freenode/#sicl - IRC Chatlog
Search
8:26:30
beach
I can not understand why an application leaking memory would make the computer grind to a sudden halt.
8:26:59
beach
I would think it would start by paging, getting slower, and then, when the application ran out of (virtual) memory, the system would kill it.
8:27:24
no-defun-allowed
i've had embarrasing pauses occur when i assume something is going way too hard on memory or processing time
8:29:35
no-defun-allowed
with riot.im, youtube and stackoverflow open my firefox is using 340 to 500mb, presumably i saw some major gcs
8:33:54
heisig
beach: I am sorry to hear that your computer is still giving you trouble. I would guess your problem is either hardware related, or a problem of the GPU driver. The other software components of a Linux system are usually quite stable.
8:35:59
beach
Sometimes it even crashes 5 minutes after I reboot, and I have not had time to do anything then.
8:39:13
shka_
beach: well, that would not help with faulty power supply, but hardware is supposed to be fine
8:39:38
no-defun-allowed
[here's my leak tester if you still think that's a problem](https://gitlab.com/snippets/1765899)
8:40:51
no-defun-allowed
mine gets killed at around 8300MiB. i have 8gb memory and 2gb swap with firefox, emacs and xfce4-terminal open.
8:41:17
beach
I need to try to get some real work done today. It is 10:40, and I haven't done anything (other than house chores and shopping for food).
8:41:35
no-defun-allowed
stuff gets very slow around 6500mb, X is laggy and my music starts to chop up
8:43:30
no-defun-allowed
mine will freeze, audio will loop the last second or so and the screen will become garbage or blank out
8:45:53
beach
Ah, that's different from mine. Here, the screen looks fine, but I can do absolutely nothing.
8:57:50
no-defun-allowed
so when (make-instance 'boot) prints out the phases it's in, are those just compiling the same code with the last phase's system?
8:58:36
beach
Often it's the same code, yes, but in different environments, with different definitions of macros and such.
8:59:00
beach
Like, typically DEFCLASS, DEFGENERIC, DEFMETHOD mean different things in different environments.
9:01:15
beach
That's the very idea behind this bootstrapping technique. Maybe something else would be possible, but that's what I came up with.
9:02:28
beach
Yes, given the constraints (no playing with package names like SBCL does), first-class global environments seem to be the solution.
9:07:31
beach
Besides, I want first-class global environments in CLOSOS anyway. Even in just SICL, I may use one environment for system stuff like the compiler and assembler and one for user stuff that has only the entry point to the compiler. This way, SICL would be much less vulnerable to installing some malignant code.
9:11:55
beach
Absolutely! Playing with new features of the system without generating an entirely new system every time.
9:14:47
no-defun-allowed
where could i find the initialization code for instances of sicl-extrinsic-environment:environment?
9:17:04
beach
The non-minimal environment contains a lot of stuff and it got confusing to debug the bootstrapping code.
9:17:24
beach
With the minimal environment, I get an error from the compiler whenever something is missing.
11:50:37
no-defun-allowed
It's low hanging fruit, but I think mentioning that you do not necessarily need 10gb of heap might encourage people to experiment and test out sicl more.
11:51:11
beach
no-defun-allowed: I'll wait with that until there is something worth experimenting with.
11:51:55
no-defun-allowed
I don't know much about compilers and much of what SICL covers, so I can only comment on the boring stuff.
11:53:19
beach
jcowan: When I give talks to industry, I actually give an alternative to using a programming language with an independent standard.
11:53:28
beach
jcowan: If they choose a different language and I see a proper risk analysis that includes the cost of maintaining an implementation of that language (in case existing implementation are abandoned, or the benevolent dictator decides to alter it in incompatible ways), then that means that they have planned for this eventuality and that they assume the risk.
11:54:27
beach
jcowan: And they just make very risky decisions anyway; decisions that might destroy the entire company if things go wrong.
11:56:20
jcowan
beach: In a climate in which people come up to you and stuff your pockets full of thousand-dollar bills merely for walking down the right street, such risk-taking is entirely rational.
11:57:00
no-defun-allowed
That's true, you could probably get venture funding if you sold toaster shoes.
11:57:22
no-defun-allowed
That doesn't mean you should expect it to continue that way, or that you should rely on it.
11:58:27
jcowan
Why not? The vast gains in productivity over the last 30 years have gone entirely to the rich, and even they can only buy so many yachts with it. The rest has to be recycled somehow.
11:59:43
jcowan
The need is to get something working quickly, and for that you don't want any more architecture than is required to construct a termite mound.
12:01:52
jcowan
Believe me, I hate the corruption of programming-as-a-discipline as much as you do, perhaps more.
12:02:21
jcowan
It's still important to be right, it's just not important to have already been right.
12:02:47
no-defun-allowed
#sicl trashtalk actually has valid arguments with inciteful points, unlike certain cafes where saying people who are there should be shot is okay.
12:04:18
no-defun-allowed
Anyways, erm, people still get money for bad ideas cause there's a way to monetise them no one else expected. Selling user data is a great one.
12:05:18
no-defun-allowed
Anyways again, I look forward to pressing M-. a lot over the weekend on SICL.
13:23:30
beach
frodef: I don't remember the details, but it was clear from when you described your objectives, they were very different from what you see in that document.
14:24:11
beach
OK, it is getting close to the end of the day and the end of the week. Soon, fatigue will prevent me from doing any more serious work today.
14:24:17
beach
I made some progress on bootstrapping, but there is not much to show for it. I am still trying to eliminate a special version of ADD-METHOD in favor of the production version in phase 3.
14:24:25
beach
When I do that, the production version invalidates the discriminating function (as it should).
14:24:26
beach
But then, when a new discriminating function must be calculated, I have some mysterious classes appear. I am getting closer to figuring out why, though.