|
|
| Monday 31st December 2001 |
|---|
resync.
Pinged Linus, and found out he was ready to resync. Spent the evening feeding up chunks of diff. Not a bad acceptance so far, probably around 95%.
The remaining 5% is either crap, or in need of rewriting. The good news is that all the 2.4.17 bits I sent so far have gone in. Pre4 had a bunch of merged bits, including dwmw2's asm/segment include crapectomy. Unfortunatly, he sent the wrong diff to me, I never noticed, neither did Linus, and it ended up in a somewhat sorry state. Pre5 got us back on track (which was put out mostly as Linus & myself lost track of what had been merged where). Sent a few more small bits, but most of the bigger bits like net core updates I'd rather leave to davem, and see whats left over afterwards.
faster builds.
Played with prefetch macros in read_file_actor() this evening whilst doing the resync. Managed to get a whole 5 seconds knocked off a kernel compile on my P3.
For every 12 kernels I build, I now save an extra minute of my life to do other things 8) Coupled with the include file shakeup that Manfred & Arnaldo are doing, hopefully we'll get the kernel build time down quite a bit in 2.5, which may be useful if/when kbuild2.5 gets merged in its current state. (It's 2x slower than kbuild-2.4)
TODO
Note to self. On tomorrows TODO list is to make a TODO of things that need doing in 2.5.
|
| Sunday 30th December 2001 |
|---|
kernels.
Put out a -dj8 fixing the reiserfs hang-on-mount problem, and a bunch of other fixes. Also included Thomas Hood's updated pnpbios work as I figured it could use some more testing. The pnpbios code should probably be mangled to fit the driverfs model thats being proposed for 2.5. If I get time I'll probably end up bending the EISA code to follow suit now that I've got a card to test with courtesy of Alan.
First step however is to get a set of screwdrivers that can remove the screws holding in the blanking plates on the quad ppro. Yep, quad ppro with EISA & PCI, sick huh ? 8)
And yup, I know I said in that interview that -dj wasn't going to turn into the next -ac. Since then I've had a steady stream of patches in the inbox from various developers saying they're using -dj as a basis for their work now, and I should keep it up. This worries me a little, but as long as people still send bits to Linus too, and don't rely on me to merge their work, then things _should_ work out ok. I hope. Did I ever say "I'll live to regret this but.." ? (Just kidding -- I'm having great fun doing this, I'm learning about areas of the kernel I never even thought about looking at before).
Philosophy.
I've found myself making posts to a number of non-technical posts to Linux-kernel today. I usually avoid these, but could not resist. Started discussing lots of things from how merging patches works, to how the development model of Linux-kernel works. As the person the conversation centered around was Larry Mcvoy, who knows what he'll come up with as the next solution to development problems 8-)
Dogma.
Rob (my housemate) and self are going through a Kevin Smith phase. "Jay and Silent bob strike back" a week or so ago, "Mall rats" last weekend, and "Dogma" this evening. Silent Bob is my personal hero of the month. |
| Saturday 29th December 2001 |
|---|
2.5.1-dj7
Pushed this out, with quite a few changes since the last release. Seems to be holding up mostly. The bug reports I've had have been bugs which are bio related, and hence also horked in Linus' tree. Oh, and 1-2 small stupid things like missing the setup-visws.c, and leaving behind an extraneous crufty CONFIG_EISA macro. The merge up until now has been fairly straightforward, but some parts have made me sit down and spend a while staring at diffs, and simultaneous code views of the same file in 3 trees. (dualhead monitors rock my world).
Reiserfs fun
One of the tricky bits of late has been reiserfs. With different sets of fixes going into 2.4 and 2.5, it's been 'fun' when they've clashed. So this evening I decided to make sure I've got things right by creating a scratch partition and running some stress tools. The plan went out the window when trying to mount the
partition after creating it caused a hang. Thankfully I could reproduce this in 2.5.2pre3 vanilla, so I set about trying to track that down before attempting to further debug my tree. Lots of stack tracing, disassembling and talking with Al Viro & Andrew Morton eventually turned up a bad one-liner that sneaked in during pre3. The good news is with this change, what will be -dj8 looks to be holding up quite well with reiserfs. Lets see what the multiple concurrent overnight runs of fsx turn up.
|
| Friday 28th December 2001 |
|---|
interview.
Got a steady stream of hits from kerneltrap.com for an hour or so, and then word got around to the slashdot crowd. The box stayed surprisingly understressed. With no TuX or other fancy bits, it happily served pages, with its loadavg never even going above 0.70. Not bad for a Celeron 366 w/256MB.
I did cheat slightly by removing the bandwidth whoring material like the photos from the indexes (I'll readd them in a few days). This seemed to be a good idea after 20mb was pumped upstream in 15 minutes.
One thing useful (other than server stress testing) from the slashdotting (*gasp* slashdot useful?) was the discovery of several old friends email addresses when a few "Is that you on slashdot?" and similar questioning emails arrived. Fun.
traveling.
Have to go to travel agent tomorrow to figure out cheapest way to get to .de during January. I'm anticipating problems here. (I *hate* dealing with travel agents).
Patchen.
No sooner had I merged 2.4.18pre1 & 2.5.2pre2, and released -dj6, when Linus puts out a pre3. Patch mayhem ensued, especially with conflicting reiserfs changes.
Changing guts of filesystems always makes me nervous, and the reiserfs changes are now quite large. Got everything under control whilst keeping an eye on server throughput. (Note: Announcing patches on a website experiencing slashdot effect causes 'interesting' effects). Started pulling together some other more experimental bits to merge, such as some of Manfreds changes that have been hanging about for a while. Managed to get his multithreaded coredump changes merged without breaking the ltp this time too. A few bits to double check before -dj7, including a strange change of behaviour with NFS file permissions that Manfred noticed.
Merging.
Tried to trick Linus into saying "I'm ready to merge some bits now". Failed.
|
| Thursday 27th December 2001 |
|---|
Xmas
Spent the past three days in Wales with the parents, and just got back to find not one, but two patches that need merging. The Linus/Marcelo tagteam patch releasing is getting to be fun to follow. Couple this with the mountain of fixes I pulled from the kernel list from the last month (I chose to catch up with old mail whilst the family watched 'the great escape' for the nth time). Looks like an afternoon of merging is called for. This should be fun.
toys
Favorite xmas present of all came from an old friend who managed to snarf some compaq dimms. v. useful for my quad ppro. Uncertain to the size, but with luck they'll be fairly useful.
interview
Kerneltrap.com published the interview with me today, judging by the referrer logs, looks like its quite popular. |
| Monday 24th December 2001 |
|---|
Ffffft.
Fed up with xmas already. Too much rushing around involved, which really isn't my preferred way of doing anything. Took the weekend off from doing anything too productive, after burning myself out by working really long hours all last week.
Met up with Leah yesterday, who stayed over our place last night on the first leg of her .uk tour. Hopefully she'll come back for new years too, or things could get exceedingly boring around here.
Wales.
So I'm off back to Wales in 8 hours or so for the first time since graduation. I'm hoping to find some time to go hang out with some old friends again (Maybe boxing day), but as I'm not staying in Wales for too long this time, time is somewhat limited. Looks like I'll be ending up there again in February anyway, for a multitude of reasons, including the SWLUG asking me to give a talk. Should be fun, but still need to find more reasons. Travelling a few hundred miles to speak at a LUG meet is a bit excessive. |
| Saturday 22nd December 2001 |
|---|
kernels.
Marcelo did 2.4.17 final. Merged up, and put out 2.5.1-dj5 also including Linus' 2.5.2pre1. Now up to 1.7mb, but thats the diff including pre1, so it hasn't grown that much recently.
battery.
Went into the office today to collect an extended battery for the vaio. Haven't really tested it out yet, as it was still charging last I looked. I'll play tomorrow sometime. Quite disturbed that it was already 38% charged when I removed it from the box. Will have to query how many times its been charged tomorrow. Annoyed that the ACPI code can't do that yet.
xmas.
Mailserver being down for a day meant a few hundred mails turned up out of the blue when it suddenly reappeared. I'm mostly caught up with the backlog, although I'll do a 2nd pass through tomorrow when I've had more sleep. Booked tickets today for the trip to Wales next week, and almost finished shopping for presents. Just one more to get, and I'll pick that up on Monday probably.
|
| Friday 21st December 2001 |
|---|
Merging.
Not half hour had passed after I put up dj4, when Linus put up pre1.
Merges most of the extra 2.5 bits I'd picked up, like Tronds NFS fixes
for the abuse that fsx put it through, and the bio bits I was carrying.
Current tree stands at 1.7mb. Merging up to latest wasn't difficult, and
-dj5-to-be looks ok so far.
More FS abuse.
Tonight, O_DIRECT was the victim. Made the necessary changes to fsx, and watched in horror as ext2, ext3 and reiserfs all blew up in just 6 operations. NFS lasted around 900 or so, but Trond informed me the code to allow O_DIRECT over NFS isn't yet there. This failure mode does seem a little dramatic though. Andrew Morton recently found a bunch of alignment bugs with regard to O_DIRECT, so it's extremely likely fsx is tickling these. Be interesting to rerun this test with a fixed O_DIRECT implementation.
Going places.
So tomorrow the two SuSE offices in Nurnburg are merging, which involves
network downtime for an unknown length of time. (At least, unknown to me)
Hopefully it'll all go ahead with minimum disruption, but if anyone
wants to contact me, using the codemonkey.org.uk address is probably
going to get a quicker response than the suse.de one.
Found out that I'm off to Nurnberg at the beginning of January. That should be fun. I always look forward to trips to Nurnburg, I'm not sure why, I just find it an interesting place to be. Off to the .uk office tomorrow briefly. Boredom^H^H^Hhamwood just isn't as interesting as Nurnburg.
|
| Thursday 20th December 2001 |
|---|
dj4
Put out 2.5.1dj4 as it was just in case something more experimental happens next release, and people can have something safe to fall back on whilst being up to date.pr0n.
The kerneltrap.com interview teaser claimed I worked for suse.org (Not linked, not work viewable). I've not laughed so hard for a while. Think it's time I found out whats happening to the perks of my job. Last I checked, the nekkid chicks hadn't arrived.
-dj kernel.
lwn.net mentioned it on the kernel page this week (and said they'll track it which is nice), it's meant another steady stream of hits to the server hosting the patches right now.
Talking with a few people about what will appear in -dj4. Currently it's just a resync with 2.4.17rc2, but there may be something interesting happening soon if Linus doesn't put out a pre1. |
| Wednesday 19th December 2001 |
|---|
Advogoat.
Finally had enough of the SNR on advogato taking a turn for the worse, and decided to start maintaining a diary on my own site instead. From my point of view it's not that much of a change, it's no harder or easier to post an entry here than it was there. I just wonder how popular this page will be.
(Another added advantage -- I can now find out how many people actually read my wibble). I've toyed with the idea of a script that takes the days entries from here and posts them to advogato, but to be honest, I've better things to do right now.
kernel.
The -dj patches continue to be popular judging by the downloads. -dj3 got most of the merge errors out of the way, and it's shaping up quite well. Synced up to -rc2 last night, but haven't put out a new patch as yet. Looks to be holding up so far. Backed out some silly bits that accidentally merged without testing. That'll teach me.
kerneltrap.
So I did an interview today for kerneltrap.com, which should go up sometime in the next week or so. Was quite fun to do, but it'd probably get quite boring to be doing them too often, especially if you get asked the same questions. Jeremy asked some pretty good questions, and tolerated my waffling in some answers, so hopefully people will find it interesting.
oprofile.
Some oprofile experiments today, trying to figure out why it loses samples on one box, but not on another when testing the same thing. Most unusual, and I've no explanation to offer right now.
Winding down.
Getting ready to go visit parents for a few days around xmas. Still need to buy presents, and sort out a lot of stuff. It's absolute chaos in my head right now. I'm so looking forward to doing nothing but overindulging for a few days. I'm tempted to take some time off between xmas & the end of the year too, but if I do so, I know I'll end up working through some of it to avoid falling behind with whats going on. Need to make some plans for new years eve too. I think it may involve more travel.
|
|
