Issues/8786 #9074
Issues/8786 #9074
Conversation
Add back manylinux support / options as well.
| try: | ||
| setup_runner.run_setup_command(source_dir=setup_dir, setup_command=split_command) | ||
| except SetupPyRunner.CommandFailure as e: | ||
| raise TaskError(f'Install failed: {e}') |
There was a problem hiding this comment.
I've heard this recommendation before from Eric and I don't understand it. In particular, a TaskError is user facing and the messages are printed to the console even with tracebacks disabled. In those cases adding the exception type to the message - which is all repr does afaict, seems generally undesirable. An internal name leaked to a user is not helpful. I'ts also not very helpful to a debugging dev with bactraces turned off. With backtraces turned on, the backtrace is what is helpful.
Concretely in this case, on the CLI the user would currently see:
$ ./pants setup-py ...
...
Install failed: Failed to execute python /a/pex setup.py ...
With repr they'd see:
$ ./pants setup-py ...
...
Install failed: CommandFailure('Failed to execute python /a/pex setup.py ...
I'm not seeing what the point of adding CommandFailure('...') wrapping the message is.
| manylinux = self.get_options().resolver_manylinux | ||
| if manylinux.lower() in ('false', 'no', 'none'): | ||
| if self.get_options().resolver_use_manylinux: | ||
| logger.warning('The [{scope}] manylinux option is explicitly set to {manylinux} ' |
| except SetupPyRunner.CommandFailure as e: | ||
| raise self.BuildLocalPythonDistributionsError( | ||
| f"Installation of python distribution from target {dist_tgt} into directory " | ||
| f"{dist_target_dir} failed using the host system's compiler and linker: {e}") |
This reverts commit ac08e4c.
Easier than figuring out how to speed them up, since we want to get rid of this entire backend anyway.
|
The remaining test failures are covered by pex-tool/pex#892 |
|
I've refreshed against upstream master and applied the pex 2.1.4 fixes. If this goes green I may switch back to #8881 to clean up and finish. |
Interpreter constraints were too broad for the default pytest setup we use which pins zipp==2.1.0 which only works for python>=3.6. Also kill legacy constraints passed to the `pants_run` IT helper.
Keep Pants to its baileywick - use pex to do this.
|
Alright, the 1 failing shard - The |
|
OK - the lint shard is unrelated and being tackled in #9155. |
|
Ok the 42e38a7 merge pulls in Eric's lint CI shard timeout fix and so should nominally go green. Still spelunking very flaky |
Since this is now done in PythonBinary internally, we can revert to the cleaner check. Additionally, this self-contained handling now removes the need for PythonInterpreter cache clearing in test_select_interpreter.py.
An attempt to get John's pex 2.x upgrade working while he's OOO. Do not review.