On Mon, Nov 23, 2015 at 3:11 PM, R. David Murray <rdmurray@bitdance.com> wrote:
On Mon, 23 Nov 2015 19:11:15 +0200, Serhiy Storchaka <storchaka@gmail.com> wrote:
On 23.11.15 18:00, R. David Murray wrote:
On Mon, 23 Nov 2015 00:58:01 -0600, Zachary Ware <zachary.ware+pydev@gmail.com> wrote: I haven't looked at this, but unless the buildbot does *not* have write access to the installed directories (ie: the install was done as another user) it isn't really doing a full installed python test.
Yes, but at least it catches cases where some files are not installed. There were few issues with this.
True. Something incomplete in this vein is better than nothing. I'm Not sure you should call it "Installed" though, as that will be a bit misleading. Most of the "can't run the tests on installed python" bugs are because the tree is read-only (obviously, not all of them!). Maybe call it "local install"? Wordy, I know, but more accurate.
I've gone with attempting to make it more like a 'real' install, by wrapping the test step with appropriate 'chmod' commands to make the install directory not writable, and confirmed during a test run that the entire installed tree is -w. I also fixed the slave's usage of usePTY (for test_curses) to avoid the failures that had been happening in the 'uninstall' step.
-- Zach