09:26 <mattip> iwkse: cool, good luck

12:40 <cfbolz> arigo: thanks for the writeup

12:50 <cfbolz> arigo: I have a small reproducer!

12:50 <cfbolz> and it crashes CPython :-/

14:14 <LarstiQ> cfbolz: is that due to "

14:14 <LarstiQ> So my conclusion is: catching a stack overflow error is not really safe, ever.

14:15 <cfbolz> I don't quite know why it crashes

14:16 <cfbolz> but basically if you catch stack overflow you don't know whether the finallys of functions on the stack successfully finished?

14:18 <cfbolz> https://foss.heptapod.net/pypy/pypy/-/issues/3596#note_165991

14:21 <cfbolz> LarstiQ: ^^

14:35 <LarstiQ> so then RuntimErrors are also in general not safe to catch?

14:37 <cfbolz> LarstiQ: no, it's specifically RecursionError

14:38 <LarstiQ> but that's a a subclass so any RuntimError might be the unsafe RecursionError?

14:38 <cfbolz> ah

14:39 <cfbolz> maybe, yes

15:50 <phlebas> cfbolz: isn't that a general issue with catching stack overflows in any system? the JVM has three different zones at the end of the stack to allow certain things to still work after stack overflows, but it's only ever improving your chances, you cannot assume in general that finally blocks will be run

15:51 <phlebas> even in Squeak, where your stack is on your heap and a stack overflow is basically an out of memory, you can recover, but you cannot be sure that all `ensure:` blocks have run, since they may have run out of memory, too

15:51 <cfbolz> phlebas: right

15:51 <cfbolz> It should be documented

15:52 <cfbolz> And we maybe want the 'several zones' approach

15:55 <phlebas> about the zones on the jvm: https://openjdk.java.net/jeps/270

15:56 <cfbolz> Thanks

16:20 <cfbolz> hahaha: https://bugs.python.org/issue45806

16:28 <cfbolz> So it seem cpython does stuff like that already

17:08 <phlebas> :D and mark opened a PR just 1 hour ago to fix it again

19:59 <iwkse> mattip: thanks :)

20:04 <iwkse> hi, do you know an efficient way to achieve something like this? https://bpa.st/KSQA Basically filtering by a key of an array of dictionaries

20:43 <iwkse> this seems to work res = map(lambda x: {'a': x['a']}, a)

20:48 <Joannah> I wonder if anyone knows a good way or can give an example of using a "xxx.a" link files with rffi. Examples given in this tutorial: https://mesapy.org/rpython-by-example/rffi/index.html are simple. In the PyPy code base, I only found a test case that is not as useful:

20:48 <Joannah> def test_link_files(self):

20:48 <Joannah>         tmpdir = udir.join('link_files' + self.__class__.__name__).ensure(dir=1)

20:48 <Joannah>         eci = ExternalCompilationInfo(link_files=['/foo/bar.a'])

20:48 <Joannah>         mk = self.platform.gen_makefile(['blip.c'], eci, path=tmpdir)

20:48 <Joannah>         mk.write()

20:48 <Joannah>         assert 'LINKFILES = /foo/bar.a' in tmpdir.join('Makefile').read()

20:49 <Joannah> I tried some things but its not recognizing some paths for header files in my C code.