Improve REPL KI #3030
Improve REPL KI #3030
Conversation
Codecov ReportAttention: Patch coverage is
❌ Your project check has failed because the head coverage (56.15561%) is below the target coverage (100.00000%). You can increase the head coverage or adjust the target coverage.
Additional details and impacted files@@ Coverage Diff @@
## main #3030 +/- ##
=====================================================
- Coverage 100.00000% 56.15561% -43.84439%
=====================================================
Files 124 246 +122
Lines 18764 37608 +18844
Branches 1269 2540 +1271
=====================================================
+ Hits 18764 21119 +2355
- Misses 0 16361 +16361
- Partials 0 128 +128
🚀 New features to boost your workflow:
|
|
Could we spawn the REPL as another process and send it a signal? (simulate ctrl+c) I think we could first send it some data over stdin to get into specific scenarios. |
|
The newline is sent to the terminal I think, rather than any specific process. Potentially we could avoid the ioctl weirdness by allowing a small amount of user discomfort in needing to press enter after Ctrl+C. |
|
Still no idea how to test the proposed fixes. I think it's relevant that asyncio isn't able to interrupt the input thread either, unless it's using the new python repl: Was there ever talk about the |
src/trio/_repl.py
Outdated
| nonlocal interrupted | ||
| interrupted = True | ||
| # Fake up a newline char as if user had typed it at terminal | ||
| fcntl.ioctl(sys.stdin, termios.TIOCSTI, b"\n") |
There was a problem hiding this comment.
you can't write to sys.stdin from a handler safely, I'm not sure if you can icotl to the .fileno() or not either, I think you need:
@disable_ki_protection
def _newline():
# Fake up a newline char as if user had typed it at terminal
fcntl.ioctl(sys.stdin, termios.TIOCSTI, b"\n")
class TrioInteractiveConsole:
token = trio.lowlevel.current_trio_token()
def handler(...):
nonlocal interrupted
interrupted = True
token.run_sync_soon(_newline, idempotent=True)There was a problem hiding this comment.
Thanks! I think I understand why, but won't a keyboard interrupt in the newline function blow up trio and end the repl? If so, Would suppressing the exception be sufficient?
richardsheridan
force-pushed
the
repl_ki
branch
from
74760af to
620a5c0
Compare
richardsheridan
force-pushed
the
repl_ki
branch
from
02ae29d to
a5dcdbf
Compare
Uh turns out pyrepl is a separate package that cpython vendored: https://pypi.org/project/pyrepl/#history I have no intention of vendoring it or making a dependency for this pr but just FYI |
This commit has a draft method to fix #3007. I have 0 ideas on how to test this thing: not only does it use the
raw_inputthat gets patched for testing, but the way it injects a newline is basically impossible for a test runner to pass in. Also, it feels fragile somehow to dip in and out of the trio loop patching the signal handler and peeking forKeyboardInterrupt.