The malloc warnings happen on OSX, with a fresh checkout of the python3
branch. Python was build using --enable-universalsdk=/ --with-universal-
archs=all, on a x86_64 capable laptop running Leopard.

The issue goes away when running from the commandline, the "-E -bb"
flags that 'make test' use seem to be involved in the issue.

test_io

Testing large file ops skipped on darwin.
It requires 2147483648 bytes and a long time.
Use 'regrtest.py -u largefile test_io' to run it.

Testing large file ops skipped on darwin.
It requires 2147483648 bytes and a long time.
Use 'regrtest.py -u largefile test_io' to run it.
python.exe(35535,0x7fff701d1720) malloc: *** mmap(size=-
9223372036854775808) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug
python.exe(35535,0x7fff701d1720) malloc: *** mmap(size=-
9223372036854775808) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug
python.exe(35535,0x7fff701d1720) malloc: *** mmap(size=-
9223372036854775808) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug
test_ioctl

What's more annoying: the error goes away while running in a debugger.

The issue happens at least in tes_constructor:

test_constructor (main.CBufferedWriterTest) ...
python.exe(35957,0x7fff701d1720) malloc: *** mmap(size=-
9223372036854775808) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug
ok

Buffer_size is set to 0x7fffffffffffffff in BufferedWriter_init at the end
of test_constructor.

I have no idea why this happens, but this definitly seems wrong to me.
Debugging is rather hard at the moment because the issue goes away in the
debugger.

Setting this as a release blocker because arguments of a wrong magnitude
to memory allocation functions are scary.

malloc: *** mmap(size=-9223372036854775808) failed

Does this error occur even if you revert Modules/mmapmodule.c to ｒ69214?
I have touched this module recently (I believe I fixed only SEGV bugs,
not changed behavior) so I want to know it.

I don't think this is related to Python's mmap module, the message says
that malloc(3) cannot mmap some extra memory space.

I'll test with the older version of mmap just in case.

Sorry, this is not related to mmap module... Forgot my comments.

If you look at test_io, calling malloc() with sys.maxsize is precisely
the purpose of the test, to check that allocation failures are detected
and reported properly. So the warnings can be safely ignored, if they
only happen in test_constructor.

PS: if the warnings are annoying, the test can be disabled selectively
under Mac OS X.

I can't believe that I completely missed that this is the purpose of the
tests, I honestly believed that I had checked everything there was to
check :-(

Disabling the test won't be necessary, I'm keeping this issue open for a
while longer to check if we can avoid getting a message on stderr during
the test. It's rather annoying that malloc logs error messages during what
turns out to be perfectly fine behaviour.

Closing the issue as this is not a bug in python (as noted in the
comments) and there doesn't seem to be a workaround for avoiding the
spurious messages on stderr.

In Jupyter Notebook, I tried to pass a large amount of data (about 2.3 GB) to Statsmodels's KDEUnivariate.fit (https://www.statsmodels.org/dev/generated/statsmodels.nonparametric.kde.KDEUnivariate.fit.html#statsmodels.nonparametric.kde.KDEUnivariate.fit) with kernel='cos'. In the Notebook, I got a MemoryError exception and traceback (which unfortunately I have since lost).

In the terminal where I was running the Jupyter server, the Python 3.7 kernel spat out:

Python(86244,0x7fff91104380) malloc: *** mach_vm_map(size=465845322670080) failed (error code=3)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug

The MemoryError coming up in the Notebook interface is the correct behavior, but I wanted to bring to your attention that the malloc junk on stderr happens even outside of Python's tests.

I'm on macOS 10.13.5.

The error is printed by libc, there's nothing we can do about that.