<nekobot>
• xeenon (cc6d84db): Crash at API::getContentRuleListSourceFromMappedFile().…
<nekobot>
• achristensen07 (dfa11c72): Expose more serialized type info for OptionalTuple types…
ialokin has left #haiku [#haiku]
nephele_xmpp has joined #haiku
jmairboeck has joined #haiku
x10z has joined #haiku
dodo75 has quit [Quit: Vision[]: i've been blurred!]
x10z has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
akashkumar has joined #haiku
<akashkumar>
hey, if mousemoved event is performed in mediaplayer application, then how does it get propagated through the haiku codebase to take appropriate actions ?
<akashkumar>
more specifically, mouse moved on seekslider
<nephele_xmpp>
Is there any reason why BToolBar AddAction constructor only takes a BBitmap? How should I at class construction time already know what the apropriate Icon Size is? Is BToolBar not allowed to resize based on gui constraints?
<nephele_xmpp>
It looks to me that ShowImage, Mail, ScreenShot all use ComposeIconSize(22) to get a size, and then use BIconUtils:GetVectorIcon to construct a bitmap
<nephele_xmpp>
So, no scaling for BToolBar? And No C++ Type/Class for Vector Icons either? It seems to be all just const uint8 arrays, or const char* arrays
<nephele_xmpp>
err, uint8 not chat
<nephele_xmpp>
For Context: I want to add BToolBar to Icon-O-Matic, with proper scalling if possible. (And also, if possible, add Hover actions for a more WYSIWYG experience -> Previewing what is going to appear)
<Habbie>
ok good, 57820 crashed after 9 hours
<Habbie>
installing 58553. had to set haikuports to master too (did not do that before) because haiku now wants openssl3 apparently
<nephele_xmpp>
there are no seperate haikuports branches Habbie
<nekobot>
• threedeyes (832d86af): ppsspp: bump version
<Habbie>
ah, hinnant-date only uses wordexp to expand "~/Downloads", lol
x10z has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
Begasus_32 has quit [Quit: Vision[]: i've been blurred!]
HaikuUser has joined #haiku
HaikuUser has quit []
<nephele_xmpp>
waddlesplash: hmm, sounds like something I could tackle for the coding sprint
<waddlesplash>
we should probably discuss its design a bit first
<nephele_xmpp>
do you suppose you are available during that time? I've not really designed apis before, and some guidance in that department would be neat :)
<waddlesplash>
when is the sprint again?
<nephele_xmpp>
the week after 2nd february
<nephele_xmpp>
so 3-7th february
<waddlesplash>
yeah I should be
<waddlesplash>
I have some notes somewhere on a class design for this
<nephele_xmpp>
excelent
Begasus_32 has joined #haiku
<waddlesplash>
basically we should support both bitmap data and HVIF data
<waddlesplash>
and maybe also BPicture data too for completeness
<waddlesplash>
and for bitmap data we need a mechanism to specify different bitmaps for different states probably
<waddlesplash>
we already have that in a number of APIs so we should just consolidate
<waddlesplash>
the QIcon class in Qt probably has some ideas we should look at here
<nephele_xmpp>
I’d say more generally Vecor data, from HVIF, but perhaps also from more complex sources like WonderBrush
<nephele_xmpp>
though for a BIcon class HVIF may be enough
<waddlesplash>
that would mean translators. Debatable whether we should have that in this. If people want translator data into icons they probably should deal with that themselves
<waddlesplash>
or maybe we can add a method to BTranslationUtils for it
<waddlesplash>
either way BIcon should be part of libbe and not need Translation Kit
<nephele_xmpp>
I agree
<waddlesplash>
we definitely need to support specifying bitmaps at least though and not just HVIF data
<nephele_xmpp>
If we replace apis that not take BBitmap, yes. If we add it in addition then maybe not
<waddlesplash>
that way we can eventually deprecate the methods that take BBitmaps in many classes in favor of ones that only take BIcons
<nephele_xmpp>
now*
<waddlesplash>
yeah
<nephele_xmpp>
Ah, okay, if you want to deprecate them that works
<waddlesplash>
for the moment it can be a private class in lib shared while we experiment with the API
<waddlesplash>
on a related note, and probably easier for a coding sprint, I agree with PulkoMandy's idea from earlier that we should do server side rendering of HVIFs
<waddlesplash>
this would have a number of advantages including better support for remote desktop
<nephele_xmpp>
Are you talking about in app_server?
<waddlesplash>
yes
<nephele_xmpp>
okay
<waddlesplash>
basically just add some new appserver protocol commands to upload icon data and render it into a BBitmap
<nephele_xmpp>
That should be fine, at rendering time the dimensions are well known already
<PulkoMandy>
Can't BPicture be used as a protocol already for that?
<nephele_xmpp>
perhapss even cache this if we want
<waddlesplash>
though hmm, may need some tricky bits around synchronization to avoid overhead
<waddlesplash>
PulkoMandy: no, HVIF can do things BPicture cannot I think?
<PulkoMandy>
And also as an interchange format for vector graphics in genral in the translation kit
<waddlesplash>
we may need to enhance BPicture a fair bit for that
<waddlesplash>
there are already TODOs
<PulkoMandy>
Yes, maybe so, but then we can extend or replace the BPicture format with something more flexible
<waddlesplash>
right
<nephele_xmpp>
(On a somewhat unrelated sidenote: We really need a *proper* on-disk format for BMessage
<nephele_xmpp>
)
<waddlesplash>
??
<waddlesplash>
what's wrong with the one we have?
<nephele_xmpp>
It’s a memory dump of the class afaik
<waddlesplash>
uh ... no
<waddlesplash>
it's definitely not
<waddlesplash>
we already have a different format than R5 here as well and we still support reading R5 format messages too
<Begasus[m]>
nice, Lazarus 3.8 still builds with my little script :)
<waddlesplash>
nephele_xmpp: oh, I see why you think this
<waddlesplash>
yes, we directly write message headers and data into the file. However these structs are special and are a constant size across architectures
<waddlesplash>
And when reading them in we validate them and bail out if they have invalid fields
gouchi has joined #haiku
<waddlesplash>
So this is a careful tradeoff here, we avoid serialization overhead and get better performance but need careful checks on the input side
<nephele_xmpp>
Ah, okay, that does make sense
gouchi has quit []
<nephele_xmpp>
As an additional thought, a self-describing format for BMessage based settingss file could be neat. So we can build a GUI directly when a user double clicks it (to show allowed values for setitngs and such)
<nephele_xmpp>
And have a class that more easily allows to save/load such confiig, places it in the right place automatically based on app name or mimetype
arjen__ has quit [Ping timeout: 480 seconds]
HaikuUser has joined #haiku
lzrd has joined #haiku
HaikuUser has quit []
AlienSoldier has joined #haiku
<PulkoMandy>
(I don't think you'll get a good user experienoe out of it)
<PulkoMandy>
Self-generating user interfaces, the dream of every developer :)
<zdykstra>
Morning all!
<x512[m]>
CMake have self-generating GUI.
x10z has joined #haiku
<nephele_xmpp>
PulkoMandy: Well, settings in the app can stay the “pretty” settings, but we would have a better interface than Loadiing up a settings file with kottan and guesssing what an app might support :)
<waddlesplash>
it is definitely the case that we could have some generic Preferences UI that would cover a majority of cases
Begasus_32 has quit [Quit: Vision[]: i've been blurred!]
<waddlesplash>
the Tracker preferences for example are very repetitive
<waddlesplash>
one can easily imagine some C++ classes that dealt with all the boilerplate and just a handful of lines of code that describe how it should be laid out instead of "manually" doing it like it is now
<nephele_xmpp>
(as a counterpoint though, the tracker settings are indeed not a great user experience)
HaikuUser has joined #haiku
<waddlesplash>
what would be better?
<Habbie>
waddlesplash, i don't actually recall if we expected improvement, but 58553 crashes on my old AMD
HaikuUser has quit []
<nephele_xmpp>
wadlesplash: 4 tabs at the top instead of a big list on the left for example
<nephele_xmpp>
disk mounting, and wether disks appear on desktop is also confusing for new users that this has anything to do with the tracker
<nephele_xmpp>
and annoying “hidden” settings like the listview columnwwidth of Desktop influencing how long labels can be in normal Desktop icon mode
x10z has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
<waddlesplash>
Habbie: there's a patch on Gerrit that I would be interested in the behavior of still
<waddlesplash>
nephele_xmpp: absolutely disagree re. tabs. I strongly prefer the list in preferences views.
<nephele_xmpp>
Lists are good for dynamic content, this content isn’t dynamici
<Habbie>
waddlesplash, ah
<AlienSoldier>
i always wanted a SCUMM like interface for OSes and apps
qwebirc12258 has joined #haiku
<waddlesplash>
nephele_xmpp: I don't agree. Most prefs interfaces on most OSes use lists. This is a good interface and works well
<qwebirc12258>
hi, can't boot haiku via usb pendrive, i did the boot via balena etcher and last iso haiku from the official website
<qwebirc12258>
after try the boot, some console appears in white colour and its all
<qwebirc12258>
Efi here. cant change to mbr
<x512[m]>
Tabs waste limited horizontal space...
<waddlesplash>
x512[m]: you mean vertical space?
<waddlesplash>
it's nice to have the vertical space distinct from horizontal space in these views yes
<x512[m]>
Horizontal space because text may be long.
<qwebirc12258>
weird, the pendrive is not acessible on windows, i thought i could see the content..
<qwebirc12258>
i'll try boot again..
qwebirc12258 has quit []
<x512[m]>
This is why Deskbar vertical mode is nice.
<nephele_xmpp>
qwebirc: please make a picture of the “white console”
<nephele_xmpp>
that is probably the kdl (or kernel debugger)
<Habbie>
waddlesplash, which patch is that?
KapiX has joined #haiku
DogBoris has joined #haiku
<DogBoris>
still can't boot
<DogBoris>
it says Panic: didn't find a partition
<DogBoris>
i see only the logo... and in that 4 image (looks like a chip) so the white console comes
<DogBoris>
and say that.. no partition... EFI
<DogBoris>
cant change to mbr legacy
<bjorkintosh>
DogBoris: gotta start over.
<bjorkintosh>
don't skip a step.
<DogBoris>
i did do nothing..
<DogBoris>
i create the boot in windows via balena etcher...
<DogBoris>
i got the image from website. the last one..
<DogBoris>
when booting.. i see the logo and that message in white "console"
<DogBoris>
bjorkintosh: what can i do
<DogBoris>
help me
Begasus has quit [Quit: Vision[]: i've been blurred!]
<DogBoris>
unbeliveable i will not know the haiku os because i am too noob to get it working
<Begasus[m]>
DogBoris tried on another device if it acts the same?
<DogBoris>
Begasus[m]: no yet, but... i tried burn the image in different ways... rufus.. balena.. include ventoy
<DogBoris>
something change in 32bit version?maybe not i guess
<Begasus[m]>
Been there and also tried several apps to put the image to the thumbdrive
<Begasus[m]>
ow
<Begasus[m]>
you downloaded the 32bit version?
<DogBoris>
no, 64.. thinking in try the 32
<Begasus[m]>
ah ok, 32bit can be another thing, not always supported on newer devices
<DogBoris>
alright.. lets try 32.. dowloading
<Begasus[m]>
mind you, I'm no expert (and got one foot out of the door) :)
<DogBoris>
haiku-r1beta5-x86_gcc2h-anyboot.iso
<Begasus[m]>
just want to have this local package for new lazarus done and checked
<Begasus[m]>
that's the 32bit one yes
<DogBoris>
Begasus[m]: any help is usefull, are you running the 32 or 64?
<Begasus[m]>
on this install 64bit, 32bit not supported
<Begasus[m]>
on another laptop I have both running
<DogBoris>
mmm.. i just wanna run in live usb
<DogBoris>
downloaded enve i have no hope i will try
<Habbie>
is 'jam -a haiku.hpkg @nightly-raw' enough?
DogBoris has joined #haiku
<DogBoris>
Begasus[m]: didn't work, but..
<DogBoris>
the 64bit version worked in an older laptop
<waddlesplash>
Habbie: other way around usually, "jam -q -jN @nightly-raw haiku.hpkg"
<Habbie>
alright
<DogBoris>
Begasus[m]: my laptop is not compatible
* jn
was summoned by a call to make -jN
<waddlesplash>
LOL
<Habbie>
haha
<jn>
ah wait, it's jam, not make :)
<waddlesplash>
nipos may also want to test with that patch on his Bulldozer hardware
<waddlesplash>
I think he was getting crashes much more easily than you and dovsienko
<waddlesplash>
dovsienko: btw I did fix that KDL
<DogBoris>
my main laptop doesn't run haiku for some reason.. but other very very old does
<DogBoris>
and my main is not new.. it's kind old too lol
<Habbie>
don't know how to make @nightly-raw
<Habbie>
but, it's building
<Habbie>
~/haiku> HAIKU_REVISION=gerrit8684 jam -q @nightly-raw haiku.hpkg
<waddlesplash>
you must be on an older tree then, or you checked out just this version
<waddlesplash>
I fixed the "don't know how to make" not long ago
<Habbie>
i checked out the actual commit
<Habbie>
ahh
<waddlesplash>
ah
<waddlesplash>
I could rebase it on master branch
<Habbie>
i know how to rebase etc.
<Habbie>
ok
<Habbie>
i was undecided, will do
<waddlesplash>
well for me it's two clicks on Gerrit
<waddlesplash>
done
<Habbie>
this was my first time using gerrit
<Habbie>
and i just took the first git command from the download panel
<waddlesplash>
well the change on Gerrit is now rebased so if you check it out again it'll be against HEAD
<augiedoggie>
DogBoris: maybe you should try a nightly build of Haiku instead of the beta
<Habbie>
already rebased here
<waddlesplash>
ok
<Habbie>
-that- "don't know how to make" is gone
<Habbie>
libsupc++ libgcc and some crt stuff still says something like that
<DogBoris>
augiedoggie: yes..
<waddlesplash>
Habbie: do you have haiku_devel installed?
<nipos>
waddlesplash: Thanks for the hint, will try installing that patch and see if compile tasks with more than one core still reliably crash the PC
<waddlesplash>
you may need to rerun ./configure to pick up new paths if you updated GCC
<Habbie>
i do
<waddlesplash>
nipos: excellent, thanks
<Habbie>
oh i bet
<Habbie>
ok
<waddlesplash>
nipos: also, you may want to retest just with nightly master, because some recent changes to GCC and vfork may also affect this
<Habbie>
also never used jam outside of haiku
<DogBoris>
augiedoggie: ahh that little images, i already tried.. and nothing :S thank you for the tip
<waddlesplash>
nipos: so make sure it still does crash regularly before applying the patch
<Habbie>
./configure reduced weird messages by 80%
<Habbie>
i'm going to let it build now
<Habbie>
don't know how to make libsupc++.a
<Habbie>
assuming this is harmless
<nipos>
Is updating to the latest nightly from repo enough or are the other fixes not yet packaged?
<waddlesplash>
Habbie: probably not. do you have gcc_syslibs_devel and haiku_devel installed
<waddlesplash>
nipos: all relevant fixes are packaged, some are in the GCC package though
KapiX has quit [Quit: KapiX]
<waddlesplash>
so make sure you full-sync or whatever and get the latest GCC package
<augiedoggie>
DogBoris: i would see if there are any options you can change in the bios, make sure secure boot is off and try turning the CSM options on or off
HaikuUser has joined #haiku
* Habbie
adds gcc_syslibs_devel
HaikuUser2 has joined #haiku
<HaikuUser2>
oi
<waddlesplash>
hello
<HaikuUser>
how u doing
<DogBoris>
augiedoggie: there is the problem, my bios doesn't have cms.... it's asus bios version 303
<Habbie>
csm
<Habbie>
it might call it legacy
<DogBoris>
no legacy too
HaikuUser3 has joined #haiku
<DogBoris>
secure and fastboot disable
<DogBoris>
stup** asus bios
vdamewood has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
<HaikuUser>
oi
<HaikuUser3>
oi
<DogBoris>
oi
<nipos>
btw, I noticed that I can't update from the repo when currently running a self-compiled system
<HaikuUser3>
como muda o nome
<DogBoris>
digite /nickname NovoNome
<waddlesplash>
nipos: you must not be building with a profile
<waddlesplash>
no profile = no SSL
<waddlesplash>
build with @nightly-raw specified to get SSL
<nipos>
Ahhh, okay, thanks
<HaikuUser3>
oi
<HaikuUser3>
deu errado
<HaikuUser3>
calmai
<nipos>
In WebPositive network still works, but maybe that's independent.
DogBoris is now known as DogBoriss
<waddlesplash>
nipos: it's CURL so yes
<DogBoriss>
HaikuUser3: desculpe é /nick NovoNOme
<nipos>
Doesn't really matter, I just revert to a official version before updating. But now I know for the next time
<HaikuUser3>
putz
<DogBoriss>
HaikuUser3: vai logo baralho
<HaikuUser3>
eu fiz issoooo
<Habbie>
ok, i ended up git cleaning with prejudice
GREGSGY has joined #haiku
<Habbie>
looks happier now
HaikuUser3 is now known as luscao
<luscao>
AEEE
<DogBoriss>
boa locão
<luscao>
KKKKKKK
<GREGSGY>
OI
<luscao>
oi desgyy
<DogBoriss>
dae
<dovsienko>
waddlesplash: do you mean #19360: VM page fault related to AVLTreeBase RAMFS?
<GREGSGY>
mano eu realmente instalei isso em hardware real
<luscao>
insano
<DogBoriss>
GREGSGY: putz veio... tente o biglinux mto melhor
<waddlesplash>
dovsienko: yes, I added a new ticket 19367 for the followup KDL
<waddlesplash>
and fixed that
<luscao>
nunca ouvi falar de biglinux
<luscao>
eh como?
<GREGSGY>
linux soq 2 tera
<luscao>
jesus amado
<DogBoriss>
luscao: melhor linux do mundo.. olha no google e entra no site po
<DogBoriss>
baixa e instala
DogBoriss has quit [Quit: Page closed]
<luscao>
ata entendi
<luscao>
talvez eu teste linux dps
<luscao>
ce tem discord?
GREGSGY has quit []
<luscao>
putz
luscao has quit [Quit: Vision[]: i've been blurred!]
HaikuUser2 has quit [Quit: Vision[]: i've been blurred!]
HaikuUser has quit [Quit: Vision[]: i've been blurred!]
B2IA has quit [Quit: Vision[]: i've been blurred!]
B2IA has joined #haiku
DogBoris has joined #haiku
<DogBoris>
voltei
<DogBoris>
worked in nightly
<DogBoris>
augiedoggie: thank you . now it worked
<augiedoggie>
nice
<DogBoris>
augiedoggie: you are the man
<DogBoris>
augiedoggie: i oppened the webpositive to use webirc.. do you know where is the irc app in this o.s
<DogBoris>
first time here
<augiedoggie>
it's called Vision
<DogBoris>
alright. thank you
<augiedoggie>
there are others you can install but Vision comes with Haiku by default
HaikuUser has joined #haiku
HaikuUser is now known as DogBoriss
<dovsienko>
waddlesplash: that's good to know, thank you
<DogBoriss>
yesss
<DogBoriss>
its great
DogBoris has quit [Quit: Page closed]
nephele_xmpp has left #haiku [Error from remote client]
<DogBoriss>
mmm no minimize
<augiedoggie>
double click the title bar or use the keyboard shortcut
<DogBoriss>
great.. just saw the side bar in the right
<nekobot>
[haiku/haiku] 5bfe88b36b3b - kernel/device_manager: Increase virtual_end of IOCache.
<nekobot>
[haiku/haiku] 71c04db60a85 - kernel/vm: Reinstate assertion in InsertPage.
<Habbie>
Stream.cpp:(.text+0x251): undefined reference to `__divdi3'
<Habbie>
bunch of things like this in the linking of bfs/bios_ia32/boot_bfs.a
<Habbie>
ring a bell for anyone?
<waddlesplash>
you can't build the 32-bit BIOS loader on 64-bit without the cross-tools
<Habbie>
ah
<waddlesplash>
open build/jam/packages/Haiku
<waddlesplash>
and comment out line 21
<waddlesplash>
by putting a # at the front of it
<Habbie>
AddFilesToPackage
<waddlesplash>
yes
<Habbie>
wonderful
<nipos>
It would be nice if there was a simple config option to not build the BIOS loader. I already know and use that hack, but messing with files that are not in .gitignore and you don't want your changes to be committed doesn't go well long-term
<waddlesplash>
really, we shouldn't include the bootloaders in haiku.hpkg anyway
<waddlesplash>
we should ship them in haiku_loader.hpkg
<waddlesplash>
however that was not done at the time because the BIOS loader MUST come first and that wasn't happening reliably when adding more than one file to the pacakge
<waddlesplash>
I have it on my TODO list to investigate this but have not yet done so
<waddlesplash>
(it's been on my TODO list for years lol)
<jmairboeck>
isn't haiku_loader.hpkg uncompressed also? or could that be specified per file?
<waddlesplash>
it's uncompressed, yes
<waddlesplash>
and no, can only be specified for a whole package
<nipos>
System updated (without the patch yet) and big build running, let's see how long that lasts
<waddlesplash>
OK, very good
<DogBoriss>
augiedoggie where can i found the instaler for apps do you know
<nipos>
Ladybird is usually big enough to hit the bug at least once, but it can take a half hour sometimes
<Habbie>
nipos, so 58553?
<nipos>
Yes
<Habbie>
alright. i crashed 57820 and 58553 today on my AMD, and am building master+the patch now :)
<augiedoggie>
there is also a local copy of the userguide installed with a link on the desktop you can read
<DogBoriss>
augiedoggie i will try some game thanks
<nipos>
I crashed a not much older hrev a few days ago while I was building haiku.hpkg with -j1 and thought building BeAIM (which is very small) at the same time also with -j1 can't hurt. I was wrong.
<nekobot>
[haiku/haiku] ab25a423870d - kernel/elf: Fix handling of the commpage in UserSymbolLookup.
<nekobot>
[haiku/haiku] 2cee2502c5d0 - kernel/vm: Unlock lower caches before allocating clean pages on page fault.
<waddlesplash>
nipos: wait, what kind of state is that? does anything appear on serial?
<waddlesplash>
most instances of this bug it's just a triple fault yeah
<DogBoriss>
augiedoggie very light os.. omg..
<nipos>
I don't have serial here. I only have serial on the server, but that's Intel and doesn't crash
<Habbie>
i ordered the parts to get serial from aliexpress but never got around to putting them together
<Habbie>
but mine reboots so probably not interesting yet
<Habbie>
(the patched version is still building)
<nipos>
Habbie: I have two different AMD machines from different years, both crashing with that weird stuff on the screen, sitting there until I pull the plug
<Habbie>
oh
<DogBoriss>
augiedoggie oh failed to download libsdl2
<Habbie>
well that is, as waddlesplash also noted, very interesting
<DogBoriss>
tried install 0 a.d game
<Habbie>
as it seems like a very different failure case from mine
<nipos>
This is a AMD Athlon II, so it's not even Bulldozer (someone told me?), but it behaves exactly the same as the AMD A4 which is Bulldozer
<Habbie>
yeah, mine is not Jaguar while the ticket says it's a Jaguar problem, i think
<waddlesplash>
nobody has ever pinpointed on what generation the problems really begin I think
<waddlesplash>
only that they don't happen on Ryzen and up
<Habbie>
yeah i believe that
<Habbie>
gerrit gives me a popup
<Habbie>
'A newer patch set has been uploaded'
<DogBoriss>
kd os br
vezhlys has joined #haiku
vezhlys has quit [Remote host closed the connection]
<waddlesplash>
Habbie: that's the rebased patchset
<Habbie>
oh that's just the rebase from earlier
<waddlesplash>
yes
<Habbie>
i guess it's not very live
<Habbie>
on a sidenote, https://voorkant.org/ runs fine on Haiku, except for the bits that rely on the hinnant-date library
<nipos>
Now building the patch. It needs to update 3550 targets, which is going to take a while on a single core
<Habbie>
i am also building without -j
<Habbie>
although my haiku is running in SMP mode
<Habbie>
it tends to last between 2 and 20 hours so i might just make it in one go
<nipos>
My Haiku is also running in SMP mode. Has never been a problem, as long as I don't try to compile stuff with multi-core or compile multiple projects at the same time
<Habbie>
ack
<nipos>
Other heavy workload is a lot less likely to trigger the bug
<waddlesplash>
based on currently available data it seems to be fork-exec related
<Habbie>
i'm on the 800th target, and i bet it has to do more than 3550, as i started from a clean dir
<waddlesplash>
the obvious potential culprit is page faults
<waddlesplash>
which is what this patch aims at improving
<waddlesplash>
returning from a syscall into a page fault immediately is a weird case, I would not be surprised if other OSes avoid doing that entirely, and this patch should too
<Habbie>
ah, right
<waddlesplash>
during a standard compile job that does lots of fork-exec, you get multiple of those per second
<DogBoriss>
in live mode i can`t install some games like wideland and 0ad. of course, maybe i need to install the haiku in usb, will it create a boot in the usb pendrive after install
<waddlesplash>
recent changes I've been trying to avoid the number of fork-execs in build jobs; the new GCC version now uses posix_spawn and cuts them out in compiler processes
<waddlesplash>
make and ninja however use more posix spawn features than GCC so that still fork-execs
<waddlesplash>
we should implement a real spawn() syscall to avoid that too, eventually
<DogBoriss>
ok i`ll try it right now
<Habbie>
waddlesplash, what's spawn()? i know posix_spawn
DogBoriss has quit [Quit: Vision[]: i've been blurred!]
<waddlesplash>
I mean an internal spawn syscall
<Habbie>
ahh
<waddlesplash>
the BSDs do this I believe
<Habbie>
for libroot to use where appropriate
<waddlesplash>
yes
<Habbie>
got it
<waddlesplash>
at some point here, once I finish this round of VM related improvements, another benchmark against Linux and also the BSDs re. compile perf may be nice
<waddlesplash>
actually against macOS would be the really interesting test
<Habbie>
why macos specifically?
<Habbie>
nipos, ah, 13480 targets for me. at 900 now
<waddlesplash>
because if we're slower than Linux people will say "who cares", but if we're slower than Linux but faster than macOS, people will take note lol
<Habbie>
oh, macos is slow?
<nipos>
I did some work on Haiku recently, so most targets here are still up to date
<waddlesplash>
because unlike with the BSDs you can't claim macOS isn't used by people lol
<waddlesplash>
Habbie: it's not too slow, but, it's definitely slower than Linux in terms of raw performance, most things are
<waddlesplash>
the question is how much slower
<Habbie>
ack
B2IA has quit [Quit: Vision[]: i've been blurred!]
<waddlesplash>
we are slowly improving here. Last summer before I started down this path, kallisti tried a build and got around 12 minutes build time for the whole tree (-j16), vs. 5m30s on Linux (same machine, however, not the same underlying SSD so it's not quite a fair comparison)
<waddlesplash>
after last summer's fixes, we got down to around 9m I think
<Habbie>
very nice
<waddlesplash>
another 20% and we will be "within striking distance"
B2IA has joined #haiku
<Habbie>
and all the pain is in kernel land, i presume, including transitions to/from it
<Habbie>
the cost of context switches, the performance of fs calls, etc.
<waddlesplash>
mostly, however we need to replace userland malloc
<Habbie>
ah!
<waddlesplash>
that will get us another few % I think
<Habbie>
which one is it now?
<waddlesplash>
a very old version of hoard
nephele_xmpp has joined #haiku
<waddlesplash>
I want to try and use OpenBSD malloc instead
<waddlesplash>
however with some additional patches and a global cache added
<Habbie>
ohhh <3
<Habbie>
the openbsd malloc author is my coworker
<waddlesplash>
I actually saw that lol
<Habbie>
:)
<Habbie>
he'll be happy to answer questions if you have them, i bet
<Habbie>
are you coming to FOSDEM?
<waddlesplash>
no
<Habbie>
ah too bad
<nephele_xmpp>
waddlessplash: maybe the haiku Mastadon account can send a message about the fosdem stand
<waddlesplash>
I am in the USA, that would be a long (and expensive) plane flight for not much time in Europe
<Habbie>
ah right
<waddlesplash>
nephele_xmpp: or maybe one of the people who is going to be there can post and the mastodon can repost :)
<nephele_xmpp>
I don’t have a mastadon account currently. I think pulkomandy and mmu_man both have one, so if they want to :)
<Habbie>
mmu_man certainly does
<Habbie>
ah yep, pulkomandy too
<Habbie>
i could save myself a day by booting the -other- box into Haiku and compiling there
<Habbie>
but that's a day of just waiting and doing this is actual work :)
<nekobot>
[haiku/haiku] 3b22516ed242 - drivers/network/pcnet: Allocate a small softc on the stack in probe().
x10z has joined #haiku
DogBoris has joined #haiku
<DogBoris>
Habbie finally.. it was a fight.. but worked
<DogBoris>
augiedoggie installed on pendrive.. running it..
<waddlesplash>
nipos: I would be interested if you have some benchmark numbers for differences. Even just at -j1, just run the same exact build of files 2-3x on each build, and see what happens
<waddlesplash>
I just tested with HaikuPorter here, and a full rebuild of the repository information (thousands and thousands of forks), goes from 47.7s -> 47.0s sys time
<waddlesplash>
so, that is an improvement, but not really too much of one
<waddlesplash>
less than 2%
<Habbie>
DogBoris, yay
<DogBoris>
Habbie laptop keyboard is not ok.. alt gr key not working .. but ok
<nipos>
I have shut down the computer for now, I can try that tomorrow
<waddlesplash>
OK
DogBoris has quit []
<Habbie>
4700 of 13480 targets
mattlacey has quit [Ping timeout: 480 seconds]
Skipp_OSX has quit [Remote host closed the connection]
nexus6 has joined #haiku
<Habbie>
ah, hangman is built. very important
<nephele_xmpp>
:)
x10z has quit [Quit: My Mac has gone to sleep. ZZZzzz…]