bbjimmy has quit [Quit: Vision[]: i've been blurred!]
<oanderso[m]>
I'd like to better understand the scenario where we could end up freeing a page table while in that code path. It's not clear to me how it's actually possible.
<x512[m]>
Large pages are currently not supported.
<oanderso[m]>
You're not doing this madness of zeroing each PTE individually and doing a partial TLB flush along the way
<x512[m]>
Why flush dead table?
<oanderso[m]>
No clue
<oanderso[m]>
The code predates me
<oanderso[m]>
It looks like it was written assuming you were freeing a live table
<oanderso[m]>
Which seems like a crazy thing to do
<x512[m]>
There is a possibility that original code is misguided and can be simplified a lot.
<oanderso[m]>
I suspect so but it would be nice to get confirmation
kinkinkijkin has joined #haiku
erysdren has joined #haiku
nipos has left #haiku [Disconnected: Replaced by new connection]
nipos has joined #haiku
kinkinkijkin has quit [Quit: Leaving]
<erysdren>
if i wanted to write a filesystem driver for Haiku... where would i start? i've written code for Haiku before, but i've never written anything for the kernel. just userspace apps.
<erysdren>
for context, i want to write a ZFS driver. i use one on my linux system
<erysdren>
iso9660 seems like the smallest one to get an idea of what glue is needed.
<OscarL>
yeah, was about to point that or cdda :-)
<erysdren>
might be better as FUSE, for haikuports, so we can pull in OpenZFS.
<OscarL>
should be easier to debug in user space too :-)
<Begasus>
Hi OscarL
<OscarL>
Good day Begasus!
<Begasus>
Hi coolcoder613 :)
<Begasus>
running over report errors due to dropping old flac today
<OscarL>
always something to fix :-)
<Begasus>
porbably will drop risidualvm, that's integrated into scummvm for a while now
<Begasus>
with csound "fixed" old fluidsynth should be good to drop too
<Begasus>
I'll probably forward you an email from Pete in regards to csound OscarL, got a new one yesterday :)
<OscarL>
There might be a way to make csound use old fluidsynt (using SOURCE_URI2, and building it just for it). But that's for someone that really wants midi out in csound :-)
<botifico>
[haikuports/haikuports] Begasus 2250037 - cmus, revbump for old flac removal (#10912)
<OscarL>
that feedgator PR surely could use a "squash_last_n_commits" when TmTFx finish with the changes :-)
<Begasus>
yeah :D
<Begasus>
well in his defense, he's not experienced with it :)
<OscarL>
Doesn't github also allows you to "squash on merge"? (not sure if it allows you to edit the commit message, thou).
<Begasus>
if you have commit acces it can be done :)
<Begasus>
did someone of the Haiku dev's enabled a thread peak when visiting github? :P
<OscarL>
indeed :-) but I meant via the web UI (I think I saw something on that regard in one of the repos I do have commit access)
<Begasus>
laptop goes into overdrive when I do
<Begasus>
the "button" mentions "squash and merge" yes
<Begasus>
and one can change the title easy
<Begasus>
for commit message one would need to push a commit --amend for it
<Begasus>
resolving build dependencies failed: requires "pillow_python39" of package "qrcode_python39-7.4.2" could not be resolved
<Begasus>
tsss
<Begasus>
OscarL, afaict pillow doesn't use any extry python packages right?
<OscarL>
"Allow squash merging" setting, gives some options for the commit message, but not direct editing... thus why it is better if the users do the squasing themselves.
<Begasus>
right, if they are familiar with it
<OscarL>
re: pillow.. at runtime, seems not (according to the .recipe at least).
<Begasus>
right, only see setuptools in build requirements, so maybe re-add python3.9 (for qrcode)?
<OscarL>
seems a bit weird that it requies lost of devel:lib* but none at runtime, thou.
<OscarL>
s/lost/lots/
<Begasus>
haven't touched it in a long while, so not sure what it really needs
jmairboeck has joined #haiku
<Begasus>
hello jmairboeck! :)
<jmairboeck>
Hi Begasus
<jmairboeck>
OscarL: Thanks for your feedback on the python recipes!
<OscarL>
TmTFx HaiQr app really uses Python 3.9? if not... I don't see why qrcode should have 3.9 support.
<OscarL>
jmairboeck: no problem! Thanks to you for your patience with me (specially when I mix thing up :-D)
<coolcoder613_>
OscarL: 0x0.st website has contact info
coolcoder613_ is now known as coolcoder613
<Begasus>
we talked about it a bit, guess it's only needed if HaiQR goes python3.9 also (ash Haiku-PyAPI supports it)
<OscarL>
Begasus: HaiQr is a standalone app, no need to have both 3.9 and 3.10 versions at the same time.
<OscarL>
it doesn't really matters here if Haiku-PyAPI supports multiple version or not.
<OscarL>
jmairboeck: LOL at a python project usiong "/dev/stdout" for output :-)
<OscarL>
good on psutil fixin' deps too!
<jmairboeck>
it won't really change anything though, but we can move some things to TEST_REQUIRES then
<jmairboeck>
wand and pytest_datafiles
<OscarL>
clearer deps at least are less puzzling :-)
<jmairboeck>
OscarL: that argparse-manpage patch is about as good as my python coding skills get though, and I had to google how to use stdout properly in Python ;)
<OscarL>
jmairboeck: psutils-3.3.5.recipe
<OscarL>
looks good to me.
<jmairboeck>
+1 OscarL :)
<jmairboeck>
I think I'll create a draft PR for that branch
<OscarL>
jmairboeck: re: python skills... and yet, you did better than upstream! way to go! :-)
<Begasus>
k, back to qrcode recipe ...
<Begasus>
hmm was looking into qmmp ...
<jmairboeck>
I only noticed it because I added tests to the recipe and they were failing
<jmairboeck>
because there is no /dev/stdout in Haiku
<Begasus>
bugger, qmmp fails with either ffmpeg version :/
<OscarL>
just noticed that TmTFx mixup is because he didn't used branches.
<Begasus>
ah right, hadn't noticed :)
<OscarL>
Added a comment with some quick tips for him. Hope is clear enough.
* Begasus
fears to take a look :P
<Begasus>
ah, not so bad :D
<Begasus>
k, think for the 2 remaining I'll just revive base flac13 package untill this can be build/bumped
<Begasus>
well risidual_vm could be dropped, but since qmmp doesn't build here for either ffmpeg version ...
<OscarL>
I really think HaiQr should handle that type of dependencies on its own. (instead of us having to add a new recipe for "pypng"... things get out of hand).
<Begasus>
that or move back to older version
<Begasus>
but granted, would be easy if TmTFX could drop in here :)
<Begasus>
k, older version still good
<Begasus>
OscarL thanks so far
<Begasus>
will drop qrcode and ask to use "pip3 install --user qrcode" instead, maybe could be done with a postinstall script
<Begasus>
keep it around for when that is updated then TmTFx :)
<Begasus>
I'm not seeing that whitespace here in the feedgator recipe
<OscarL>
Seems so Begasus, there is now an empty line separating the REQUIRES from INSTALL().
<OscarL>
Hey there TmTFx :-)
<TmTFx>
Hi @OscarL!
<OscarL>
TmTFx: I replied to your last comment on Gerrit.
<Begasus>
ah! No empty line between PROVIDES and REQUIRES TmTFx :)
<Begasus>
ps TmTFx, if you use "git commit --amend" it doesn't add another commit
<OscarL>
TmTFx: and don't worry about making "newbie mistakes", we are all learning here :-)
<Begasus>
instead of "git commit -m *" :)
<Begasus>
yeah, OscarL always keeps me on edge with his python recipe changes :P
<OscarL>
TmTFx: and to be honest, git can be a PITA, and starting with .recipes can be a challenge too.
<Begasus>
be glad fbrosson isn't around :D
<Begasus>
but yeah, at first it can be a bit scary
<Begasus>
but I'm glad to have got some new help in that area, so kudos there too :)
<TmTFx>
@OscarL I see, well, I don't remember which pc I used to push the changes ^^'''''' I know it's not nice to pass the buck, but I'm currently swamped with work. I have to go to work soon, among other things.
<OscarL>
TmTFx: I don't mind makeing the changes on the intel driver. Just wanted to make sure to not step on your toes, so to speak.
<Begasus>
OscarL, can you remove the suggestion to feedgator or should I do?
<botifico>
[haikuports/haikuports] tmtfx 3b786a6 - remove empty line (#10918)
<coolcoder613>
Also it would be nice if it was cleaned up a little
<ilzu[m]>
coolcoder613: You linked a link to Pragtical editor few days a go. It complies without problems on Haiku, though the plugin manager needs some changes and never libgit2. Some of the plugins need some work too. https://ilzu-dev.hilppa.org/~ilzu/files/pragtical2.png
<Begasus>
that that was fixed TmTFx :)
<coolcoder613>
ilzu[m]: I had just seen it on HN
<coolcoder613>
I've never used it
<TmTFx>
bye!
TmTFx has quit [Quit: Vision[]: i've been blurred!]
<botifico>
[haikuports/haikuports] Begasus 91a7cbd - poppler24, bump version (#10920)
<coolcoder613>
Hello zard
<zard>
Hello coolcoder613 :)
<zard>
I read your two new blog posts
<coolcoder613>
Did you like them?
<zard>
Eh, nothing particularly interesting tbh
<coolcoder613>
I'm currently installing NetBSD on the Deskpro RN
<zard>
Ofc, I doubt there's much else that you have to write about, so they're perfectly fine :)
<coolcoder613>
The blog is mainly there to document my projects
<zard>
+1
<coolcoder613>
So I'm not explaining them seperately in lot's of different places
<coolcoder613>
Plus, I get some writing practice, which is a good thing
<milek7>
oanderso[m]: when software access/dirty fault is taken
<milek7>
if page table is freed between the moment handler walks the page table and actually modifies the entry
<milek7>
and backing page is reallocated to something else, and happens to accidentally contain the same bitpattern, then it ends up flipping some bits it shouldn't
<milek7>
as for when it could happen, something like calling munmap to yank memory under another running thread
<milek7>
and this is a problem only for software handler, hardware access/dirty management is fine I think
cyrus_ has joined #haiku
<milek7>
as for tlb flushing in FreeTable, yeah you could move that out to FreeTable callee
cyrus_ has quit []
<milek7>
*caller
cyrusbuilt has joined #haiku
<milek7>
but va range tlb shootdown is in architecture extension, so either you need to keep that as fallback or flush everything when extension is not available
walkingdisaster has joined #haiku
tuaris has joined #haiku
HaikuUser has joined #haiku
bbjimmy has quit [Quit: Vision[]: i've been blurred!]
<Begasus>
if it doesn't work an issue should be reported at haikuports
<JackDaniel>
woooow, it works! Thanks, man, you save my day! :)
<Begasus>
np :)
<JackDaniel>
ah, another weird thing regarding telegram
<Begasus>
upstream should get a revbump then (or I could push the newer version)
<JackDaniel>
I use my phone as a hotspot, because I dont have internet here other than 4G
<JackDaniel>
and if someone calls me when I have telegram opened, it crashes
<JackDaniel>
this is happen only with the last version, which is 5.4.10-1
<JackDaniel>
this is happen only with the last version, which is 5.4.1-1
<Begasus>
no idea there, got an account but I only got it to check it out, never use it
<JackDaniel>
I reported it as a bug
<Begasus>
+1
<JackDaniel>
I just wanted to inform you
<Begasus>
3dEyes will tackle it I guess
<Begasus>
np thanks :)
walkingdisaster has quit [Quit: Vision[]: i've been blurred!]
<Begasus>
I try to tackle those I know how to testdrive good enough :)
<JackDaniel>
@Begasus: Thanks a lot for helping me!
<Begasus>
no problem :)
JackDaniel has quit [Quit: Vision[]: i've been blurred!]
<Begasus>
now what was I doing ...
<Begasus>
k, will leave old qca recipe alive for now
<Begasus>
qgis will need a revbump for the correct library name first
HaikuUser2 has quit [Quit: Vision[]: i've been blurred!]
<Begasus>
bbl
kescher has quit [Quit: Bye]
kescher has joined #haiku
smalltalkman__ has joined #haiku
HaikuUser has joined #haiku
HaikuUser has quit []
freddietilley has quit [Quit: WeeChat 4.2.2]
flag_ has joined #haiku
_-Caleb-_ has left #haiku [#haiku]
_-Caleb-_ has joined #haiku
flag has quit [Ping timeout: 480 seconds]
Coldfirex has joined #haiku
<mbrumbelow>
Is the HaikuDepot not functioning correctly?
gouchi has joined #haiku
frkazoid333 has joined #haiku
<Skipp_OSX>
I believe we turned r1b4 off, not enough resources to maintain both r1b4 and r1b5 is what I've been told.
cyrusbuilt has quit [Quit: Client exiting]
cyrusbuilt has joined #haiku
Coldfirex has quit [Remote host closed the connection]
dalme has joined #haiku
DigitalBox98 has joined #haiku
DigitalBox98 has quit []
<mbrumbelow>
Thanks
Coldfirex has joined #haiku
<x512[m]>
milek7: What is a fundamental difference between riscv64 and arm64 that makes things so complicated with arm64?
<kallisti5[m]>
Skipp_OSX: correct(ish). There are some limitations with haikuporter buildmaster in our infrastructure making maintaining another branch repo difficult
<kallisti5[m]>
tldr; we have to double our haikuporter footprint which has a lot of volume mounts which exceeds the maximum PV mount on a single kubernetes node at Digital Ocean
<x512[m]>
As I understand page tables works nearly the same in all modern architectures. 32 bit PowerPC seems different.
<kallisti5[m]>
x512: lol. "Object Storage" is the industry term for "Non-AWS s3"
<Skipp_OSX>
FAT does, but do we, I'm looking at the commit just pushed and it seems like it restricts to ASCII but maybe not I can't really tell for certain
<kallisti5[m]>
* "Non-AWS s3 services"
<Skipp_OSX>
I do know that I had to do some UTF-16 => UTF-8 conversion work on exfat to get the volume label to appear correctly when Unicode characters are used.
<x512[m]>
kallisti5: Is it some kind of network file system?
<kallisti5[m]>
S3 is pretty much pragmatic file storage and management over https.
<kallisti5[m]>
(and rest calls)
<kallisti5[m]>
storage areas are divided into buckets.. you put files into buckets via http rest calls (actually.. it might be soap? I don't remember)
<kallisti5[m]>
and they can show up via a url
<x512[m]>
So what is the problem? Package builder can have a cache of HPKGs needed as dependencies for building packages and download then on demand by using some protocol.
<kallisti5[m]>
yup. However, someone has to add that feature to haikuporter
<kallisti5[m]>
and haikuporter makes a LOT of assumptions about local files
<kallisti5[m]>
and repos.. etc
<kallisti5[m]>
it's a solvable problem, but it's difficult to attract people to work on it
<x512[m]>
Its Python, so I am out.
<kallisti5[m]>
^^^ that's pretty much what everyone says LOL
<kallisti5[m]>
I need my { }
<x512[m]>
I am slowly making replacement tool in C++>
<kallisti5[m]>
nice
<x512[m]>
s/++>/++./
<kallisti5[m]>
fun fact: haiku's repo is actually served by a neat Rest microservice I wrote a few years ago
<kallisti5[m]>
yeah... C++ is bad at anything that needs substantial libraries
<kallisti5[m]>
just from the weight of hauling them around
<phschafft>
x512[m], Skipp_OSX: ah, so exfat vs. fat. thanks for the clarification. :)
<x512[m]>
Exfat is still patented.
<x512[m]>
Implementing it may be legally dangerous in USA.
<x512[m]>
Linux have spacial permission from MS.
<waddlesplash>
Microsoft contributed the patent licenses to the Linux Foundation
<waddlesplash>
but we aren't part of that
<Skipp_OSX>
correct, currently we only have read support. It would be nice to have write support even if we had to bury it under a patent encumbered flag
<x512[m]>
But not to Haiku. And Haiku is not a member of Linux Foundation.
<x512[m]>
There are no significant benefits of ExFat comparing to Fat. Still no journal.
<Skipp_OSX>
right ok but this isn't about exfat it's about naming your fat volumes with unicode characters in Winders and then plugging them into Haiku and it goes fizzle pop
<Skipp_OSX>
or not, I don't know that's why I'm asking
<x512[m]>
Haiku bug in FAT driver?
<Skipp_OSX>
possibly
<Skipp_OSX>
yes
<Skipp_OSX>
There was a Haiku bug in the exfat driver, I fixed it, wonder if same fix needed for regular fat
<kallisti5[m]>
Oh look, the 👻 of me warning y'all about merging a new fat driver before a major release 😅
PetePete has quit [Ping timeout: 480 seconds]
<Skipp_OSX>
hehe
<kallisti5[m]>
(not discounting how nice the new one is... I was just worried about timing)
PetePete has joined #haiku
<dovsienko>
the only two reasons I ever needed to use exFAT were: 1. support for large disks 2. a smart TV that supported playing media files from USB only if it was in exFAT
<x512[m]>
Are the any objections to removing GUI from CMake package? It makes terrible cyclic dependency.
<Skipp_OSX>
I'm trying to fix this guy: https://dev.haiku-os.org/ticket/18999 I set the MaxContentWidth on the menubar inside the menu field and that almost fixes the issues, but it still overwrites the drop down and idk why.
<x512[m]>
> smart TV that supported playing media files from USB only if it was in exFAT
<x512[m]>
Really? Even FAT is not supported?
<dovsienko>
for the latter one, after I needed to play a file on another smart TV, I prepared an exFAT USB stick, but it didn't like it. it turned out the second TV accepted ext2 because it was running Linux
<phschafft>
I generally don't care much for fat. but I was wondering as I was looking into filesystem identification earlier.
<Skipp_OSX>
You don't have to be in the Dutch languages to trigger the bug, just select Application => "OpenDocument Database (vnd.oasis.opendocument.database)" or other similarly long titled mime type.
<x512[m]>
kallisti5: It seems that Rust is needed to be added in haikuports.cross.
<dovsienko>
x512[m]: AFAIR, the problem was that FAT does not support files larger than 2GB, and the file was larger than 2GB. so perhaps it was support for large files, not support for large disks
<Skipp_OSX>
apparently Microsoft recently fixed this in Winders 11
<Skipp_OSX>
Meaning, allows you to create larger than 2GB FAT32 partitions
<Skipp_OSX>
It says 4gb on the article but i think in reality it is 2gb
<phschafft>
last time I checked with $some_version_of_fat the problem was that the filesystem could handle 4GB but most implementations used 32 bit *signed* types for file size.
<phschafft>
e.g. the classic off_t is 32bit signed. so only allowing for 2GB files.
<Skipp_OSX>
https://0x0.st/Xtc5.png <= Set the MaxContentWidth() that makes it closer, but even if I subtract off the end to leave space for the drop down it's not showing it.
frkazoid333 has quit [Read error: Connection reset by peer]
frkazoid333 has joined #haiku
<Skipp_OSX>
I messed with the IconMenuItem::GetContentSize width but no joy there either... _BMCMenuBar_ should be handling leaving space for us but idk why it isn't here. I vaguely remember this being a thing I wanted to fix before but then didn't, something to do with menu insets.
<kallisti5[m]>
<x512[m]> "kallisti5: It seems that Rust is..." <- errrrr. what needs Rust in build-packages?
<x512[m]>
avif1.0
<kallisti5[m]>
oh...
<kallisti5[m]>
neat. So.... I think our rust is still cross-compiled?
<kallisti5[m]>
we're not native yet
<x512[m]>
It is impossible to compile Rust compiler without Rust compiler.
<milek7>
oanderso[m]: that could work, but then you're stuck with these pages for address space lifetime? I thought also about delaying releasing the pages, but how to determine how long is enough
<kallisti5[m]>
x512: anyway, you're correct. We need to ping nielx and get the instructions into the haikuports repo at minimum
<kallisti5[m]>
It's a standard process to cross-compile rust, i just don't know it
<x512[m]>
milek7: It is already so for x86. Non-leaf page table entries are freed only on process termination.
<x512[m]>
Non leaf page table entries freeing can be implemented, but it is not a priority and can be done later. Some common solution should be found for x86 and riscv64 first.
<oanderso[m]>
You could do a prune pass on context switch or something
<oanderso[m]>
I guess that could still race
nosycat has quit [Quit: Leaving]
<oanderso[m]>
Maybe if all threads are unscheduled then you could do it
<waddlesplash>
that's probably not easy to determine
<waddlesplash>
I vaguely recall talking about mmlr about something like this at some point
<waddlesplash>
I think it was about this anyway. He mentioned some strategy we could use for implementing it, but this was years ago and I don't recall the details
<waddlesplash>
I could possibly go dig up those chat logs...
<milek7>
maybe keep atomic counter of currently walking handlers
<x512[m]>
Priority is getting working arm64 port, optimizations later.
<milek7>
then when it's 0 you can acquire spinlock to prevent new handlers from beginning the walk
<dovsienko>
1. make it work 2. make it right. 3. make it fast
<waddlesplash>
yeah, doesn't really matter right now, we can just add TODOS
yann64 has joined #haiku
<Skipp_OSX>
waddlesplash can you merge in https://review.haiku-os.org/c/haiku/+/8100 pls, you can cherry-pick just the "Tracker: Change kMoveSelectionTo to kCopySelectionTo on paste" commit into r1b5 if you are worried about the stack variable causing a problem (although I know it doesn't)
<Skipp_OSX>
The commits are completely independent of each other so it shouldn't make a difference either way... but I would like the stupid Cut items staying semi-transparent bug fixed for R1B5.