DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.tw/irclogs/ Connected with DDNet's IRC channel, Matrix room and GitHub repositories — IRC: #ddnet on Quakenet | Matrix: #ddnet-developer:matrix.org GitHub: https://github.com/ddnet

Between 2017-10-11 00:00:00Z and 2017-10-12 00:00:00Z

[ddnet-maps:master] 1 new commit

4f274f8 M Insane - def-

<Nimda> Offpanic by Themix just released on Brutal at 2017-10-11 19:52

<laxa> !ddnet

18:50

<laxa> !ddnetpeak

18:50

<Nimda> Current players on DDNet : 368

18:50

<Nimda> Current DDNet peak : 810 users online at 2015-04-26 20:11:01

https://github.com/ddnet/ddnet/pull/880/commits Now i've failed 2 out of 3 what the heck

Jekyll clip by hpjohn · Pull Request #880 · ddnet/ddnet

Create modifiable group clipping Add a list of clipping areas to map file using new editor window (enable cl_editorcliptrigger 1) -These consist of tune zone, trigger id, clip x,y,w&h, disableclipp...

std:: ?

19:06

or maybe you are usign function which are not included

It compiles locally

soo maybe, you have newer version

19:07

don't think so

can you sactually see where the errors are occurring

in editor.cpp

19:09

there's no #include <string.h> but

19:09

im not C dev, so maybe it should inherit from somewhere else

Where have i added anything from there in editor.cpp

19:10

strcmp?

yes

hmmmmmmmmmm

and memset in

19:11

anyway

19:11

https://circleci.com/gh/ddnet/ddnet/1790

19:11

just go there

19:11

and expand bam release error log

@Dr. Jekyll you shouldn't use string.h in Teeworlds code, use functions from system.h instead

#include <base/system.h> ?

str_comp replaces strcmp, memset(x, 0, y) is replaced by mem_zero(x, y)

both memset and strcmp are used im other places not by me

only in code that doesn't adhere to the common style – we'd accept PRs to fix that

19:40

where?

server.cpp 78

19:41

478

introduced by GreYFoX

19:42

that's been a while. he didn't care too much about code style

file_score.cpp 166

19:42

ill just use your qrapper then

19:42

wrapper

memset is used 5 times in the code, mem_zero 162 times

19:45

strcmp 3, str_comp 319

]

19:45

:]

I found interesting site http://styleguides.io/

generally: work with the style of the surrounding cod

19:54

e

doppel peace an @heinrich5991

@heinrich5991 each ASYNCIO gets it's own thread?

yes

22:07

did it like that because they can block for arbitrary amount of time and you can't check whether they'll block

22:07

(no nonblocking mode for files)

they?

the disk IO operations

22:08

or at least that's what deen told me

can't you return if block on disk IO?

no

22:08

I checked that

hmm stupid assumption they made

22:08

all disk io doesn't have to be fast

anyway, it shouldn't be a lot of overhead. the threads take up no CPU until woken up by more input

looks like a perfect place to use condition variables

shame they don't work on windows xp

yes :/

just use c++11 ?

not in system.c you don't

(edited)

22:11

@heinrich5991 if you feel up to it maybe a good time to start system.cpp

1

mh. I was happy that I could do without

You know me I'm all for keeping to C

because this makes it more likely that teehistorian is merged into other mods

22:14

I could check out the condition-variable implementation of gcc or clang on windows

I think they use semaphores

22:15

https://www.microsoft.com/en-us/research/wp-content/uploads/2004/12/ImplementingCVs.pdf

if you refuse to switch to system.cpp what about adding CVs using c++11 here: https://github.com/ddnet/ddnet/blob/master/src/base/tl/threading.h ?

ddnet - DDraceNetwork, a mod of Teeworlds

@heinrich5991 what is the two buffers for?

@HMH feel free to add system.cpp once you need it, I won't block it

22:19

@Learath2 when the thread that writes to the queue needs to create a bigger buffer, it can't delete the old buffer right away

22:19

because the thread reading from the queue might still have pointers into the buffer

wait wait wait there is a thread that writes to the queue?

the main thread

oh

mh?

22:20

I mean there should be someone writing, right? ^^

I thought you had another thread that was coordinating the writing threads, was like how did I miss that one

nono

22:21

the "writer" is the main thread that wants to output stuff into files, the "reader" is the dedicated thread that forwards the things into a file

22:22

(in my terminology)

is this even safe? you write to buf1 from one thread while the other is possibly io_writeing what's in buf1

I'm writing disjoint areas of the same array

shouldn't atleast the length associated to the buffer be lock protected?

yes, it is

22:23

well

22:23

so

22:23

the lock is held everytime someone does something with the array, except while io_writeing the stuff

22:24

(because that would make the extra thread useless)

22:25

the queue reader only updates its (lock-protected) read pointer when it is done io_writeing the contents into the file

22:25

the queue writer only writes until the buffer is full, and at that point, it increases the buffer size by switchting to a new buffer

side-line: all the extra braces makes this really hard to read for me

extra braces?

the ones in buffer_len e.g.

ah

22:26

yea, always writing braces doesn't look too well in our braces style

so used to omiting them for oneliners it looks disjointed

I decided to just always write them and never look for braces errors anymore

22:28

it works better in the

22:28

if(abc) {
    return 1;
} else {
    foobar();
}

22:28

style

eeh, don't think I ever made a brace error

22:31

been a while since I've seen a brace mistake in ##c either

don't really want to get sidetracked into this discussion unless you really want to

anyways, what is "Don't allow full queue to distinguish between empty and full queue." supposed to mean?

22:31

we can get sidetracked about braces later

22:32

don't fill the queue completely – full queue looks the same as empty queue, read_pos == write_pos

how does that work? write_pos wraps back?

yes

22:33

#########
  W   R

(edited)

22:34

this means that the queue is filled from the R, wrapping around to the W

why is it filled from R? read_pos sounds like where you are io_writing from

yes

22:35

I'm io_writeing from R to W

22:35

the reader is behind the writer

22:35

it still has to read all the stuff the writer has written

In your picture about R is ahead of W, how does that happen?

when the buffer has wrapped around

22:36

the writer wrote 6 bytes, the reader read 6 bytes, the writer writes 6 bytes

so you are kinda using it like a circular buffer?

yes

22:36

except when it gets filled, then I create a larger buffer

do you have a max size somewhere?

no

22:39

I hopefully crash on allocation failure

22:39

but not checked, as indicated in the comment

Is it a good idea to crash on a logging failure?

it is a good idea to crash on teehistorian failure

22:40

logging -- maybe not

22:40

mhh

maybe a forced blocking flush for teehistorian and a skip for logging?

crashing vs blocking probably doesn't matter too much, in any case the race is fucked up

22:42

I could perhaps introduce a max size

22:42

for the queue

well would blocking kill everyones connection?

you have blocked long enough to have filled the teehistorian buffer, which is probably at least multiple 10MBs

22:43

that's probably as much as such a server produces in a day

22:44

I'll see if I can introduce a max size

22:44

(that will need another notification in the other direction, to let the reading thread wake up the writing thread)

the buffer is somehow full, you blocking flush in tick, no more data is generated by teehistorian, how bad can io be honestly

22:45

frozen server is better then crashed server no?

yea

actually if it never unfreezes it's worse then crashing

true

is there a reason you don't do a realloc?

yes. I still need the old buffer for the reader

22:48

it might currently do a io_write with the old buffer

22:49

oh, another counter-argument against checking for OOM: none of the other code does it, so that will crash when we check for it (?)

22:49

we run our servers on linux with overcommit, so malloc failure isn't really a thing there

true

22:52

@heinrich5991 so buf1 and buf2 are just convenience pointers right?

yes

22:52

I just wrote that code to reuse it somewhere else

22:52

I think I call that function in two places

you use old_buffer to make sure the buffer doesn't get free'd while io_writeing right?

yes

Q:resizing the buffer "straightens out" the buffer, if io_write was using the old_buffer how do you sync up the read_pos and write_pos to the new buffer? You also don't seem to check the return of io_write is it right to assume the entire buffer was written?

I sync the read_pos and write_pos to the new buffer by just setting read_pos to 0

23:02

the reading thread will always just add the length of the buffer it just wrote to read_pos after it is done

23:02

(while the stuff is locked, so the writing thread isn't interfering)

23:04

unless an error occurs, I think fwrite always writes the full number of bytes

what is refcount for?

depending on what exits first, the reading thread or the writing thread

23:08

the aio structure should only be freed after both exited

you decrement refcount at only one function and only call that function once per aio no?

I call that function in two places

23:11

once from the reading thread, and once from the writing thread

23:11

(just search for the function name)

oh missed that one, sry gh website doesn't like my searches

23:11

sometimes it decides not to show stuff that I can't currently see

mh ~~

you know what would be cute? a goto for cleanup in async_new

thought so too

<3 goto

but then I remembered our lack of gotos

we only use it in the sql stuff and that stuff all looks horrible anyways

23:21

especially the save/load thing, awful style, weird use of return codes

@Learath2 any more questions? probably going to bed soon

nope, submitted my first review

ty, will check that out tomorrow

I should also get some sleep, gn8

nite

Exported 186 message(s)