<waddlesplash>
yes, but they at least have all the BSD patches intree
<Nephele>
and it is also not toooo sure they will even survive with the recent antitrust suite
<waddlesplash>
nah, it will survive, one way or another
<Nephele>
maybe it will become a proper community project again :P
<waddlesplash>
if Mozilla dies, Firefox will get reborn yet again as a community project
<waddlesplash>
yeah
<waddlesplash>
maybe it will become possible to donate to Firefox development again...
<Nephele>
maybe that would not be too bad an outcome
<Nephele>
no crypto scams, forced advertising... an open webrowser xD
<Nephele>
i was quite amused at jwz insulting mozilla on twitter for accepting crypto donations
<Nephele>
waddlesplash: speaking of browsers, i think the netsurf native chrome could use some love
<waddlesplash>
everything around here could "use some love"
<waddlesplash>
we are building an OS with pennies and sweat
<Nephele>
Me too! :D
<waddlesplash>
lol
<Nephele>
well, curently you get the most pennies
<waddlesplash>
yes, I do
<Nephele>
:)
<waddlesplash>
but they're still pennies compared to what OS devs at MS or Apple get paid :P
<Nephele>
clearly you chose the wrong profession
<waddlesplash>
I think the Haiku community would not agree with you :-p
<Nephele>
I actually looked at how safari performs on apples new platform two days ago, they redesigned the scrollbar in a good way. but still autohide it
<waddlesplash>
but yes, it is very impressive how much we can do with our "pennies and sweat"
<Skipp_OSX>
more than most do
<Nephele>
indeed
<waddlesplash>
the pittance of "man-hours" poured into Haiku compared to Linux, and yet for a variety of usecases we are a real competitor to desktop Linux
<waddlesplash>
and the gap is, slowly, closing
<Nephele>
But i think this is because we talk to each other, and don't develop 20 alternatives and let the user "choose"
<Skipp_OSX>
well I'm happy, assuming we get the download in Web+ bug solved, and I'm pretty sure it already is.
<Nephele>
it is for 64bit yeah
<Nephele>
32bit apparently doesn't build but PulkoMandy wanted to investigate that
<Nephele>
but ran into trouble with git
<Nephele>
I want to properly set up my webkit build server with haiku again :(
<Nephele>
waddlesplash: i sure wish it would boot with mbr for framebuffer
<waddlesplash>
?
<Nephele>
but barring that, maybe somebody has a recomendation for "some" graphics card pci-e i can use for native resolution
<Nephele>
well, mbr boot is broken on that machine
<Nephele>
and efi gives me a trashy resolution
<waddlesplash>
what resolutions does the bootloader offer?
<waddlesplash>
EFI can change resolutions, just not after boot services exited
<waddlesplash>
so, check what the bootloader has in the modelist
<Nephele>
uhh, i think a trashy one only
<waddlesplash>
just one? really?
<waddlesplash>
weird
<Nephele>
i can check though
<waddlesplash>
you can also try the VESA settings file
<Nephele>
one moment
<waddlesplash>
the EFI loader does read this and tries to set that mode
<Nephele>
but this efi has given me lots of trouble in the past
<waddlesplash>
ah
<Nephele>
if you disable cms you won't even get video output for some reason
<waddlesplash>
Skipp_OSX: yeah, at this point, we need WebKit fix, and then Userguide sync, and that's it
<waddlesplash>
so, I should work on the release notes
<Skipp_OSX>
I opened up the contract reports and monthly updates to get a gist... but there's a lot
<Skipp_OSX>
I have updated release notes a bit on trac... maybe you see my 256MB min change and will be mad.
<Nephele>
huh? wasn't it more in the previous release?
<Nephele>
364mbß
<Nephele>
?
<Skipp_OSX>
yeah that's why he's be mad
<Nephele>
well if you tested it and it works that should be fine
<Skipp_OSX>
yep, and it didn't before
<Skipp_OSX>
so something got improved
<Nephele>
on a related note, i am happy of not having any recent packagefs sucks discussions in the forum
<Skipp_OSX>
who needs package management when you can just unzip a directory of files on top of home?
<waddlesplash>
shh... don't jinx it
<waddlesplash>
:P
<waddlesplash>
nephele: I have a WIP change to reduce packagefs memory usage by 40 bytes per file visible on the filesystem
<waddlesplash>
didn't get it in to beta5 because it's a bit dangerous and needs more work
<Nephele>
that's ridicilous, just delete some files >:( /s
<Skipp_OSX>
40 bytes, what is this bloatware?
<Skipp_OSX>
(hehe)
<waddlesplash>
no, 40 bytes *per file*
<Skipp_OSX>
oh I suppose memory just grows on trees where you're from
<waddlesplash>
if you have 2000 installed files, it'll save 78 MB
<waddlesplash>
etc.
<Skipp_OSX>
yeah, cool.
<Nephele>
I want less installed files sometimes
B2IA has quit [Quit: Vision[]: i've been blurred!]
<Nephele>
but quite good a change indeed
B2IA has joined #haiku
<Nephele>
most of my memory keeps getting consumed my webkit memory leaks though
<Skipp_OSX>
yeah of course
<Nephele>
waddlesplash: one cool thing to add would be to, like FreeBSD, reject writing to a device file that contains a mounted filesystem
<waddlesplash>
that would be nice
<waddlesplash>
we have some other bugs around exclusive access
<waddlesplash>
to open devices. but that's another storyt
<Nephele>
though currently if you unmount both filesystems on my flash drive it dissapears from the device tree and i dunno why
<Nephele>
so i have to keep it mounted to write to it which is silly
<Nephele>
also stealing SIGINFO would be cool :D
HaikuUser has joined #haiku
<waddlesplash>
hmm, does only FreeBSD have that, or do others?
<Nephele>
linux has it on *some* obscure cpu arch, i think the other BSDs have it too and illumos or so? not too sure about illumos
<Nephele>
but i know gnu coreutils has some support for it
<Nephele>
linux has it on alpha if i read this correctly
<jhj>
If the system is under memory pressure, you get a SIGDANGER, which you can choose to handle, usually taking a memory saving action or saving your work in case of a crash.
<Nephele>
jhj: that's silly, just kill the process that has the longest time on low memory. Like the X11 server ;)
<Nephele>
waddlesplash: yeah i'm getting 1024x768 and 800x600
<jhj>
hah
<Nephele>
the card is also a nvidia card currently inserted, so no chance for drivers on that one
<Nephele>
but the bios loader freezes with native resolution :D
<Nephele>
freebsd extended filesystem flags like undeleteable (by root) are also quite cool, you can protect special files with it
<Nephele>
but probably harder to add to bfs
<phschafft>
why?
<Nephele>
phschafft: compared to not modifying the disk format :)
HaikuUser has quit [Quit: Vision[]: i've been blurred!]
<Nephele>
I'd assume ufs2 is designed with those flags in mind
<phschafft>
I mean they are just a set of bits in the inode.
<jhj>
Linux has that too
<jhj>
chattr, etc.
<phschafft>
but you could just use an extented attribute. ;)
<Nephele>
jhj: hard to wrap my head around making it case insensitive beeing the hard complex case, from a user perspective it sounds like "One less thing to worry about" xD
<waddlesplash>
jhj: SIGDANGER/psdanger looks like a userspace version of what in Haiku is the kernel low_resource system
<Nephele>
hmm, wasn't there some hardware DB i could check for supported gpus for haiku
<jhj>
waddlesplash: a way to ask running processes to try to clean up for themselves (before invoking an OOM killer) is just polite.
<Nephele>
I agree, but on Haiku that also is less relevant because we don't overcommit by default
<waddlesplash>
jhj: we don't have an OOM killer
<jhj>
It could be used to kill backgrounded tabs sitting open on a browser for example.
<waddlesplash>
apps that request overcommitted memory are expected to deal with SIGSEGV themselves
<jhj>
I know, just, in general :)
<Nephele>
Well telling the browser, and it deciding this for itself, sounds much better to me than just shooting a process somewhere
<jhj>
exactly
<Nephele>
it can neatly store it's content even and restore the state if it wants
<jhj>
For a browser, also dumping any memory cache, etc.
<Nephele>
I kind of want a BColumnListView in team monitor and some generic cpu% mem% tabs
<jhj>
I can see a lot of potentially friendly behaviors
<Nephele>
Indeed
<Nephele>
waddlesplash: mind making a ticket for this one? I'm not too versed in our memory system, I think you can add more relevant info here :)
<Nephele>
Skipp_OSX: how hard would it be to implement alpha channel support for trackers desktop mode background images?
<phschafft>
I mean there is memory pressure. as in if you run low on memory (or swap) it might be worth to let applications store stuff in an application specific way and drop memory. it will most likely be better than to swap random pages.
<phschafft>
but I also think software should use stuff like madvise() some more.
<Nephele>
ah yes, discussions about HaikuOS are what i needed instead of packagekit .-.
dqk_ has joined #haiku
* Nephele
heads to bed
Nephele has quit [Quit: Vision[]: i've been blurred!]
dqk has quit [Ping timeout: 480 seconds]
wicknix_ has joined #haiku
_-Caleb-_ has left #haiku [#haiku]
_-Caleb-_ has joined #haiku
<Skipp_OSX>
alpha is possible...
<Skipp_OSX>
how hard? Well, not hard, but not implemented yet.
<Skipp_OSX>
(unfortunately they are all really hard to fix)
mmu_man has quit [Ping timeout: 480 seconds]
marzzbar has quit [Ping timeout: 480 seconds]
_0xdd has quit [Quit: WeeChat 4.2.2]
_0xdd has joined #haiku
<jhj>
Oh, speaking of useful signals, Sun has SIGPWR (and I think Linux does too but it terminates by default, or,it used to)
<jhj>
The idea is on power failure a UPS monitoring daemon sends SIGPWR to notify of the power failure, and processes can do something special to temporarily reduce their power usage and/or save user data.
<jhj>
It's hampered because there wasnt ever any corresponding "power restored" signal.
AlienSoldier has joined #haiku
<AlienSoldier>
latest qtwebengine broke falcon. Id that not the point of a package manager to manage dependency?
<AlienSoldier>
*is
vdamewood has joined #haiku
<jhj>
These days that functionality has been taken up by watching power events on dbus or what not, on linux
<jhj>
AlienSoldier: I'd make an issue just in case, but I'm sure it wasnt intentional :/
<AlienSoldier>
i am glad it is worked on at least, facebook messenger is not working reliably with it. Web+ a bit better but can't rplay audio message.
<AlienSoldier>
web epiphany is also still broken.
AlienSoldier has quit [Quit: Vision[]: i've been blurred!]
<Begasus>
OscarL! grabbing cmake-3.28.3-2-x86_64.hpkg and moving it to /Opslag/haikuports/packages/cmake-3.28.3-2-x86_64.hpkg
<Begasus>
:D
_-Caleb-_ has left #haiku [#haiku]
<OscarL>
hsp: but as Begasus said... before removing them all, make sure your system is working fine (so you don't need to go back to those old states)
_-Caleb-_ has joined #haiku
<Begasus>
mostly before I do it I do a reboot to be sure to boot with latest changes
<OscarL>
Begasus: nice!, do the VBox guest additions next! :-)
<Begasus>
and keep the last month around
<Begasus>
nah OscarL, not in my list atm :P
<Begasus>
have to be sure it's operational eg also testing on 32bit
<Begasus>
had to use CPPFLAGS and LDFLAGS to get it to build
<Begasus>
bugger, now I can't use nexcloud to share files :/
* OscarL
adds another "#define _DEFAULT_SOURCE" :-/
<Begasus>
export CPPFLAGS="_GNU_SOURCE"
<Begasus>
export LDFLAGS="-lbsd"
<Begasus>
tried that?
jmairboeck has joined #haiku
<Begasus>
hi jmairboeck
<jmairboeck>
Hi Begasus
<OscarL>
this one needs _DEFAULT_SOURCE, tried with export first, no go... and... I just want to make it compile first, then make it better :-D
<OscarL>
good day jmairboeck :-)
<jmairboeck>
OscarL: the proper way to deal with _DEFAULT_SOURCE, _GNU_SOURCE etc is the -std compiler flag: when using -std=c++11 or something, you don't get them. You need -std=gnu++11, or no -std at all.
<jmairboeck>
so if something in the build system sets -std, I think the easiest would be to patch that
<OscarL>
jmairboeck: will keep that in mind, thanks!
<jmairboeck>
or the equivalent for C without the ++
<OscarL>
problem is... this thing uses "kbuild", those .mk are pretty large, and as usual... I barely know what I'm doing :-)
<OscarL>
(context: virtualbox_guest_additions)
<Begasus>
if(CMAKE_SYSTEM_NAME MATCHES "^(Linux)$" OR "Haiku")
<OscarL>
yeah, that was why I mentioned it early (but was too lazy to link it :-D).
<Begasus>
just commented there
MrSunshine has joined #haiku
<Begasus>
with all the non prior PR's atm lost it on the way :)
<Begasus>
k, importing patch changes from madmax ...
<Begasus>
hp -c ...
<Begasus>
how big is this webkit_gtk thing?
<OscarL>
no clue (and hope to never find out on my own).
<OscarL>
how long will this darn vbox-ga build will take?
<OscarL>
I miss those [x of y] progress from some other build systems :-)
<Begasus>
nuked build on 32bit, rebuild with madmax patch
MrSunshine_ has joined #haiku
_-Caleb-_ has left #haiku [#haiku]
MrSunshine_ has quit []
MrSunshine_ has joined #haiku
MrSunshine has quit [Ping timeout: 480 seconds]
<Begasus>
checking cmake with konqueror :)
<Begasus>
or konqueror with cmake? :P
<OscarL>
darn, OSS was enabled on the main Config.kmk ... here we go again.
Nephele has joined #haiku
<Begasus>
heh
<Begasus>
Hi nephele
<Nephele>
hi there
<Nephele>
qmplay2 works fine .9
<Nephele>
:)
<Begasus>
nice!
<Begasus>
uninstalled it already here :P
<Begasus>
nephele, do you have a 32bit install where you could check something that uses qtwebengine?
<Nephele>
no
<Nephele>
atleast, not at this time
<Begasus>
looks like it's broken on 32bit, gives error in syslog for falkon and nexcloud (2 that I mostly use)
<OscarL>
nephele: hi... earlier you wrote: "though currently if you unmount both filesystems on my flash drive it dissapears from the device tree and i dunno why"
<OscarL>
maybe due to "Eject when unmounting" setting from Tracker preferences?
<Nephele>
OscarL: maybe? but why would that affect thumb drives?
<OscarL>
guess it affects all "removable" devices?
<Nephele>
But I do have it set, I guess I will uncheck it. I was having this checked since I assumed the dvd would eject on unmount
<Nephele>
and it does too :)
<Nephele>
but for thumb drives this makes no sense as they can't be ejected as such
<OscarL>
sounds like it needs a ticket then :-P
<Nephele>
that just means you can't remount them without having to manually unplug them or reboot the computer :/
<OscarL>
yeah, that's why I keep it unchecked too.
<Begasus>
"now" I have kdesu on 32bit also for KF5 :)
_-Caleb-_ has left #haiku [#haiku]
_-Caleb-_ has joined #haiku
mmu_man has quit [Read error: No route to host]
diver has quit [Ping timeout: 480 seconds]
<Begasus>
konqueror same issue with qtwebengine
diver has joined #haiku
Coldfirex has joined #haiku
<Begasus>
PulkoMandy, is there a way to use cmd:autoconf >= 2.69 for postgresql? can't seem to trigger it :/
<Begasus>
v11*
<PulkoMandy>
I don't know; I think the autoconf269 package doesn't provide a command named just autoconf?
<Begasus>
no it doesn't
<Begasus>
I see in the issue postgresql12 also needs it
<PulkoMandy>
maybe we can create a symlink autoconf->autoconf269 in the postgresql build if there is no other way, or maybe we can create a package autoconf269_autoconf wiht just a symlink from "bin/autoconf" to "bin/autoconf269". But the simpler way is if postgresql buildsystem has a way to force what autoconf binary to use
<Begasus>
checking repology ... maybe there is something there
<Begasus>
whoot, lacking behind there :D
<Begasus>
missing libdap_r in latest openldap
<Begasus>
and in the older openldap the devel packages are nuked
Nephele has joined #haiku
Jixt has joined #haiku
<hsp>
Begasus, you mean rm -r /system/packages/*
<OscarL>
do *NOT* do that.
<Begasus>
jikes no!!!
<Begasus>
maybe better use the tool to clean up some administrative states
<Begasus>
k, got postgresql building to end up with undefined errors :/
<hsp>
OscarL, which tool?
bbjimmy has quit [Quit: Vision[]: i've been blurred!]
<OscarL>
FilWip
<hsp>
ok
<Nephele>
you can do that, but only once ;) your OS won't boot afterwards since you'll have deleted all system packages
<Begasus>
don't think it will go down without a warning nephele :)
* OscarL
signs off, wishining everyone a nice day.
OscarL has quit [Quit: Gone with the wind]
<Begasus>
cu ...
<Begasus>
grabbing postgresql11-11.1-5-x86_64.hpkg and moving it to /Opslag/haikuports/packages/postgresql11-11.1-5-x86_64.hpkg ...
<Begasus>
maybe not the best atm, but it build (openssl3 and icu74)
<Begasus>
PulkoMandy, disabled autoconf and added -lnetwork with LDFLAGS in BUILD() and option --disable-thread-safety
<Begasus>
also re-enabled devel package for older openldap*
<PulkoMandy>
well, if it works that way, there is no need to run autoconf
<Begasus>
guess those could be dropped on an uptodate postgresql, for now it should be good enough
<PulkoMandy>
we had to do this back when most configure scripts didn't know about haiku, but now they do
<Begasus>
yeah been there :)
<Begasus>
checking v12
frkazoid333 has quit [Read error: Connection reset by peer]
<Begasus>
nice :) grabbing postgresql12-12.0-5-x86_64.hpkg and moving it to /Opslag/haikuports/packages/postgresql12-12.0-5-x86_64.hpkg
deneel has quit [Remote host closed the connection]
<Begasus>
Warning: POLICY WARNING: "/packaging/postgresql12_server/bin/postgres" needs library "libldap-2.4.so.2", but the package doesn't seem to declare that as a requirement
<Begasus>
probably could switch that definition to make it use the proper devel package also
<Begasus>
k, let's see what we still need for postgresql
<Begasus>
jmairboeck, if you read this, changes to openldap/postresql* is only done to have "current" verrsions still around, no hastle further allong the line :)
<waddlesplash>
and we clear the 64BIT_RETURN before calling the debugger (if we call the debugger at all)
<waddlesplash>
so we can't just check this in the debugger itself
<waddlesplash>
I guess it's possible that EDX is only caller-saved when the caller has a 64bit return value?
<PulkoMandy>
waddlesplash: I think the change in that commit is not just changing the register value, but replacing it in the stackframe (where maybe a register value was saved during the call)
<waddlesplash>
PulkoMandy: hmm. maybe so
<PulkoMandy>
But I don't know enough 32bit x86 to be sure how this all works...
<waddlesplash>
I guess this isn't a safe change after all. let's rework it to not do that then
<mmlr>
I mean if you can easily reproduce it, backing out the change and checking would be easy
<waddlesplash>
I can, but it's annoyingly intermittent
<waddlesplash>
but EDX does seem to basically always be involved
<waddlesplash>
so I don't know what else it could be
<PulkoMandy>
and I'm not setup for cross-development so I don't know if I can manage to build a test image. But I can try
<waddlesplash>
I am set up, I'll test it
<Begasus>
closing down
mmlr has quit [Remote host closed the connection]
<Begasus>
cu peeps!
Begasus has quit [Quit: Vision[]: i've been blurred!]
_-Caleb-_ has left #haiku [#haiku]
mmlr has joined #haiku
<mmlr>
just had one of those boot hangs again on x86_64 where it just sits at the boot screen with everything lit
<waddlesplash>
(this is not a problem on beta branch however)
<waddlesplash>
(because on beta branch the network module isn't unloaded)
<PulkoMandy>
I'm trying to understand what the initial hroblem in strace was. As far as I can see, the return type of syscalls is known (from gensyscall data), why would strace try to print 32 bit values as 64 bit then?
<waddlesplash>
strace just prints whatever the debugger gave us
<waddlesplash>
if there's a 64-bit value, it prints that
<mmlr>
waddlesplash: I did not check enough threads to know, but the timing and what was running at that time would fit
<waddlesplash>
we could maybe fix this on the strace side but it seems bad to leak random kernel values
<waddlesplash>
what would be ideal would be for the debugger to know if we had a 64 bit syscall value
<waddlesplash>
but we clear the flag by then
<waddlesplash>
and we have to use an atomic loop to do it, so, it's not easy to move that
_-Caleb-_ has joined #haiku
<waddlesplash>
the basic patch to not clobber EDX is easy enough
<waddlesplash>
but fixing the original bug again is tough
<PulkoMandy>
strace knows about the type from gensyscalls data, and uses that in a variety of ways already (to decode status_t into errors for example), so it could use that to know the size of the return value as well
<PulkoMandy>
Anyway, I think the asm code can be changed so that the jnz skips over the modification of edx in the iframe, instead of setting edx to 0 and then overwriting the iframe with that
<waddlesplash>
yes
<waddlesplash>
I made that edit, but I wanted to fix the other half before rebooting to test
<waddlesplash>
maybe we can just move the THREAD_FLAGS_DEBUGGER_INSTALLED check
OrangeBomb has quit [Read error: Connection reset by peer]
<PulkoMandy>
I'd rather fix it in strace and not add extra code in the syscall handler, if possible. We don't really want to make it slower
<waddlesplash>
we need to not leak kernel memory here no matter what
<waddlesplash>
so we need to have the testl and then avoid setting EDX if we can
<waddlesplash>
but the extra code for the DEBUGGER_INSTALLED case doesn't matter because it won't be executed if the flag is unset
<waddlesplash>
that case will already be much slower, it won't make much difference
<waddlesplash>
mmlr: btw, if you create areas with PROT_NONE now, they won't reserve any memory
<waddlesplash>
they'll reserve it on demand when you mprotect
<waddlesplash>
for your guarded heap changes
OrangeBomb has joined #haiku
HaikuUser has joined #haiku
<mmlr>
it already uses B_OVERCOMMITTING_AREA
<jhj>
Would it be OK to open a ticket to add MADV_DONTDUMP and MADV_DODUMP to madvise?
<jhj>
These are supported decently widely elsewhere.
<waddlesplash>
how widely?
<waddlesplash>
core files on Haiku are a bit different than elsewhere
<waddlesplash>
mmlr: ah
<jhj>
Linux, AIX, FrseBSD, and current Solaris.
<waddlesplash>
right now most of our MADV do nothing
<waddlesplash>
I think MADV_FREE may be the only one that does anything
<jhj>
It's MADV_NOCORE and DOCORE on FreeBSD.
<jhj>
It's super useful to exclude hundreds of megs of stuff (virtual machine memory) from emulators for example.
<waddlesplash>
I don't know how useful this would be on Haiku because most of the time we just generate text reports
<waddlesplash>
if you really want a core file why wouldn't you want everything?
<jhj>
waddlesplash: A virtual machine that has 8GB of memory assigned for example.
<waddlesplash>
okay but if most of that's zeroes, what's the problem
<waddlesplash>
and under what scenarios would you want a full core file but not that memory?
<PulkoMandy>
Why not? Anyway, opening a ticket sounds good. Not sure if anyone will get to it soon
<jhj>
LLVM uses it in a few places.
<jhj>
Some of those automatic crash reporters user it to try to avoid egregiously sending user data.
<waddlesplash>
automatic crash reporters should use the text report format on Haiku
<jhj>
I'll open a ticket later, it's nothing that's urgent.
<waddlesplash>
doesn't send any user data (well, besides what's in stack frame memory and registers)
<waddlesplash>
this is the patch I'm going to test
marzzbar has joined #haiku
Jixt has quit [Ping timeout: 480 seconds]
<mmlr>
waddlesplash: but that always clears edx regardless of the 64 bit flag as the xor is before the test?
<waddlesplash>
mmlr: in post_syscall_work() yes
<waddlesplash>
we either want it to be 0 or the value from the syscall
<waddlesplash>
this is after we've stored it inside the iframe field, that's not affected
<waddlesplash>
well, the patch boots anyway
<mmlr>
I read it as it being stored into the iframe at line 60 but cleared at line 57
<waddlesplash>
unless I'm mistaken, line 60 is reading it, not storing it
<waddlesplash>
lines 28 and 30 store it
<waddlesplash>
this is GNU assembler syntax, I think
<mmlr>
yeah, that's what confused me
<waddlesplash>
anyway considering how rare the crashes were, I wonder if this code was actually broken before, but it wasn't noticed because EDX didn't get properly clobbered very often
<waddlesplash>
anyway with the patch, strace still works and a full rebuild of the haikuports repository directory did not crash at all when run with libroot_debug (not guarded heap)
<waddlesplash>
previously it crashed at least 3-4 times (regular malloc) or almost every single recipe (libroot debug)
<mmlr>
so the difference to before is that we leave edx in the iframe alone if it's not a 64 bit syscall instead of resetting it to 0?
<waddlesplash>
so, I guess that fixes it
<waddlesplash>
mmlr: yes
<waddlesplash>
and before my previous change, we clobbered it with whatever we had locally for EDX
AD_Haiku_onPC has joined #haiku
<waddlesplash>
which maybe was always correct if we loaded EDX from the iframe?
<waddlesplash>
at the beginning that is, and then nothing later on clobbered it without restoring it
<mmlr>
maybe
<mmlr>
if that wasn't the case, it should have crashed as well at least sometimes
<waddlesplash>
yes
<waddlesplash>
but less often
<waddlesplash>
I wonder if there are any random crashes on x86 that this will clear up