18:42 <earnestly> Is the amount of memory required to build ponylang mentioned anywhere? Once I get the 'Generating Reachability' memory usage spikes and eventually the oomkiller in invoked "Killed process 6584 (ponyc) total-vm:3840264kB, anon-rss:3427436kB, file-rss:0kB, shmem-rss:0kB". I only have 4G (3.9GiB) available and it seems like this is not enough

18:43 <earnestly> The rest of the build seems to require very little though, around 343M (similar to glibc), so I'm wondering if there is more likely a leak as there are a few issues open about such things

18:44 <earnestly> It's also not something I can easily test locally as it grinds my machine to a halt for almost 10 minutes

19:27 <TwoNotes> Those "Killed" messages are a sign of running out of memory. On my Rpi3 I have to have a swapfile defined on a USB external disk

19:28 <TwoNotes> earnestly, I use 4GB

19:29 <TwoNotes> My RPi3 has just 1GB of RAM. But with the swap file it can compile ponyc

19:29 <earnestly> TwoNotes: Well of course, I have some things using some of that memory already, I've tried killing off some non-essentials so I had 3.4G remaining (with most of it cached)

19:29 <earnestly> TwoNotes: Yeah, that's also a different architecture

19:30 <TwoNotes> I doubt the compiler uses THAT much less memory on ARM

19:30 <earnestly> This only happens when 'Generating Reachability' after most of the tests and even after building the packages

19:31 <earnestly> I have no idea what 'Generating' means, the Makefile is by default configured to provide very little information, probably for aesthetic reasons

19:32 <TwoNotes> ponyc has a phase where it eliminates code that can never be reached. perhaps it is that

19:32 <TwoNotes> Are you talking about building ponyc itself (which is written in C) or in compiling Pony programs?

19:33 <TwoNotes> The 'packages' are recompiled every time you compile a Pony program.

19:35 <earnestly> TwoNotes: Itself

19:36 <earnestly> TwoNotes: I should have said, 'build ponyc' instead of 'build ponylang', that was not as clear

19:53 <TwoNotes> When the 'Main' actor has nothing more to do, does that exit the program, even if other actors are still running?

19:57 <Praetonus> earnestly: This message about reachability only appears when building a pony program, and not when compiling the compiler itself, so your problem is unclear to me

19:57 <Praetonus> TwoNotes: No, the program terminates when the system is quiescent, i.e. no actor is executing a behaviour and all message queues are empty

20:12 <earnestly> Praetonus: https://ptpb.pw/kzic

20:12 <earnestly> Praetonus: These are all the points at which anything is built

20:12 <earnestly> Praetonus: Could it be from generating the docs for stdlib then?

20:14 <Praetonus> Looks like there is a missing space between `-r` and `expr` at line 6

20:14 <earnestly> Should that matter?

20:14 <earnestly> I'll try with a space either way

20:15 <Praetonus> With that typo, the full test suite for the standard library is built, going through the reachability pass and consuming a lot of memory

20:15 <Praetonus> Reducing the memory usage here is a hard problem

20:15 <earnestly> Praetonus: Ah, so when flags take arguments, the remaining flags aren't treated as the argument?

20:15 <Praetonus> I don't think they are

20:16 <earnestly> For single flags, -a -b -c and -abc should be equivalent, but if -a takes an argument then -abc and -a bc are equivalent

20:16 <earnestly> At least, this is "normal" for most tools, gcc/clang included

20:16 <earnestly> Praetonus: Anyway, here goes. I have a bunch of stuff going so hopefully no lockups :p

20:17 <Praetonus> I don't have a strong opinion on flag positioning. You can open an issue on github if you wish to discuss that further

20:28 <earnestly> Praetonus: Unfortuantely, same problem

20:29 <earnestly> I guess for now I'll just skip generating the documentation

20:55 <earnestly> Praetonus: So this message wasn't coming from that command for generating the docs, from instead make config=release prefix=/usr test

20:55 <earnestly> Praetonus: Here's the log: ~1800 lines long: https://ptpb.pw/VqRR

20:58 <earnestly> I'm slightly surprised no one else is having this issue

20:59 <earnestly> (That said, this has been an issue for a few months, I should have said something earlier but thought it was an obvious regression)

21:00 <earnestly> (And since it hasn't changed, maybe it's a consequence of something and nothing can be fixed. At which point it'd be nice to know what the requirements to build are)

21:13 <Praetonus> The memory required to compile the test suite (or any other big pony program) is pretty high. This comes from the way the reachability pass works. Improving that would require some work on the compiler, but it would be nice to do that. In the meantime, the only way to fix the problem is to have more memory available

21:13 <Praetonus> IIRC, the standard library tests require between 2 and 3 GB of memory to compile

21:14 <TwoNotes> Praetonus, my program goes into background and so its output does not appear. If I run it under GDB, the output *does* appear. I do not do anything that would make it background itself..

21:15 <TwoNotes> My largest Pony program is about 1700 lines of code, plus liraries, and it will not compile on 1GB machine without a swapfile

21:18 <earnestly> Praetonus: I don't mind requirements on memory being available, but it'd be nice to know how much. Some projects do kindly mention what kind of memory might be needed, such as Rust and Chromium

21:18 <earnestly> Or android stuff

21:18 <earnestly> Praetonus: It clearly needs more than 3.8G on x86_64

21:19 <TwoNotes> Most languages do not recompile their whole library each time. So the Pony compiler is looking at a lot more stuff

21:19 <TwoNotes> Well, the part of the library you actually use