Advogato blog for redi

17 Jul 2008

Thu, 17 Jul 2008 14:50:02 GMT

I agree with lkcl that you can avoid being marked as a spammer by taking 30 seconds to say what your involvement is in open-source software. Linking to a brand new Wordpress account that has no content doesn't convince me you're doing anything except preparing to spam (see one of today's new accounts.)

Create a new account could recommend being more descriptive, to protect against the trigger-happy spam police (I include myself in that category.)

11 Jul 2008

Fri, 11 Jul 2008 12:43:25 GMT

ncm, yep, I only mentioned the one that most annoyed me, as I'd already done one hatchet job on it in the #cplusplus channel at work. There are a few sensible rules, though they tend to be in the blindingly-obvious category, rather than useful e.g. "Use the C++ keyword explicit for constructors with one argument" - not really a style tip, more like "remember to tie your laces when you put your shoes on."

Update: I hadn't noticed they require ALL_CAPS for enumerator names. Epic fail.

10 Jul 2008

Thu, 10 Jul 2008 16:41:28 GMT

Having read the Google C++ Style Guide I can safely say I would never work for Google. Any style guide that recommends 2-stage construction "if at all possible" is shit. That style is, as Stroustrup said eight years ago, a relic of pre-exception C++.

10 Jul 2008

Thu, 10 Jul 2008 15:44:24 GMT

Dear Advogato,

I've been too busy to read you recently, except for a few people's diaries that I watch for updates. I hope you don't mind.

I thought you might like to know I released a new version of PStreams yesterday, but you probably don't care. I've also resumed porting STLsoft and Pantheios to Solaris, and made my biggest change yet to GCC: GNU Compiler Collection, fixing something I helped break back in February this year.

Until next time.

22 Apr 2008

Tue, 22 Apr 2008 05:07:36 GMT

movement, brilliant! Humm, Pastor Roger Byrd ... "Roger" is so close to "Roget" I think he may be a thesaurus and therefore doesn't recognise god.

ncm, thanks for sharing that. As an ex-physicist who is very skeptical of the entire astronomy/cosmology field I think the poster's explanation for the behaviour makes a lot of sense.

17 Apr 2008

Thu, 17 Apr 2008 12:05:27 GMT

First there is a mountain, then there is no mountain, then there is a mountain.

The three stages of expertise

Ruby is a cult.

You can ignore one or two fanboys who want to teach you why they're right and you're wrong, but after the hundredth one raving about Ruby's unique features it just gets laughable.

Hint: Smalltalk. C'mon, kids, try to explain why Ruby is good without the rhetoric, you're just exposing your ignorance.

1 Apr 2008

Tue, 1 Apr 2008 17:55:11 GMT

I'm off to the ACCU conference tomorrow and have just bought an Asus Eee on which I plan to install GHC in time for Simon Peyton-Jones' talk. I'm pretty impressed with it so far - it meets all my criteria for a convenient way to ssh to my home PC (cheap, lightweight, wireless, *nix, proper keyboard, fits in my larger pockets.)

Over the past week I've been porting STLsoft to Sun's compilers and Solaris. Pantheios now works with Sun compilers on Linux or Solaris. Once I submit those patches my next step is to get it working with GCC on Solaris. I must say, Solaris system headers are a lot nicer to read than GNU ones. Solving a glibc include-order bug happened in spite of having access to the full source code, not because of it!

19 Mar 2008

Wed, 19 Mar 2008 17:34:26 GMT

Hank Paulson, who I had assumed to be a fairly well-educated man, said yesterday the US economy is in a "downcline". This is rather conturbing information, and a matter of real discern. We must keep our crongers fissed that it does not become a more serious deturn, I suppose.

Thinking about that stupid word make me regret recently coining the term "deducerator" for C++ functions such as std::back_inserter and std::make_pair. Although it was suggested as a joke, I do think it's more memorable than Object Generator, or the less specific "creator function" used by Matthew Wilson in his books. I think functional languages call them constructors, but that would obviously be confusing in C++.

jtauber, that's due to GNU bash. Bash remembers that you changed dir via the symlink, and preserves that information in your pwd, which is why bash's pwd command has a -P option to see the physical pwd, not symbolic one.

20 Feb 2008

Wed, 20 Feb 2008 15:26:50 GMT

cannam, to uncertify someone just certify them as Observer. You make a very good point otherwise, making it optional would be an improvement.

9 Feb 2008

Sat, 9 Feb 2008 17:56:23 GMT

f*** logging: a dead horse

(disclaimer: I'm suffering from a chronic case of man flu and have been cooped up indoors for 48 hours, this is a rant.)

I find the subject of logging as dull as ditchwater, so I don't want to do it myself, but I wish someone would write a decent, free logging framework for C++.

Please don't tell me about the "Boost" logging lib - it's been rejected from Boost at least once (twice I think.) You'd be forgiven for thinking it was part of Boost from the documentation or the author's cringe-worthy self-promotional website ('hey, doc, is there a cure for me?' no, you'll probably always be like that.)

The good thing about the non-boost lib is it tries to be a natural fit for C++, whereas the popular C++ logging libs are ports of log4j. And I guess being written by a raving egomaniac works well for some software projects! Let's see how its current boost review goes.

log4cpp & log4cplus are OK, as ports of log4j they have pretty stable and mature APIs, but both have some pretty hairy (and inefficient) aspects of their implementation. I checked Log4cxx out from subversion to have a look at that one. The INSTALL file says "Apache log4cxx is a C++ transliteration of Apache log4j" which is a bad start - C++ and Java have enough differences that porting from Java to C++ is not as simple as replacing Integer with int, the whole design should be evaluated and idioms native to the target language should be used. (Xerces/C is another project from Apache that makes the same mistake, as ncm has mentioned once or twice here.) The transliteration shows itself immediately in terms of broken const-correctness, which is easily explained when you see:


template<typename T>
class ObjectPtrT {
public:
    // ...
    T* operator->() const {return (T*) p; }
    T& operator*() const {return (T&) *p; }
    operator T*() const {return (T*) p; }
private:
    T* p;
};

I'll hazard a guess and say those casts were put there by someone who simply doesn't understand const in C++: p isn't being modified, so there's no need to cast away the const (in the context of those member functions, p has type T * const i.e. const pointer to non-const T)

Well, I thought, this is open-source software and I'm no stranger to writing reference-counted smart pointers in C++. It won't take me long to knock that smart pointer into shape, then I can continue evaluating the rest of the library. Even if I decide not to use it, the project will benefit from a better smart pointer. So I tried to build it. The project offers a GNU autotools build system which fails due to missing files in the svn repo, and an ant-based build system which requires a jar from ant-contrib which isn't included in the Fedora package, or the latest download on sourceforge. Harrumph. I've subscribed to the log4cxx mailing list to ask if I'm being dumb and am proceeding by compiling the code in my head (the project page says it's been tested with GCC 3.3, which handled templates in a non-standard way - the process of adding missing typename keywords and re-compiling is much quicker using a virtual compiler in my head! ;-)

There are various other things that would prevent me from using it for the moment, such as the function for atomic exchange of a pointer value being non-atomic on 64-bit systems (but hey, who uses them? oh, hang on, I do.)

Although I haven't given up on fixing log4cxx, it might be a slow process, and it's not an itch I want to scratch. Someone, please write a logging framework that doesn't suck. Or if you've done it already, publicise it better!

Update: following my email to the list the build system has been fixed within hours - that's a good sign! I'm continuing to review it, reassured it's an active and responsive project.