Add tab-completion support to PEX repls. #2321
Conversation
jsirois
force-pushed
the
repl/support-completion
branch
2 times, most recently
from
9a13dbb
to
4ad367e
Compare
| @@ -234,37 +235,95 @@ def test_pex_repl_built(): | |||
| assert b">>>" in stdout | |||
|
|
|||
|
|
|||
| @pytest.mark.skipif( | |||
| IS_PYPY or IS_MAC, | |||
| reason="REPL history is only supported on CPython. It works on macOS in an interactive " | |||
There was a problem hiding this comment.
This appears to never have been true. In fact, using PEX_INTERPRETER_HISTORY / PEX_INTERPRETER_HISTORY_FILE on any Python interpreter without readline support, even CPython, previously led to:
$ PEX_INTERPRETER_HISTORY=1 PEX_INTERPRETER_HISTORY_FILE=/tmp/check $PWD/install/bin/python3.12 empty.pex
Traceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/__main__.py", line 106, in <module>
bootstrap_pex(__entry_point__, execute=__execute__, venv_dir=__venv_dir__)
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex_bootstrapper.py", line 627, in bootstrap_pex
pex.PEX(entry_point).execute()
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex.py", line 561, in execute
sys.exit(self._wrap_coverage(self._wrap_profiling, self._execute))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex.py", line 468, in _wrap_coverage
return runner(*args)
^^^^^^^^^^^^^
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex.py", line 499, in _wrap_profiling
return runner(*args)
^^^^^^^^^^^^^
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex.py", line 585, in _execute
return self.execute_interpreter()
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/jsirois/.pex/unzipped_pexes/e2c4c91895f250a4f203b1d575cb3c39208ea5a7/.bootstrap/pex/pex.py", line 661, in execute_interpreter
import readline
ModuleNotFoundError: No module named 'readline'
(Here I hand-built CPython3.12 with ./configure --prefix $PWD/install --without-readline)
There was a problem hiding this comment.
Hmm, archaeology on #2018 (whose CI logs have unfortunately expired) finds a commit message stating that the issue on pypy (in CI) was apparently related to terminfo, so may have been due to interaction between pypy and the CI environment rather than pypy per se, but AFAICT wasn't due to unavailability of readline. Similarly, the macos ENOTTY issue only manifested in CI, and was not related to the presence of readline.
Anyway, if this is now deemed false and the actual underlying issue is just readline then the mentions of CPython in pex/variables.py#531 and pex/variables.py#542 should be removed.
There was a problem hiding this comment.
Yeah, definitely false. CI passes for pypy3.10 and mac and I've personally tested for pypy{2.7,3.{5,6,7,8,9,10}}. This PR had to plumb TERM to tests to work around 1/2 of the tty issues (using pexpect / ptys was the other half).
Variables docs fixed.
There was a problem hiding this comment.
I should have dived deeper in #2018 then, sounds like TERM would have made things work.
There was a problem hiding this comment.
Well, you'd have still needed to use psuedo terminals for the notty issue as in this PR and then figured out the libedit _HiStOrY_V2_ magic to work around error 22 on mac. It was a full 3 days of swearing for me anyway to move from works on my machine to fully tested.
This works for all supported CPython versions, although automated testing is limited to CPython 3.6+.
jsirois
force-pushed
the
repl/support-completion
branch
from
2784fea
to
bfd8817
Compare
| ) | ||
| else: |
There was a problem hiding this comment.
python has -S that (among other things) disables completion. It looks like this is unconditional, if the modules exist.
Is there potential risks/downsides to doing this unconditionally without a knob to turn it off, beyond the (seemingly small) performance cost of this set-up?
I assume that it's harmless enough, just double checking.
There was a problem hiding this comment.
Yeah, I got familiar with the CPython site.py code that -S disables in order to get this all working. I can't see any harm.
This works for all supported Python versions even when those versions
own REPL does not support tab-completion out of the box.