17:16 <headius[m]> yeah I can run some numbers

17:16 <headius[m]> this is the issue the jetbrains folks opened yeah?

17:22 <headius[m]> I'm setting up windows to investigate some other perf stuff there but I can circle back...not critical for 9.2.8

17:58 <Ryctolagus> I have a jruby on rails web application using puma, and I had a question on IIS best practices. which would be industry preferred: using URL rewrite and host the JVM independent of IIS, or have an IIS worker process manage the JVM and puma?

20:21 <enebo[m]> https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/setmode?view=vs-2019

20:21 <enebo[m]> If we can dig handle I think this can change to O_TEXT

20:21 <enebo[m]> we need to make fd out of handle though

20:21 <enebo[m]> Not sure about how Java 9+ deals with our handle grabbing

20:22 <headius[m]> Ryctolagus: hey there

20:23 <Ryctolagus> headius: Good Day :)

20:23 <headius[m]> Ryctolagus: I'm not sure anyone here is qualified to answer that...I have not used IIS in 15+ years

20:24 <Ryctolagus> headius: I kind of figured that, I also asked in #windows-server. However I think this is an edge case so I will just have to test and profile my enviroment to determine which is a more robust solution.

20:26 <headius[m]> If there's some simple way to set up IIS to reverse proxy to a free-standing JVM I think that's what I'd prefer

20:26 <headius[m]> but that might not fit your organization security/management policies

20:26 <headius[m]> we recommend people deploy JRuby apps with Puma, but I would never recommend Puma being the web server...so from there it's the same question as hosting any other language runtime on Windows

20:26 <headius[m]> enebo: yeah digging handle out I'm sure we can find a way to twiddle the right bits

20:26 <headius[m]> and then turn off the crlf translation

20:26 <headius[m]> I'm trying to see if there's anything in JDK itself that we could use

20:26 <headius[m]> omg windows trackpad palm detection is actually worse than Mac

20:27 <headius[m]> Ryctolagus: yeah let us know what you figure out...I feel like this has been asked before and we didn't have a good answer then

20:27 <enebo[m]> headius: yeah that would be better but I believe we already have the handle to fd function exposed in jnr-posix but I am not so sure about _setmode

20:28 <Ryctolagus> headius: Setting up IIS reverse proxy as a reverse proxy to puma is actually what both solutions do. The difference is in who (me, or IIS) controls the JVM.

20:28 <Ryctolagus> headius: I am kind of a documentation nut, so I will be sure to report back on my findings :)

20:30 <headius[m]> actually we merged some code into JRuby proper for the handle/fd thing...if we have that in jnr-posix as well we might be able to ditch it

20:30 <headius[m]> it's in JRuby because we need to be able to release the fd when the handle closes, or vice versa or something

20:30 <headius[m]> Ryctolagus: ahh ok

20:30 <kares[m]> why would the web server control the app server, is that common these days?

20:33 <headius[m]> https://stackoverflow.com/questions/3776923/how-can-i-normalize-the-eol-character-in-java

20:35 <headius[m]> Java people have no idea

20:35 <headius[m]> I can't believe there's no O_TEXT equivalent in Java IO

20:35 <headius[m]> Java on Windows would probably be much faster for file access if they used it

20:35 <headius[m]> kares: these days I don't know but I do remember using IIS plugins that let it start and stop a JVM

20:35 <headius[m]> it's so that that JVM fits into service controls

20:35 <headius[m]> I guess equivalent would be having init.d scripts for your puma app

20:35 <headius[m]> sort of

20:36 <enebo[m]> headius: yeah although the old guard of Java really did not like adding platform useful shit

20:36 <enebo[m]> We are past that now so perhaps someone should add it to Java23

20:38 <enebo[m]> seems like _setmode is a path to good perf so long as internally Java is not doing something weird like constantly calling _setmode :)

20:38 <kares[m]> thx, thought there would be 2 services for every server but guess win has its ways ... maybe they are 2 acting as one through dependency mngmt

20:39 <headius[m]> enebo: when we can do that yeah I suppose that's the only path right now

20:42 <enebo[m]> headius: when is that not possible?

20:46 <headius[m]> well when we can't dig out the handle

20:46 <headius[m]> I'm back on MacOS again

20:46 <headius[m]> I was about to break my machine with all the ghost clicks and rogue inputs on Windows

20:50 <enebo[m]> ah yeah java 9+ but other file types as well?

20:50 <enebo[m]> I suppose in those cases we just punt to what we are doing for now

20:53 <headius[m]> well this is really just files I think

20:54 <headius[m]> other channels can do CRLF but you have to ask for it...this is a file open(2) mode

20:54 <headius[m]> but because we can't do the mode I'm using explicit CRLF translation all the time

20:54 <headius[m]> so that does limit the scope of this

20:56 <headius[m]> enebo: look at FilenoUtil

20:57 <enebo[m]> yeah

20:58 <headius[m]> I believe with that logic we should be getting real FD when possible...so perhaps we can do this setmode thing now

21:01 <headius[m]> hmm good on me for commenting out all the places MRI IO code does setmode()