- Oct 06, 2022
- Oct 05, 2022
-
-
Sabar Dasgupta authored
-
Sabar Dasgupta authored
-
- Mar 18, 2020
-
- Sep 03, 2019
-
-
- Add PEP 518 compliant build requirements. This fixes the chicken and egg requirement on numpy - Convert byte array to string. Fixes distutils incompatibility
- Feb 15, 2019
-
-
Mat authored
-
- Oct 14, 2018
-
- Oct 13, 2018
-
-
Mat authored
When creating arrays where the element type is not trivial (for instance strings), it is necessary to pass the element itemsize (elsize) to numpy. Without it numpy has no way of knowing the full size in memory of the array, and in fact it segfaults.
-
- Jul 02, 2018
-
-
Mat authored
-
- Apr 25, 2018
-
-
Mat authored
-
Mat authored
-
Mat authored
-
Mat authored
-
Mat authored
These functions were just wrappers to methods of the numpy array base object which can be accessed directly.
-
Mat authored
Given an array it is now possible to see the name of the SHM block or the name of the file that owns its memory. This name can be fed to SharedArray.delete() directly.
-
- Apr 21, 2018
-
-
Mat authored
-
- Feb 19, 2018
-
-
Kirill Kouzoubov authored
At least on a mac `lseek` and `read` fail when operating on a shared memory file descriptor. Seems like POSIX only promises resize and memory map, but not read or seek. Also on a mac resizing shared memory file rounds up to nearest 4K, so we have to query actual file size after resize before writing header information to the end of the file.
-
- Dec 18, 2017
-
-
Mat authored
-
Update setup.py to list numpy as a pre-requisite and skip the correct guess of the include directory if numpy hasn't been installed yet. To test, start a fresh virtual environment and run `pip install SharedArray`. It crashes without this fix and installs with this fix.
- Oct 08, 2017
-
-
Mat authored
There is no portable way to list SHM segments, that's more or less a flaw in the SHM API. On Linux SHM segments are implemented as regular files on a tmpfs filesystem, usually mounted as /dev/shm. The list method scans this mount point and looks for SharedArray headers. This won't work on other operating systems (*BSD, Darwin, Solaris, etc.), so disable the method on non-Linux targets.
-
Mat authored
On Linux the shm_open(3) and shm_link(3) functions are implemented in librt. This is specific to Linux, other targets typically don't need (nor have) librt. This is the case specifically for *BSD and Darwin.
-
Mat authored
Add rules to build the extension in CI with python2 and python3, and another rule to release to PyPI.
- Oct 07, 2017
- Nov 22, 2016
-
-
Mat authored
The README.md file is converted to reStructuredText when uploading the package to PyPI, because they don't support markdown. The conversion uses pypandoc and use to fail when pypandoc wasn't installed. Make this conditional on the availability of the pypandoc package. Requiring the installation of pypandoc (quite large) just for that purpose would be too much.
- Nov 20, 2016
-
-
Mat authored
PyPI doesn't support markdown. Convert the README.md file to reStructuredText and use that as long_description for the package.
-
Mat authored
-
Mat authored
-
Mat authored
The wrappers are accessible both as module functions and as methods of the MapOwner class which is the base class of the numpy array. The two examples below are identical in behavious: >>> SharedArray.msync(array, SharedArray.MS_ASYNC) >>> array.base.msync(SharedArray.MS_ASYNC)