Re: Red Hat, 16%. Canonical, 1%.

While some people are busy complaining that Canonical doesn't contribute as much as others, I'd like for everyone to take a step back and ask themselves, "what would Gunny say?"

I'll tell you what he'd say. He'd say,

Hmm, that's interesting. Do you know what makes me sad? YOU DO! Maybe we should chug on over to mamby-pamby land where maybe we can find some self-confidence for you, you jack-wagon! Want a tissue?

After he finished ripping someone a new one, he'd point out that in that very same GNOME Census slide deck, I am ranked #8 in the top contributors list and I haven't contributed much of anything to any of the core GNOME components in about 5 years.

Yeah, that's right you cry babies, I put all y'all to shame.

On a more serious note, for better or worse, we all knew what we were getting into when we decided to take up the Free Software baton and start running with it. This is just how Free Software works. Don't like it? Cry me a river.

tl;dr

Update: It seems my post here has been misinterpreted. Allow me to try and rectify this. I'm not pissed off at Greg for voicing his opinion. We all do it. I'm just making a mockery of the whole situation because I was itching to trollcat and because I wanted to pat myself on the back for being #8 on the individual contributors list in terms of commits (as meaningless as that is). Also because I was up late last night and saw that Geico commercial with R. Lee Ermey which just cracks me up every time I see it.

Early Morning Sunrays

IMG_1886, originally uploaded by jstedfast.

Took this photograph around 6:30 in the morning on June 18th. The way the sunlight could be seen streaking through the trees and the complete stillness of the water was surreal, so I had to take a photograph of it ;-)

Dawns of June

Dawn at Wollaston Beach, MA on June 4th.



Dawn at Core Creek Park in Bucks County, PA on June 7th.



Dawn at Milton Pond, NH on June 17th.



Dawn at Milton Pond, NH on June 17th.

Reflecting on 10 Years at Ximian

IMG_1047, originally uploaded by jstedfast.

Today marks my 10th anniversary since I was hired at Helix Code to work on Evolution.

In that time I've gotten to work closely with and learn from some of the most talented developers in the Free Software community including Michael Zucchi, Miguel de Icaza, Federico Mena-Quintero, Chris Toshok, Larry Ewing, Michael Meeks, Dan Winship, Radek Doulik, Joe Shaw, Vlad Vukićević, Dave Camp, Dan Mills, and many others (too many to name!).

Back in the early days of Helix Code/Ximian, we all worked tirelessly to put together the very best GNOME distribution we could and make Evolution the very best groupware client we could.

We had big dreams and I like to think we succeeded. GNOME has become mainstream and the Linux Desktop can most definitely be measured as a success.

Of course, we couldn't have done it without all the help and hard work from the entire GNOME community.

Today, a younger generation of hackers are taking up the reigns to make GNOME awesome with things like the Paper Cuts project by David Siegel's team at Canonical and the work being done by Red Hat on the new GNOME-Shell.

I'm still working with a couple of my old Evolution teammates like Larry Ewing and Chris Toshok on the Moonlight project in the hopes of making it possible for Linux Desktop users to view Silverlight content on the web. We've been making great progress with implementing Silverlight 3.0 and 4.0 features in svn and I've got some accomplishments on that front that I should really blog about but I've just been so busy.

Much love to all you GNOMEies and thanks for all the support and love over the past decade!

Sundown over the Charles River

IMG_1022, originally uploaded by jstedfast.

Weird bugs due to gcc 4.4 and strict aliasing

I was just running the unit tests for GMime and I got a couple of failures on my openSUSE 11.2 machine that I have never gotten before. I started debugging and I noticed something very odd. One of my functions that returned a linked-list of 'word' tokens was returning NULL for something that it should not be returning NULL from, especially since I had just stepped through that method and seen with my own eyes that it was creating and appending nodes to my linked list as it should!

At this point I split out a tiny test case:

#include <stdio.h>
#include <stdlib.h>

typedef struct _node {
    struct _node *next;
    int value;
} Node;

int main (int argc, char **argv)
{
    Node *list, *node, *tail;
    int i;
    
    list = NULL;
    tail = (Node *) &list;
    
    for (i = 0; i < 10; i++) {
        node = malloc (sizeof (Node));
        node->next = NULL;
        node->value = i;
        
        tail->next = node;
        tail = node;
    }
    
    if (list == NULL)
        printf ("oops, list is null???\n");
    
    return 0;
}

I then built this test program using gcc -Wall -g -o list-test list-test.c. When I ran the program, no error.

Huh.

Luckily I thought to rebuild with -O2. This time when I ran my test program, it printed out the error I expected. The list was NULL.

Doing a bit of research suggests that gcc 4.4 has decided to enforce strict aliasing for -O2 and higher, which explains why it works without -O2.

Oddly enough, I get no strict aliasing warnings from gcc even when I use -Wall -O2.

So be warned... gcc 4.4 may break your perfectly valid code in mysterious ways.

Update: Looks like MySQL had similar problems and according to this article, it is suggested that the gcc developers have interpreted the c99 specification far too strictly.

Update 2: Josh's comment had an excellent explanation of the problem this code hits, so I thought I'd add it to my blog post for anyone who may otherwise miss his comment:

I imagine that gcc's aliasing thinks that dereferencing a Node* can only write to Node objects, so it won't have any effect on Node*s like list. Thus it thinks that list is never modified in that function, and it propagates the NULL constant throughout.

I also liked Josh's explanation of why the above code works (since a number of people have been confused by it already, I'll post that too):

The only reason it "works" in the -O0 case is because the next field happens to be the first field, as Fabian pointed out. The first time you write to tail->next, you're writing to &list with no offset, so list happens to get the value of node. When next does have an offset, then the value of list is not changed, but something else on the stack after it will be clobbered.

Trolls

Thanks go out to Stormy Peters and Jo Shields for tweeting a link to Seth Godin's blog post entitled: Trolls. Had they not tweeted it, I likely would have missed these marvelous words of wisdom:

Lots of things about work are hard. Dealing with trolls is one of them. Trolls are critics who gain perverse pleasure in relentlessly tearing you and your ideas down. Here's the thing(s):

1. trolls will always be trolling
2. critics rarely create
3. they live in a tiny echo chamber, ignored by everyone except the trolled and the other trolls
4. professionals (that's you) get paid to ignore them. It's part of your job.

"Can't please everyone," isn't just an aphorism, it's the secret of being remarkable.

Thanks, Seth. I needed to be reminded of this fact.

And yea, Team Mono is remarkable ;-)