gh-202: Simplify Windows installation #203
Conversation
| Windows notes | ||
| ------------- | ||
|
|
||
| On Windows, to allow pyperformance to build dependencies from source | ||
| like ``greenlet``, ``dulwich`` or ``psutil``, if you want to use a | ||
| ``python.exe`` built from source, you should not use the ``python.exe`` | ||
| directly. Instead, you must run the little-known command ``PC\layout`` | ||
| to create a filesystem layout that resembles an installed Python:: | ||
|
|
||
| .\python.bat -m PC.layout --preset-default --copy installed -v | ||
|
|
||
| (Use the ``--help`` flag for more info about ``PC\layout``.) | ||
|
|
||
| Now you can use the "installed" Python executable:: | ||
|
|
||
| installed\python.exe -m pip install pyperformance | ||
| installed\python.exe -m pyperformance run ... | ||
|
|
||
| Using an *actually* installed Python executable (e.g. via ``py``) | ||
| works fine too. | ||
|
|
There was a problem hiding this comment.
It seems like this should remain for the sake of Python versions that don't have the fix. So a note should be added identifying the last Python version to which this applies.
There was a problem hiding this comment.
I was going with the assumption that using a source build on Windows is such a rare occurrence that this would be in the realm of CPython developers already operating out of a GH clone. This would imply an up to date (-ish) checkout, therefore containing the bug fix.
There was a problem hiding this comment.
Yeah, that makes sense. The catch is that we run benchmarks against older CPython versions too.
There was a problem hiding this comment.
Do note, though, the removed verbiage only applies to source builds. Other means of running Python (nuget, winget, installer) work as is. I would say workflows that clone a released Python branch and build it to run pyperformance would be better served using the released binary for the comparison.
For example:
nuget.exe install python -Version 3.10 [-OutputDirectory <dir>]
will download and unpack the released distribution for 3.10.0 to <dir>\python as a full usable interpreter (<dir>\python\tools\python.exe). This already is done in the build scripts just to build the interpreter now[1].
nuget.exe is also available in the <srcdir>\externals directory (see [1]). This has the added benefit of guaranteeing isolation from previous builds.
[1] Unless there is an existing Python installation usable for building.
There was a problem hiding this comment.
I'll admit, however, I have no idea how any current workflows using pyperformance on Windows are implemented. Although the current guidance seems to be quite heavy already.
There was a problem hiding this comment.
Very few have managed to successfully run PyPerformance on Windows, because it has been so broken for so long (and the test suite is still broken -- as soon as you start it, the process detaches and runs in the background, which I didn't even know was possible in DOS Windows :-).
I have a reason for building past versions from source though: You never know which aspects of your current setup (compiler version, Windows version, hardware, version of other tooling installed) might affect the performance of the binary. It wouldn't be the first time that someone reported a significant but mysterious/irreproducible speedup or slowdown between two releases only to find out that they were comparing binaries built with different compiler versions. If the goal is to measure the effect of the source code changes rather than simply the progress in compiler technology, it's important to try to keep as many variables unchanged as possible.
| # Recent setuptools versions have a bug that prevents being | ||
| # used to compile from a source build virtual environment. | ||
| # [BUG](https://github.com/pypa/setuptools/issues/3325) | ||
| # The stdlib distutils does not have this problem. |
There was a problem hiding this comment.
What happens when the stdlib distutils goes away?
There was a problem hiding this comment.
Presumably the bug would have to be fixed in setuptools?
|
@gvanrossum, what do you think? |
gvanrossum
changed the title
|
Removing that docs section seems premature, the root cause hasn't even been fixed (seems the OP is on a quest here :-). I'm okay with adding the env var to work around the setuptools problem, we can just remove it again (and require a high enough setuptools version) once that is fixed in setuptools. I don't think stdlib distutils is going away in 3.12 yet, is it? If it removed before setuptools fixes the bug on their side, we'd have to lean on setuptools. |
|
I wouldn't say quest but attempting to run pyperformance from a source checkout is what started this whole chain of changes. I just found it unreasonable to need to remember to do the additional setup for testing an in-progress build. So, I leveraged my Windows build system knowledge to hunt down the cause of this issue. |
See python/cpython#92584 (and https://peps.python.org/pep-0632), distutils is going away in 3.12. That has not happened quite yet, but probably will relatively soon. |
PR for issue #202