a library I can’t change, which uses the system default time zone for various calculations in Java. When you’re running server code, the default time zone is almost never the one you want to use, and it certainly isn’t in my case.

A similar problem is Java’s decision to use the system default encoding in all kinds of bizarre places – FileReader doesn’t even let you specify the encoding, which makes it almost entirely useless in my view.

So I’ve been wondering how we could fix this and problems like it. One option is to completely remove the defaults. If you always had to pass in a CultureInfo/Locale, TimeZoneInfo/TimeZone, Encoding/Charset when you call any method which might be culturally sensitive.

Making life easier (in .NET)

It strikes me that .NET has a useful abstraction here: the assembly as the unit of deployment. (Java’s closest equivalent is probably a jar file, which probably gets messier.)

Within one assembly, I suspect in many cases you always want to make the same decision. For example, in protocol buffers I would like to use the invariant culture all the time. It would be nice if I could say that, and then get the right behaviour by default. Here are the options I’d like to be able to apply (for each of culture, time zone and character encoding – there may be others):

  • Use a culture-neutral default (the invariant culture, UTF-8, UTC)
  • Use a specific set of values (e.g. en-GB, Windows-1252, "Europe/London")
  • Use the system default values
  • Use whatever the calling assembly is using

Of course you should still have the option of specifying overrides on a per call basis, but I think this might be a way forward.

Thoughts? I realise it’s almost certainly too late for this to actually be implemented now, but would it have been a good idea? Or is it just an alternative source of confusion?

C#, F#, General, Java

Language proliferation

16 Comments

I’ve always been aware that .NET supports multiple languages (obviously) and that Microsoft has been experimenting with this to some extent. It’s only recently struck me just to what extent this is the case though.

Here’s a list – almost certainly incomplete – of .NET languages from Microsoft alone.

Some of these are research languages which are more important for the ideas they’ve contributed to more mainstream ones at a later date than for anything else – but there’s still a lot of effort represented in the list.

In addition, there are third party languages targeting .NET, such as Boo, IronScheme and Scala. (Wikipedia lists loads of them.)

Now, think back to the time before .NET. Was Microsoft actively experimenting with languages back then? Plenty of people were trying things against the JVM, but Sun was pretty much absent from that party. .NET seems to be a "missing ingredient" that has allowed smart folk at Microsoft to let their imaginations loose in ways which they couldn’t previously. (Of course, not everyone in the language business at MS started there: Jim Hugunin was hired by Microsoft precisely because of his work on IronPython.)

I wonder how long this will continue.

Tower of Babel, or land of polyglots?

What does this mean for the average developer? Currently, if you’re writing a non-web application in .NET, you really only need to know a single language – and any of them will do. (Plus potentially SQL of course…) Compare this with web developers who have to be intimately familiar with HTML, CSS and JavaScript – and the differences between various implementations.

How long will it be before backend developers are expected to know a dynamic language, a static OO language and a functional language? Does the benefit of mixing several languages in a project worth the impedance mismatch and the increased skillset requirements? I’m not going to make any predictions on that front – I can certainly see the benefits of each of these approaches in certain situations. They’ve been designed to play well together, but there are bound to be limitations and oddities: times when you need to change how you write your F# so that it’s easily callable from C#, for example.

Whether or not you learn multiple languages to a professional level is one thing, but becoming familiar with them is a different matter. In the course of co-authoring Functional Programming for the Real World (where "co-author" is a bit of a stretch title – I’ve played more of an editorial role really, with the added bonus of picking on Tomas whenever I felt he was perhaps a little harsh towards C#) I’ve learned to appreciate many of F#’s qualities, but I don’t really know the language. If someone asked me to write a complete application in it (rather than just a toy experiment) I’d be reaching for books every other minute. I hope I’ll learn more over the course of time, but I doubt that I’ll ever be sufficiently experienced in it to put it on my CV. The same goes for IronPython, although I’m considerably more likely to need Python at work than I am F#. (Python is one of the three "approved" languages at Google, along with Java and C++.) None of this means that time spent in these languages is wasted: I’ll be able to apply a lot of what I’ve learned about F# to my C# coding, even if it will make me pine for things like pattern matching and asynchronous workflows periodically.

I think it’s pretty much a given that these days we all need to bring a wide range of technologies to bear in most jobs. While it used to be just about feasible in the .NET 1.1 days to have a pretty good grasp of all the major aspects (ASP.NET for sites and web services, ADO.NET, WinForms, Windows services, class libraries, interop) it’s just impossible these days. We learn something new when we need to – but usually against the background of a familiar language. How well would we cope if we had to learn whole new languages (to the level of being able to use them for production code) as often as we have to learn new libraries?

This worries me a little. I’m pleased to see that C# 4 is a much smaller change than the previous versions were. Admittedly I’d rather have had immutability support than dynamic, but that’s just me… and that’s the problem, too. While I worry about our ability to actually learn everything that’s becoming available, it’s all good stuff. Can there be "too much of a good thing"?

What I really don’t want to see is developers having to know multiple languages, and everyone knowing them poorly. I’m a big believer in having a thorough understanding of your language, so that even if everything else is new, you can rely on your understanding of that aspect of your code. It would be a shame if the pressure of knowing many languages turned many of us into cargo cult programmers. The utopia would be for us all to turn into language renaissance developers. I suspect the reality will be somewhere between the two.

Still, as long as I get to keep helping authors write about languages I know almost nothing about, I’m sure I’ll be happy…

Books, C#

RFC: C# in Depth 2nd edition, proposed changes and additions

42 Comments

As I’ve mentioned in passing before now, I’ve started working on the 2nd edition of C# in Depth, to roughly coincide with the release of C# 4 and Visual Studio 2010. So far I’ve just been thinking about what should be added and what should be changed or removed. That’s what I’d like to share today, and I’m really eager for feedback. In particular, I’d like to think of three audiences:

  • Developers (possibly hobbyists) who have read one introductory C# book (Head First C#, Microsoft Visual C# 2008 Step by Step or something similar) and are now looking to follow it up. I’m thinking of people with less than a year of full-time development in C#, and not a lot of experience in similar languages. Is C# in Depth too “hard” for this audience? If so, should I try to make it more accessible for them, or is the material inherently too advanced? What might attract such a person to the book in terms of the table of contents?
  • Reasonably experienced developers who know C# 2 and 3 already, but haven’t read C# in Depth. They may have heard of it before (I’ve been very pleased in terms of word-of-mouth for the first edition, so thanks to everyone who’s contributed to that!) but not had much real reason to buy another book when they already have something like C# 3.0 in a Nutshell. Now that C# 4 is around, they may be interested in buying something which will explain the new features, and they might as well get a different book rather than just a fresh edition of something they already own.
  • Existing readers of C# in Depth. If you already own the first edition, what might persuade you to buy the second edition? How much “new” stuff is required? Would you rather get “more bang for the buck” from a significantly thicker book, or does the relative slimness of C# in Depth hold real appeal? (I know I’m a fan of slim books, but I don’t know quite how important it is. The second edition certainly will be thicker than the first as I’m unlikely to remove much, but I’m sure there’s scope for varying just how much we add.)

Here’s a draft table of contents – only down to headings within a chapter, but with some notes. Changes and notes are highlighted in blue.

Part 1: Preparing for the journey

Chapter 1: The changing face of C# development
  • Evolution in action: examples of code change
    I won’t be able to include the C# 4 features by just evolving the existing code… at least, not the big ones. Options:
    • Stick with the existing example for 1 to 2 and 2 to 3, but change to a different problem for 4
    • Try to work out a different example to show all of 1 to 2, 2 to 3 and 3 to 4
    • Stick with the existing example, but take it in a different direction for 4, rather than just rewriting the existing code
  • A brief history of C# (and related technologies)
    Update this with what’s been happening since 2008 and possibly slim down what’s already there
  • The .NET platform
    Add the new versions, including .NET 3.5 SP1, and introduce the DLR in the terminology part.
  • Fully functional code in snippet form
    Nothing new to add about Snippy, but there may be another tool to talk about as well.
  • Summary
Chapter 2: Core foundations: building on C# 1
  • Delegates
  • Type system characteristics
  • Value types and reference types
  • Beyond C# 1: new features on a solid base
    Introduce interface/delegate variance and the dynamic type.
  • Summary

Part 2: Solving the issues of C# 1

Chapter 3: Parameterized typing with generics

  • Why generics are necessary
  • Simple generics for everyday use
  • Beyond the basics
    Add a tip and example about using generic static methods in a non-generic type to take advantage of type inference when creating instances of generic types.
  • Advanced generics
  • Generic collection classes
    It would probably make sense to include HashSet<T> here, and possibly any new collection classes in .NET 4.0. Version warnings would be given, of course!
  • Limitations of generics in C# and other languages
    Refer to part 4 in terms of variance, and possibly trim down the coverage of alternative approaches.
    Add an example of Marc Gravell’s work for generic operators, or at least refer to it. Possibly mention the idea of     static interfaces. Both give food for thought, but aren’t really part of C#.
  • Summary
Chapter 4: Saying nothing with nullable types
  • What do you do when you just don’t have a value?
  • System.Nullable<T> and System.Nullable
  • C# 2’s syntactic sugar for nullable types
  • Novel uses of nullable types
    Possibly include Marc Gravell’s teaser about calling all the Object methods on a null value.
  • Summary
    Make sure the summary is typeset as 4.5 instead of as part of 4.4. Oops!
Chapter 5: Fast-tracked delegates
  • Saying goodbye to awkward delegate syntax
  • Method group conversions
  • Covariance and contravariance
    Mention generic delegate variance in C# 4, referring to part 4.
  • Inline delegate actions with anonymous methods
  • Capturing variables in anonymous methods
  • Summary
Chapter 6: Implementing iterators the easy way
  • C# 1: the pain of handwritten iterators
  • C# 2: simple iterators with yield statements
  • Iteration in the real world
    Renamed section, but keep first example – possibly make it shorter.
    Add iteration over lines in a file.
    Add generation iterator, referring to LINQ.
  • Pseudo-synchronous code with the Concurrency and Coordination Runtime
    Fix sample code and explanation, with full article on the web
  • Summary
Chapter 7: Concluding C# 2: the final features
  • Partial types
  • Static classes
  • Separate getter/setter property access
  • Namespace aliases
  • Pragma directives
  • Fixed-size buffers in unsafe code

Part 3: New title TBD

Chapter 8: Cutting fluff with a smart compiler
  • Automatically implemented properties
  • Implicit typing of local variables
  • Simplified initialization
  • Implicitly typed arrays
  • Anonymous types
  • Summary
Chapter 9: Lambda expressions and expression trees
  • Lambda expressions as delegates
  • Simple examples using List<T> and events
  • Expression trees
    Expand for new expression support in .NET 4.0 (clearly labeled as such)
  • Changes to type inference and overload resolution
    Check if/how this has changed in C# 4.0 spec
  • Summary

 

Chapter 10: Extension methods
  • Life before extension methods
  • Extension method syntax
  • Extension methods in .NET 3.5
  • Usage idea and guidelines
    Update advice here and give some more examples.
  • Summary
Chapter 11: Query expressions and LINQ to Objects

Look at changing how the diagrams are formatted in this chapter.

  • Introducing LINQ
  • Simple beginnings: selecting elements
  • Filtering and ordering a sequence
  • Let clauses and transparent identifiers
  • Joins
  • Groupings and continuations
  • New section: extending LINQ to Objects
    • Advice for writing your own extension methods
    • Plug for MoreLINQ and other libraries
  • Summary
Chapter 12: LINQ beyond collections
  • LINQ to SQL
    Despite the rumours of its doom, I think this is the best place to start.
  • Translations using IQueryable and IQueryProvider
  • LINQ to DataSet
  • LINQ to XML
  • LINQ to Entities
    Again, not much depth – still a whirlwind tour.
  • Third-party LINQ (renamed from LINQ beyond .NET 3.5)
    • Update state of play with third-party providers
    • Example of PushLINQ – still using simple delegates, but against different interface
    • Remove Parallel Extensions – moved to part 4
  • Summary
Chapter 13: Removed (new chapter at end of part 4)

Part 4: C# 4 (full title TBD)

Chapter 13: Dynamic binding in a static language
  • Introduction to the DLR, IronPython, IronRuby
  • Calling dynamically – the dynamic keyword
  • Reacting dynamically – implementing IDynamicObject
  • Applications for dynamic code
  • Summary
Chapter 14: More minor tweaks
  • Named and optional arguments
  • Variance of interfaces and generic delegates
  • Simplifying COM interoperability
  • Summary
Chapter 15: Major new features of .NET 4.0

Even though these don’t affect the language directly, they change the “shape” of code, the way we think about problems. I’m hoping there will be more…

  • Bullet-proofing your code with Code Contracts
  • Simplifying concurrency with Parallel Extensions
  • F# (?)
  • Summary
Chapter 16: Whither now?
  • Educated guesses about C# 5
  • The Renaissance developer
  • Until we meet again
Appendix: LINQ standard query operators

So, how does that sound? What else needs changing from the first edition? Are there any particular sections which would benefit from a neat example? Have I missed anything important that you’d want to see in part 4?

Right now I’ve got a lot of flexibility – so please, I’d much rather hear ideas now than later.

 

Stack Overflow, Wacky Ideas

Go on, ask me anything

28 Comments

This afternoon, I found a comment which had been trapped in the spam bin for this blog. It was from Andrew Rimmer, in reply to my “micro-celebrity” post, pointing me at http://askjonskeet.com

The world has officially become extremely silly. The surprising thing is, it’s actually useful – at least for me. A number of times I’ve wanted to find my old answers to questions, so I can either just refer to them in a new answer or mark the new question as a dupe. You might have thought that a simple search such as

exceptions site:stackoverflow.com “jon skeet”

would suffice – but that finds what other people have said about exceptions in the same questions that I’ve been active in, and it also picks up any questions which happened to get one of the FinalBuilder adverts when the spider fetched them. The equivalent search on AskJonSkeet.com gets good results.

I’ve no idea how useful it will be for anyone else, but personally I love it. Ego? What ego?

Side note to self, puncturing ego slightly: don’t blog on the tube. It’s way too easy to miss your stop…

Benchmarking, Parallelization

Buffering vs streaming benchmark: first results

11 Comments

My poor laptop’s had a busy weekend. It’s run 72 tests, rebooting between each test. Most of these tests have kept both the CPU and disk busy for a lot of the time. I expect to update this blog post with more numbers – and possibly more strategies – as time goes on, but I wanted to get the numbers out as quickly as possible. Graphs should be coming when I’ve got a decent network to experiment with Google graphing.

Source code and setup

While I don’t really expect anyone else to sacrifice their computer for the best part of 24 hours, it doesn’t take very much effort to get the tests running. This zip file contains the source code for three programs, a batch file to generate the test data, and a command file.

The CreateFiles executable creates input files based on its command line arguments. (This is run multiple times by the batch file.) EncryptFiles “encrypts” (I use the term loosely) all the files in a given directory, writing the results to another directory. The class design of EncryptFiles is a little bit funny, but it’s really tailored for these tests. It should be easy to write other strategies in the same way too.

ExecuteAndReboot is used to automate the testing. Basically it reads a command file (from a hard-coded directory; simplicity trumps flexibility here). If the file isn’t empty, it waits five minutes (so that after a reboot the computer has time to settle down after loading all its services) and runs the command specified in the first line of the file, appending the output to a results file. When the command has finished executing, ExecuteAndReboot rewrites the command file, removing the first line, and schedules a reboot. The idea is to drop ExecuteAndReboot into a Startup folder, set Windows to log in automatically, reboot once and then just let it go. If you want to use the computer, just wait until it’s in the “waiting five minutes” bit at the start of the cycle and kill ExecuteAndReboot. Next time you reboot it will start testing again.

That’s the general principle – if anyone does want to run the tests on their mail and needs a bit of help setting it up, just let me know.

Test environment and workload

I expect the details of the execution environment to affect the outcome significantly. For the sake of the record then, my laptop specs are:

Dell Inspiron 1720 (unlikely to be relevant, but…)
Intel Core 2 Duo T7250 2.0GHz
Vista Home Premium 32 bit
3 GB main memory, 32K L1 cache, 2048K L2 cache
2 disks, both 120GB (Fujitsu MHY2120BH, 5400RPM, 8MB cache)

Even though there are two drives in the system, the tests only ever read from and wrote to a single disk.

I created four sets of input data; I’ve presented the results for each one separately below, and included the description of the files along with the results. Basically in each case there’s a set of files where each file is the same size, consisting of a fixed number of fixed-length lines.

The “encryption” that is performed in each case (on the encoded version of the line) is just to XOR each byte with a value. This is repeated a specified number of times, and the XOR value on each pass is the current index, i.e. it first XORs with 0, then 1, then 2 etc. I’ve used work levels of 0 (the loop never runs), 100 and 1000. These were picked somewhat arbitrarily, but they give interesting results. At 0 the process is clearly IO-bound. At 1000 it’s clearly CPU-bound. At 100 it’s not pegging the CPU, but there’s significant activity. I may try 200 as well at some point.

For each scenario I tried using 1, 2 and 3 threads (and 4 for the big tests; I’ll go back and repeat earlier ones with 4 threads soon). The two strategies used are “streaming”: read a line, encrypt, write the line, repeat and “buffering”: read a whole file, encrypt it all, write it all out. After a thread has started, it needs no shared data at all – it’s given the complete list of files to encrypt at the very start.

Results

For every table, the number of threads is shown by the leftmost column, and the work level is shown by the topmost row. The cells are of the form “x/y” where “x” is the time taken for the streaming strategy, and “y” is the time taken for the buffering strategy. All times are in seconds. For each column, I’ve highlighted the optimal test.

Small tests

There are 10000 test files. Each is 100 lines long, with 100 characters per line.

Threads/Work 0 100 1000
1 47/90 94/79 163/158
2 95/112 113/157 129/146
3 179/128 130/144 152/160
4 144/139 163/193 156/204

Medium tests (short lines)

There are 100 test files. Each is 500,000 lines long, with 20 characters per line.

Threads/Work 0 100 1000
1 138/157 219/267 1125/1205
2 196/238 259/263 604/691
3 292/250 312/236 624/676
4 302/291 321/281 602/666

Medium tests (long lines)

There are 100 test files. Each is 100 lines long, with 100,000 characters per line.

Threads/Work 0 100 1000
1 203/213 249/286 1027/1107
2 283/264 311/296 602/696
3 319/284 314/278 608/618
4 360/300 336/297 607/598

Big tests

There are 20 files. Each file is 100,000 lines long, with 1000 characters per line.

Threads/Work 0 100 1000
1 190/246 349/441 2047/2165
2 365/375 392/473 1097/1387
3 456/379 484/399 1161/1296
4 517/442 521/431 1113/1214

Conclusions

I’m loathe to draw too many conclusions so far. After all, we’ve only got data for a single test environment. It would be interesting to see what would happen on a quad core machine. However, I think a few things can be said with reasonable confidence:

  • Sometimes the streaming strategy wins, sometimes the buffering strategy wins. The difference can be quite significant either way. For a given input and workload, the streaming solution almost always wins on my machine.
  • In ideal scenario, the bottlenecked resource should be busy for as much of the time as possible. For example, in a high-CPU task it makes little sense to have idle cores when you’ve already loaded data. Likewise in a disk-bound task we should always be fetching (or writing) data even if all our cores are momentarily busy. Personally I think this is the operating system’s pre-fetch systems’s job. We could do it ourselves, but it’s nicer if we don’t have to.
  • If we’re disk-bound, it makes sense to read (or write) one file at a time, to avoid seeking. This is obvious from the results – for work levels of 0 and 100, a single-thread solution is consistently best (and streaming usually works better buffering). Based on this assumption, I suspect a cleaner solution would be to have a single thread reading largish chunks (but not the whole file) and feeding the other threads – or to use async IO. However, this all gets very complicated. One nice aspect of both of the current solutions is that they’re really simple to implement safely. Reading line-by-line in an async manner is a pain. I’d quite like to write an async “execute for every line” helper at some time, but not just yet.
  • If we’re CPU bound, the buffering solution’s sweet spot is 3 threads (remember we’re using 2 cores) and the streaming solution works better with just 2 threads – introducing more threads will make both the IO and context switching less efficient.
  • The streaming strategy always uses less application memory than the buffering strategy. When running the buffering strategy against the “big” data set with 4 threads, the memory varied between 800MB and 1.8GB (as measured by the “Total bytes in all heaps” performance counter), vs around 165KB with the streaming strategy. (Obviously the process itself took more than 165KB, but that was the memory used in “normal” managed heap memory.) Using more memory to improve performance is a valid technique, but I’d say it’s only worth it when the improvement is very significant. Additionally, the streaming technique instantly scales to huge data files. While they’re not in the current requirements, it doesn’t hurt to get that ability for free.

Next steps…

I’ve included quite a few “I’m going to try to […]” points in this post. Just for reference, my plans are:

  • Turn off my virus checker! If this makes a significant difference, I’ll rerun all the current tests
  • Try to visualise the results with Google graphs – although I think it may get cluttered
  • Rerun big/buffering/1000 test – odd results
  • Rerun tests with a work level of 200 and keep an eye on CPU usage – I’d like to spot where we tip between CPU and IO being the bottleneck
  • Try to optimize the Stream/StreamReader being used – we should be able to skip the FileStream buffer completely, and specifying FileOptions.SequentialScan should give more hints to Windows about how we’re intending to read the data. I’ll try a few different buffer sizes, probably 4K, 8K and 64K, just to see what happens.

This is a really interesting little example problem, because it sounds so simple (and it is simple in terms of the implementation) but it has a lot of environment variables. Of course it would be nicer if we didn’t have to reboot the machine between test runs in order to get a fair result, but never mind…

Benchmarking, C#, Parallelization

Benchmarking IO: buffering vs streaming

24 Comments

I mentioned in my recent book review that I was concerned about a recommendation to load all of the data from an input file before processing all of it. This seems to me to be a bad idea in an age where Windows prefetch will anticipate what data you need next, etc – allowing you to process efficiently in a streaming fashion.

However, without any benchmarks I’m just guessing. I’d like to set up a benchmark to test this – it’s an interesting problem which I suspect has lots of nuances. This isn’t about trying to prove the book wrong – it’s about investigating a problem which sounds relatively simple, but could well not be. I wouldn’t be at all surprised to see that in some cases the streaming solution is faster, and in other cases the buffered solution is faster.

The Task

The situation presented is like this:

  • We have a bunch of input files, either locally or on the network (I’m probably just going to test locally for now)
  • Each file is less than 100MB
  • We need to encrypt each line of text in each input file, writing it to a corresponding output file

The method suggested in the book is for each thread to:

  1. Load a file into a List<string>
  2. Encrypt every line (replacing it in the list)
  3. Save to a new file

My alternative option is:

  1. Open a TextReader and a TextWriter for the input/output
  2. Repeatedly read a line, encrypt, write the encrypted line until we’ve exhausted the input file
  3. Close both the reader and the writer

These are the two implementations I want to test. I strongly suspect that the optimal solution would involve async IO, but doing an async version of ReadLine is a real pain for various reasons. I’m going to keep it simple – using plain threading, no TPL etc.

I haven’t written any code yet. This is where you come in – not to write the code for me, but to make sure I test in a useful way.

Environmental variations

My plan of attack is to first write a small program to generate the input files. These will just be random text files, and the program will have a few command line parameters:

  • Directory to put files under (one per test variation, basically)
  • Number of files to create
  • Number of lines per file
  • Number of characters per line

I’ll probably test a high and a low number for each of the last three parameters, possibly omitting a few variations for practical reasons.

In an ideal world I’d test on several different computers, locally and networked, but that just isn’t practical. In particular I’d be interested to see how much difference an SSD (low seek time) makes to this test. I’ll be using my normal laptop, which is a dual core Core Duo with two normal laptop disks. I may well try using different drives for reading and writing to see how much difference that makes.

Benchmarking

The benchmark program will also have a few command line parameters:

  • Directory to read files from
  • Directory to write files to
  • Number of threads to use (in some cases I suspect that more threads than cores will be useful, to avoid cores idling while data is read for a blocking thread)
  • Strategy to use (buffered or streaming)
  • Encryption work level

The first three parameters here are pretty self-explanatory, but the encryption work level isn’t. Basically I want to be able to vary the difficulty of the task, which will vary whether it ends up being CPU-bound or IO-bound (I expect). So, for a particular line I will:

  • Convert to binary (using Encoding.ASCII – I’ll generate just ASCII files)
  • Encrypt the binary data
  • Encrypt the encrypted binary data
  • Encrypt the encrypted encrypted […] etc until we’ve hit the number given by the encryption work level
  • Base64 encode the result – this will be the output line

So with an encryption work level of 1 I’ll just encrypt once. With a work level of 2 I’ll encrypt twice, etc. This is purely for the sake of giving the computer something to do. I’ll use AES unless anyone has a better suggestion. (Another option would be to just use an XOR or something else incredibly simple.) The key/IV will be fixed for all tests, just in case that has a bearing on anything.

The benchmarking program is going to be as simple as I can possibly make it:

  • Start a stopwatch
  • Read the names of all the files in the directory
  • Create a list of files for each thread to encrypt
  • Create and start the threads
  • Use Thread.Join on all the threads
  • Stop the stopwatch and report the time taken

No rendezvous required at all, which certainly simplifies things. By creating the work list before the thread, I don’t need to worry about memory model issues. It should all just be fine.

In the absence of a better way of emptying all the file read caches (at the Windows and disk levels) I plan to reboot my computer between test runs (which makes it pretty expensive in terms of time spent – hence omitting some variations). I wasn’t planning on shutting services etc down: I really hope that Vista won’t do anything silly like trying to index the disk while I’ve got a heavy load going. Obviously I won’t run any other applications at the same time.

If anyone has any suggested changes, I’d be very glad to hear them. Have I missed anything? Should I run a test where the file sizes vary? Is there a better way of flushing all caches than rebooting?

I don’t know exactly when I’m going to find time to do all of this, but I’ll get there eventually :)

C#, General, Wacky Ideas

Breaking Liskov

31 Comments

Very recently, Barbara Liskov won the Turing award, which makes it a highly appropriate time to ponder when it’s reasonable to ignore her most famous piece of work, the Liskov Substitution (or Substitutability) Principle. This is not idle speculation: I’ve had a feature request for MiscUtil. The request makes sense, simplifies the code, and is good all round – but it breaks substitutability and documented APIs.

The substitutability principle is in some ways just common sense. It says (in paraphrase) that if your code works for some base type T, it should be able to work with subtype of T, S. If it doesn’t, S is breaking substitutability. This principle is at the heart of inheritance and polymorphism – I should be able to use a Stream without knowing the details of what its underlying storage is, for example.

Liskov’s formulation is:

Let q(x) be a property provable about objects x of type T. Then q(y) should be true for objects y of type S where S is a subtype of T.

So, that’s the rule. Sounds like a good idea, right?

Breaking BinaryReader’s contract

My case in point is EndianBinaryReader (and EndianBinaryWriter, but the arguments will all be the same – it’s better to focus on a single type). This is simply an equivalent to System.IO.BinaryReader, but it lets you specify the endianness to use when converting values.

Currently, EndianBinaryReader is a completely separate class to BinaryReader. They have no inheritance relationship. However, as it happens, BinaryReader isn’t sealed, and all of the appropriate methods are virtual. So, can we make EndianBinaryReader derive from BinaryReader and use it as a drop-in replacement? Well… that’s where the trouble starts.

There’s no difficulty technically in doing it. The implementation is fairly straightforward – indeed, it means we can drop a bunch of methods from EndianBinaryReader and let BinaryReader handle it instead. (This is particularly handy for text, which is fiddly to get right.) I currently have the code in another branch, and it works fine.

And I would have gotten away with it if it weren’t for that pesky inheritance…

The problem is whether or not it’s the right thing to do. To start with, it breaks Liskov’s substitutability principle, if the “property” we consider is “the result of calling ReadInt32 when the next four bytes of the underlying stream are 00, 00, 00, 01” for example. Not having read Liskov’s paper for myself (I really should, some time) I’m not sure whether this is the intended kind of use or not. More on that later.

The second problem is that it contradicts the documentation for BinaryReader. For example, the docs for ReadInt32 state: “BinaryReader reads this data type in little-endian format.” That’s a tricky bit of documentation to understand precisely – it’s correct for BinaryReader itself, but does that mean it should be true for all subclasses too?

When I’ve written in various places about the problems of inheritance, and why if you design a class to be unsealed that means doing more design work, this is the kind of thing I’ve been talking about. How much detail does it make sense to specify here? How much leeway is there for classes overriding ReadInt32? Could a different implementation read a “compressed” Int32 instead of always reading four bytes, for example? Should the client care, if they make sure they’ve obtained an appropriate BinaryReader for their data source in the first place? This is basically the same as asking how strictly we should apply Liskov’s substitutability principle. If two types are the same in every property, surely we can’t distinguish between them at all.

I wonder whether most design questions of inheritance basically boil down to defining which properties should obey Liskov’s substitutability principle and which needn’t, for the type you’re designing. Of course, it’s not just black and white – there will always be exceptions and awkward points. Programming is often about nuance, even if we might wish that not to be the case.

Blow it, let’s do it anyway…

Coming back to BinaryReader, I think (unless I can be persuaded otherwise) that the benefits from going against the documentation (and strict substitutability) outweigh the downsides. In particular, BinaryReaders don’t tend to be passed around in my experience – the code which creates it is usually the code which uses it too, or it’s at least closely related. The risk of breaking code by passing it a BinaryReader using an unexpected endianness is therefore quite low, even though it’s theoretically possible.

So, am I miles off track? This is for a class library, after all – should I be more punctilious about playing by the rules? Or is pragmatism the more important principle here?